Kimi K3 on Korea Invest Insights

Weekly Semiconductor Deep Dive: Strong Earnings, Compressed Multiples, and the 2028 Inflection

Sun, 19 Jul 2026 15:53:00 +0900

On July 13, KOSPI fell 9.05% and SK Hynix dropped 15.37%. Yet in the same week, TSMC reported Q2 revenue of $40.2 billion and a gross margin of 67.7%, while ASML raised its annual revenue outlook to €43–45 billion. Commodity DRAM and NAND contract prices also rose sharply. It was a week in which physical orders and stock prices moved in opposite directions.

This divergence is less a signal that the semiconductor boom is over, and more a sign that the market’s question has changed. Investors are no longer looking only at how much will be earned in 2026 and 2027. They are first calculating how much supply will expand in 2028, whether customers can sustain high memory prices and data center capex, and how much inference efficiency gains will erode the scarcity premium of HBM.

Connected Context This post consolidates Five Clocks of the Semiconductor Bull and Bear Case, Memory Fair Value via FCFE and Normalized Earnings, Samsung Electronics and SK Hynix: 2027–2028 Scenarios, and HBM 2030 Supply-Demand Model Cross-Check (26.7 EB) into a weekly synthesis updated with new data from the third week of July. Related posts are available at the AI HBM Hub and the Exclusive Analysis Hub.

TL;DR

Official data confirmed from July 13–19 does not support the claim that AI semiconductor demand has peaked. TSMC and ASML results and guidance, DRAM and NAND prices, and server memory demand all reinforce the earnings-strength thesis for 2026–2027.
The same data also amplifies 2028 supply risk. ASML’s capacity expansion plans and TSMC’s heavy capex are evidence of current orders — and simultaneously a reservation of future supply and depreciation.
HBM long-term contracts can raise the floor on volume and earnings, but may slow the pace at which spot price spikes translate into blended ASPs. Volume stability and a cap on price upside coexist.
China plays three roles simultaneously: an end-demand destination, a competitor in commodity memory, and a market that could be politically decoupled. Its larger impact is on 2028 commodity DRAM and NAND price ceilings and global supply discipline, not on near-term HBM substitution.
Efficiency gains from open models such as Kimi K3 do not simply eliminate semiconductor demand. They can shift the value of some HBM toward server DRAM, eSSD, networking, and power. The key question is not whether cost-per-task falls, but how much total workload expands and what share is deployed on self-built infrastructure.
Current base-case scenario probabilities: Extended Scarcity 25%, Strong Earnings / Compressed Multiple 40%, Efficiency and Supply Normalization 25%, Financial / Demand Dislocation 10%. The most probable path is one where strong earnings in 2026–2027 and duration discounting for 2028 coexist.

Key Statement of the Week

The physical evidence underpinning the memory supercycle passed its first full-scale stress test. Stock prices, however, have begun asking not about the magnitude of scarcity but about how long excess earnings will last. 2026–2027 is the period for earnings delivery; 2028 is the period that simultaneously tests supply, efficiency, and customer profitability.

1. Scope and Evidence Rules

The analysis covers July 13–19, 2026. Company IR materials, government and congressional filings, industry data, broker estimates, and market supply-demand figures drawn from two weekly analysis documents were re-categorized. Where multiple posts cited the same underlying source, the confirmation count was not incremented.

Grade	Source	How Used in This Report
A	Company IR, government / congressional / regulatory filings, SEC disclosures	Baseline for facts and official plans
B	Industry research such as TrendForce, primary broker reports	Supporting evidence for pricing, share, and company-level estimates
C	Cross-referenced press reports, channel check summaries	Used for directional confirmation only; not used as standalone investment signals
D	Social media, circulating summary images	Used only to identify questions requiring further verification

Orders from TSMC, ASML, and memory manufacturers can reflect the same hyperscaler capex plan multiple times. Official plans signal demand intent but do not guarantee actual utilization or cash flow. The sharp stock price declines were not attributed solely to technical supply-demand dynamics. Flow imbalances can amplify drawdowns, but they cannot eliminate a fundamental re-rating of earnings duration.

2. A Map of the Week’s Events

Date	Confirmed Developments	Questions the Market Began Asking	Current Assessment
July 13	HBM 2030 demand model at 26.7 EB re-verified; SK Hynix Q2 earnings estimate cut; KOSPI sharp decline	Long-term shortage, or excessive extrapolation?	2027 shortage has strong support; 2030 shortage magnitude carries low conviction
July 14	SK Hynix estimates adjusted to reflect HBM long-term contracts; IBM and Ericsson flagging memory cost burden; reports of Apple testing CXMT chips	Are long-term contracts an earnings floor or a price ceiling?	Volume visibility is high, but ASP upside constraints and customer cost burden emerge alongside
July 15	ASML Q2 results and 2027–2028 capacity expansion plans	Strong orders, or the start of a supply response?	Both are correct — simultaneously reinforces current demand and amplifies 2028 supply risk
July 16	TSMC Q2 results and Q3 guidance; U.S. House letter requesting restrictions on Chinese memory procurement	AI demand is strong — why is memory stock price weak?	Debate centers on earnings attribution, supply, and cost of capital — not demand collapse
July 17	Kimi K3 launch; Chinese open-model efficiency gains and API pricing repricing	Does efficiency reduce semiconductor demand?	Cost-per-task falls, but aggregate usage and self-built deployments may expand
July 18	China AI API cost analysis; review of memory trio’s China exposure	Is China a customer or a competitor?	Both roles simultaneously; impacts 2028 normalized value more than 2026 EPS
July 19	Channel checks circulating claiming strong server DRAM and HBM4 orders	Did the sell-off price in a Q3 earnings collapse?	Probability appears lower, but primary source unconfirmed — maintained at Grade C

The most significant shift in this table is not the direction of individual events, but the time horizon. Until early July, the debate centered on whether H2 2026 earnings would beat expectations. It has since shifted to whether 2028 earnings represent a normalized earnings level, and how long elevated prices and margins can be sustained.

3. Evidence Strengthened This Week

3.1 TSMC and ASML Showed That AI Orders Have Translated into Real Activity

TSMC’s Q2 results demonstrate that AI demand has moved beyond the planning stage.

Item	TSMC Q2 2026 and Q3 2026 Outlook
Q2 revenue	$40.2 billion
Q2 gross margin	67.7%
Q2 operating margin	60.3%
Q3 revenue guidance	$44.6–$45.8 billion
Q3 gross margin guidance	65–67%

The high gross margins and next-quarter growth guidance indicate that the quality of orders for leading-edge process nodes and advanced packaging remains intact. TSMC Q2 2026 Official Results

ASML confirmed the same direction. Q2 revenue of €9.326 billion and net income of €2.918 billion were reported, and the 2026 revenue outlook was set at €43–45 billion. The capacity roadmap reveals the timeline at which current orders translate into future supply.

Equipment	2026 Capacity	2027 Plan	2028 Under Review
Low-NA EUV	~65 units	~30% expansion	Additional ~30% expansion under review
DUV immersion	~130 units	~30% expansion	Additional ~30% expansion under review

ASML stated that AI investment is driving demand for leading-edge logic and memory, and that customer capacity commitments are converting into long-term demand visibility. These figures weaken the strong bear case that AI orders consist entirely of duplicate bookings. ASML Q2 2026 Official Results

3.2 Memory Prices Remain Elevated

TrendForce projected Q2 2026 commodity DRAM contract prices to rise 58–63% quarter-over-quarter, and NAND prices to rise 70–75%. The Q3 server DRAM price growth forecast was revised to 13–18%, but the upward trajectory is maintained. Q2 DRAM/NAND Outlook, Q3 Server DRAM Outlook

It is important to distinguish between a deceleration in the rate of price increases and an actual price decline. Given how rapidly prices rose in Q2, the quarter-over-quarter growth rate in Q3 may slow. Nonetheless, if absolute prices remain elevated, suppliers’ operating income can remain strong. The risk the market is pricing is not a decline in earnings, but a deceleration in earnings growth.

3.3 HBM Ramp Cannibalizes Commodity DRAM Supply

HBM requires more wafer area, stacking, packaging, and testing capacity than commodity DRAM to produce the same number of bits. As the three memory manufacturers increase their HBM mix, HBM shipments expand while the effective bit supply of commodity DRAM can shrink.

As a result, the bottleneck broadens in the following sequence:

GPU / ASIC
-> HBM
-> Server DRAM
-> eSSD and NAND
-> Packaging / Substrate / Test
-> Networking / Power / Cooling

AI agents and long-context inference do not rely solely on HBM for model weights. They also consume server DRAM and eSSD for KV cache and data serving, and consume more networking to connect multiple accelerators and sparse mixture-of-expert models. This is why aggregate memory tier volumes can continue to expand even if HBM unit pricing comes under pressure.

3.4 2027 Shortage Is a Strong Conclusion; 2.5× Shortage by 2030 Is a Scenario

Comparing 2030 HBM demand of 26.7 EB against supply of 10.6 EB yields demand at 2.52× supply. The arithmetic is reproducible. The issue lies not in the result but in whether the underlying assumptions hold simultaneously. The base model assumes 24× token growth, 5× model size expansion, 4× context length extension, and a 70% HBM retention rate for KV cache — all simultaneously.

Assumption Change	2030 HBM Demand	vs. 10.6 EB Supply
Base model	26.7 EB	2.52×
Model size 2×	18.5 EB	1.75×
KV efficiency 6×	22.3 EB	2.10×
HBM retention rate 50%	22.9 EB	2.16×
Tokens 12×	16.2 EB	1.53×
Composite bear assumptions	6.5 EB	0.62×

Reducing any single variable still leaves a shortage, but if multiple bearish assumptions hold simultaneously, oversupply becomes possible. The investable statement is therefore: “The market is likely to remain tight through 2027.” The statement “Supply will be definitively 2.5× short by 2030” cannot be directly plugged into a price target.

4. Risks Strengthened This Week

4.1 Earnings Growth Rate Has Become More Important Than Strong Absolute Earnings

The SK Hynix Q2 operating income estimate cut originated from adjustments to ASP assumptions rather than from a collapse in shipment volumes.

Report	Q2 2026 Operating Income Estimate	Key Adjustment
Mirae Asset Securities	From ₩70.7 trillion to ₩62.3 trillion	DRAM ASP from +40.6% to +32.9%; NAND from +55% to +50%
Korea Investment & Securities	₩60.4 trillion	DRAM ASP +28.9%; NAND +50.9%

An operating income in the ₩60 trillion range is very large in absolute terms. However, stock prices responded more to the magnitude of the revision from prior estimates than to the absolute level of earnings. The 15.37% decline in SK Hynix on July 13 is better explained by the combination of shifting expectations and concentrated technical supply-demand dynamics than by any conclusion that the upcycle has ended.

This is the central logic of the weekly analysis. Strong demand alone does not drive stock prices higher. Demand must beat market expectations, and evidence that elevated earnings persist into 2028 is required.

4.2 Long-Term Contracts Create Both an Earnings Floor and a Price Ceiling

HBM long-term supply agreements provide volume visibility, customer lock-in, predictability of payback periods, and downside protection when spot prices fall. On the other side, the following concerns apply:

Even if spot prices spike, a company’s blended ASP may rise only gradually.
Pricing formulas, renegotiation mechanisms, minimum purchase commitments, and cancellation terms are not disclosed.
Customer compliance with minimum purchase obligations and credit risk remain with the supplier.
Up-front payments and long-term volume commitments may not flow through to accounting earnings and operating cash flow at the same pace.

Interpreting long-term contracts as an unconditional reinforcement of pricing power is only half the picture. Contracts raise the earnings floor and reduce volatility, but they may not capture the full upside of a spot price spike. The SK Hynix Q2 estimate revision was the first instance of this gap appearing in reported numbers. The detailed figures are discussed in SK Hynix Q2 Earnings Cut and Maintained Price Targets.

4.3 Equipment Orders Are Both Current Demand and Future Supply

Reading ASML’s capacity expansion solely as near-term bullish evidence misses the time dimension.

Low-NA EUV 2027 capacity ≈ 65 units × 1.30 = 84.5 units
DUV immersion 2027 capacity ≈ 130 units × 1.30 = 169 units

Equipment manufacturers recognize orders as revenue in 2026–2027 first. Actual bit production at memory manufacturers can increase from 2028 onward, after equipment installation, yield stabilization, and packaging capacity additions. The same capex that initially boosts equipment and materials earnings later pressures memory prices and margins.

The metrics to watch for 2028 are therefore not the announced investment totals. They are equipment installation timing, new wafer starts, yield ramp, packaging throughput, the increase in depreciation, and actual bit output.

4.4 High Memory Prices Can Erode Customer Demand

When memory prices rise, customers can respond in four ways:

Pull forward inventory purchases before prices rise further.
Raise server and end-product prices.
Reduce memory content or downgrade specifications.
Delay purchases and projects.

The first behavior makes near-term orders appear stronger but simply pulls forward future demand. The other three represent pathways through which suppliers’ pricing power ultimately destroys customer demand. The IBM and Ericsson examples showed that memory supply constraints have begun to affect end-customer budget allocation and margins. We are currently in a phase where pre-buying and specification adjustments are visible before large-scale order cancellations.

4.5 AI Capex Bottleneck Is Shifting to Cost of Capital

Even if AI demand is real, if the investing entity’s cash flow cannot keep pace with depreciation and financing costs, the pace of investment will slow. Going forward, the following four items matter more than the absolute capex figure from hyperscalers:

Whether AI and cloud gross margin growth outpaces the increase in depreciation.
Whether data center power delivery and GPU utilization ramp according to plan.
Whether returns on new projects, including borrowing costs, exceed the cost of capital.
Whether the credit quality and up-front payments from customers that have signed HBM long-term contracts remain stable.

The strength of orders at TSMC and ASML means they are not evidence of current demand collapse. However, in U.S. Big Tech earnings releases at the end of July, investors should examine AI gross margins, free cash flow, and the slope of 2027 investment guidance alongside — not instead of — total capex figures.

This week’s sharp decline in Korean semiconductor stocks bears the hallmarks of leverage unwinding, program selling, and a market-cap structure heavily concentrated in a handful of large-cap names — all of which amplified the drawdown. Simultaneous selling by foreign and institutional investors caused single-day price moves far in excess of any change in underlying business value.

Yet concluding that “it was merely a supply-demand shock, so everything rebounds” carries its own danger. If relative strength fails to recover even after strong earnings and pricing data, it may signal that the market is actively lowering its estimates for 2028 EPS and normalized multiples.

The diagnostic framework is straightforward.

Signals pointing toward a supply-demand shock	Signals pointing toward a fundamental repricing
Foreign and program selling decelerates quickly	Selling continues despite positive news
Relative strength recovers on volume	Semiconductors continue to underperform the broader market
12-month forward EPS revisions remain positive	Samsung Electronics, SK Hynix, and Micron EPS all revised down simultaneously
Contract prices rise and inventories stabilize	Inventories build while contract prices decline simultaneously
Big-Tech CAPEX and AI profitability both improve	CAPEX cuts or deteriorating AI margins

This distinction is precisely why price and earnings estimates must be tracked together. Watching only supply-demand dynamics risks missing long-term risks; watching only results risks missing the additional downside created by crowded positioning.

6. China Is Three Variables at Once

6.1 China Is a Major Demand Market

China exposure figures disclosed in company filings use different definitions, making direct comparisons difficult.

Company	China Exposure Metric	Caveat
Samsung Electronics	30.1%	Based on standalone consolidated revenue — not memory-only share
SK Hynix	1Q26 24.3%, full-year 2025 19.7%	Based on sales entity location
Micron	FY25 China + Hong Kong 10.1%	Based on customer headquarters location

Samsung and SK Hynix are directly exposed if Chinese demand contracts. Micron’s direct exposure is comparatively low, but it cannot escape the impact if global pricing declines. Revenue-share figures alone are insufficient to assess China risk.

6.2 China Is a Commodity Memory Competitor

CXMT has announced mass production of LPDDR5X at 8,533 Mbps and 9,600 Mbps, with its 10,667 Mbps product reportedly in customer qualification. Its estimated share of global DRAM revenue for Q1 2026 stands at approximately 8%. However, an Apple supply agreement, broad adoption in flagship global products, and HBM yield have not been confirmed. CXMT LPDDR5X Official Announcement

The near-term threat is concentrated in two channels rather than HBM displacement:

Localization of commodity DRAM and NAND in Chinese smartphones and PCs.
Using CXMT as a fourth-supplier card in price negotiations with the established memory trio.

If volume displaced from China migrates to other markets, it could pull down global contract prices for Samsung, SK Hynix, and Micron. This price contagion may matter more as a second-order effect than a direct decline in China revenue. A detailed exposure breakdown is available in China Memory Localization and the Big Three.

6.3 China Can Be Segmented by Policy

On July 16, members of the House Select Committee on China sent a letter to the Department of Commerce calling for tightened controls on YMTC, accelerated review of CXMT for Entity List inclusion, restrictions on Chinese DRAM and HBM procurement in AI systems, data centers, federal IT, and critical infrastructure, and coordination with South Korea, Japan, and the EU. This is a policy demand — not an implemented purchasing ban. House Select Committee on China Official Letter

If enacted, the global memory market could bifurcate along the following lines:

Market	Possible Structure
United States and allies	Verified supply chains, higher prices, market-share defense by the memory trio
Chinese domestic market	Localization, lower prices, medium-term oversupply risk

Micron stands to benefit most directly from U.S. policy. Samsung and SK Hynix carry both the advantage of being allied-nation suppliers and the exposure of their China businesses. This policy factor is less a catalyst that directly lifts 2026 HBM earnings and more a variable that reduces the China-oversupply discount rate priced into 2028 valuations.

7. Kimi K3 and Efficiency: Demand Shifting, Not Demand Destruction

According to Kimi’s official announcement, K3 has 2.8 trillion total parameters, activates 16 of 896 experts per token, and supports a 1 million-token context window. The sparse expert architecture reduces compute per request, but still requires large memory and high-bandwidth networking to store all weights and route expert selection. Kimi K3 Official Announcement

The semiconductor impact of efficiency gains is better understood through the following equation:

Total semiconductor demand = silicon cost per task × total number of tasks × fraction self-hosted

Sparse architectures, caching, and quantization reduce GPU time and HBM footprint per task.
Falling API prices increase AI usage volumes and the number of agents.
Open-weight models may increase self-hosted server deployments by enterprises and governments.
Memory tiering shifts some HBM workload onto server DRAM, eSSD, and remote caches.

Current firm orders and CAPEX data suggest that volume growth is outpacing efficiency gains. Over a longer horizon, however, the volume elasticity of server DRAM, eSSD, networking, and power may prove more favorable than the pricing power of premium GPUs and HBM. Efficiency is less a variable that “reduces total semiconductor consumption” and more one that reshapes “which semiconductors capture the revenue.”

Two caveats apply here. The KV-cache savings rates and theoretical throughput from Kimi’s linear research models cannot be directly applied to K3 in production. Additionally, self-reported performance benchmarks should not be weighted equally with independent verification. Weights, licensing terms, independent throughput results, and actual tokens and total cost per task require confirmation.

8. Five Clocks — When They Are Watched Together, the Contradictions Resolve

Clock	This Week’s Observation	Investment Interpretation
Pricing	Sharp 2Q DRAM and NAND price increases; 3Q pace of gains decelerating	2026 earnings look strong, but the rate of increase may slow
Investment	TSMC expanding investment; ASML orders and capacity growing	Current demand confirmed while 2027–2028 supply response is booked
Construction	Power, land, and packaging bottlenecks persist	Announced investment lags before translating into actual production ramp
Monetization	AI usage rising; customer costs and financing charges also rising	AI gross margin and free cash flow require scrutiny alongside revenue
Share price	Sluggish reaction to positive news; concentrated supply-demand shock in Korea	Market is discounting duration and positioning rather than current earnings

The pricing and investment clocks are running fast, while the construction and monetization clocks are lagging. The share-price clock is the forward-looking reflection of whether that gap narrows or widens ahead. That is why strong results and raised CAPEX guidance can trigger selling on the same day.

The “55% industry strength / valuation weakness coexistence” from the prior piece and “P2 at 40%” below are not two different readings of the same table. The 55% was a broad category in which multiples remain compressed even as the industry stays strong. The more granular distribution here splits that category further into the early phases of P2 and P3. Because new data from TSMC and ASML confirmed current demand while simultaneously expanding the implied supply response, a portion of the long-term scarcity probability has been shifted toward supply normalization.

9. Scenario Update

The probabilities below are judgment values reflecting information confirmed as of July 19 — not historical base rates.

Scenario	Probability	Conditions for Validity	Expected Market Behavior
P1. Sustained Scarcity	25%	AI usage, CAPEX, and utilization rates continue to overwhelm supply and efficiency gains; HBM4 pricing and orders hold	High-beta memory names such as SK Hynix and Micron, alongside packaging and equipment, outperform
P2. Strong Earnings, Compressed Multiples	40%	Earnings remain strong in 2026–2027 but concerns over 2028 supply and customer profitability persist	Range-bound trading and elevated volatility despite earnings growth; Samsung Electronics carries relative advantage
P3. Efficiency and Supply Normalization	25%	2027–2028 capacity expansion, Chinese commodity supply, customer price resistance, and declining AI costs converge	HBM scarcity premium narrows; equipment and materials orders peak before memory itself
P4. Financial or Demand Dislocation	10%	Big-Tech CAPEX cuts, a credit event, rising inventories, and simultaneous contract price and EPS downgrades	Sector-wide semiconductor decline; cash and non-semiconductor diversification become more important

The previous detailed distribution was 35% / 40% / 15% / 10%. This revision moves 10 percentage points from P1 to P3. TSMC and ASML order data did not raise the probability of P4. Conversely, equipment capacity expansion and the pricing burden on customers have increased the probability of supply and efficiency normalization relative to prolonged scarcity.

10. Roles by Stock and Value Chain

Samsung Electronics: Risk-Adjusted Core

Samsung absorbs commodity DRAM and NAND price appreciation across a broad market share base while retaining the option value of catching up in HBM and recovering its foundry business. It also holds relative defensive characteristics should open models and low-cost inference proliferate, shifting some HBM workload toward commodity memory and storage.

Key risks include exposure to China and commodity memory, uncertainty around HBM4 customer qualification, and the high capital burden of foundry operations. Samsung offers more defensive characteristics than SK Hynix in P2 and P3, but diversification does not equal profitability if HBM catch-up has not been validated in earnings.

SK Hynix: Pure Beta on Sustained Scarcity

SK Hynix’s leading HBM market share, co-development relationships with customers, and long-term supply agreements give it the strongest earnings leverage under P1. Conversely, it also carries the highest sensitivity to earnings and multiple compression if supply expansion and efficiency gains accelerate under P3.

Additional judgment requires tracking HBM4 and HBM4E customer qualifications, yield, average selling price terms embedded in long-term contracts, and foreign-investor and program-trading flows alongside relative strength. Business quality and price-embedded margin of safety must be assessed separately.

Micron: Intersection of U.S. Policy and Pure-Play Memory

Micron is the most direct beneficiary of U.S.-based manufacturing, policy support, and restrictions on Chinese memory procurement. It offers a clean read-through to the HBM, DRAM, and NAND upcycle. However, its limited business diversification means it is also directly exposed to any normalization in global average selling prices. Even with a low China revenue share, Micron cannot easily avoid the impact if surplus Chinese-origin volumes depress global pricing.

SanDisk: Tactical Beta on NAND and eSSD

SanDisk is directly exposed to NAND supply constraints and AI server eSSD demand. It stands to benefit most if low-cost inference and memory tiering shift incremental storage volumes away from HBM. At the same time, NAND exhibits greater price volatility than DRAM and carries more direct YMTC exposure, giving it a more tactical character relative to core positions.

TSMC and ASML: Demand Validators and Leading Indicators of Future Supply

TSMC demonstrates the quality of AI demand and the depth of the leading-edge foundry moat. ASML is the earliest observable data point for customers’ long-term capacity expansion intentions. Both companies’ results serve as supporting evidence for the memory thesis, but they are not simple coincident indicators. High CAPEX and growing equipment supply signal not only future memory demand but also future supply expansion.

Korean Equipment and Materials: Actual Orders and Recurring Revenue Over Published Plans

Category	Candidate Names	Data to Confirm	Current Role
Recurring consumables	Hana Materials, Wooldex	Customer utilization rates, OEM channels, revenue and operating margin correlation	Priority research group
Process equipment	KC Tech, Nextin	Actual orders, customer qualification, backlog	Conditional confirmation group
HBM back-end	Techwing	Additional Cube Prober orders and production revenue	Event confirmation group
Infrastructure construction	Hanyang E&E, others	Major contract wins, order-to-revenue ratio, margins	Order-event group rather than core

The confirmation sequence for investments is: capacity announcement → actual order placement → backlog build → revenue recognition → recurring consumables. Equipment names may peak in earnings earlier than memory production itself. Rather than buying broadly on fab plans alone, the focus should narrow to companies where actual orders and margins are confirmed.

11. Counter-Arguments: Where This Thesis Could Be Wrong

Demand Evidence May Have Been Double-Counted

Orders from TSMC, ASML, and the memory trio may all originate from the same Big-Tech capital expenditure plans. Adding orders across multiple tiers of the supply chain as separate demand signals overstates actual end demand. Power connection timelines, GPU utilization, server shipments, and cloud AI gross margins are the terminal verification data points.

The 2030 Demand Model May Underestimate Efficiency Gains

If any one of — token count, model size, context length, or HBM retention rate — is reduced, a shortage persists. But if all four decline simultaneously, a 2030 oversupply scenario becomes plausible. The directional case for long-term shortage and the numerical magnitude of that shortage must be assessed separately.

CAPEX May Grow Faster Than Returns

Suppliers must spend more on new fabs and packaging even when order books are full. Customers must absorb data center depreciation and interest expense. If revenue growth fails to translate into free cash flow growth, high earnings may attract depressed multiples.

China Policy May Not Be Implemented

The House Committee letter is a policy signal — not an executive order or coordinated allied action. If CXMT enters only a limited set of consumer products within China and enforcement remains loose, pricing leverage for the incumbents may erode before any meaningful market-share shift occurs.

Open-Model Company Disclosures May Not Be Independently Reproducible

Kimi K3’s official specifications are confirmed, but throughput, real-world token consumption, and total serving cost require independent verification. Conversely, if efficiency gains arrive faster than expected and usage elasticity is lower than assumed, the HBM scarcity premium could compress more severely.

The Entire Price Decline May Have Been Attributed to Forced Liquidation

Leverage unwinding and program selling explain the magnitude of the drop. But if prices fail to recover even in the face of positive news, the market may be actively lowering its 2028 earnings and normalized-multiple assumptions. A supply-demand explanation cannot substitute for validating earnings durability.

12. Observable Conditions That Would Change the Thesis

Rather than any single article or a one-day bounce, the key is whether two independent data axes move together.

Conditions That Raise the Probability of a Bullish Outcome

Big-Tech CAPEX growth and improvement in AI and cloud gross margins are confirmed simultaneously.
Data center power connections, GPU utilization rates, and server shipments all rise together.
HBM4, server DRAM, and eSSD prices, along with 12-month forward EPS, continue to be revised upward.
Samsung Electronics, SK Hynix, and Micron rally on positive news and recover relative strength versus the semiconductor index.
Foreign-investor and program selling decelerates, and a trend recovery accompanied by volume emerges.

Conditions That Raise the Probability of P3 Normalization

2027–2028 equipment delivery and new bit capacity come online faster than planned.
HBM yields and packaging productivity improve sharply while supply from all three incumbents expands simultaneously.
Following memory price increases, customers reduce their content per device and purchasing deferrals become widespread.
CXMT DDR5 and LPDDR5X customer qualifications and sales outside China increase.
Server DRAM and HBM contract price growth rates and 2028 EPS are simultaneously revised lower.

P4 Warning Conditions

U.S. Big-Tech companies cut CAPEX or cancel or delay data center projects.
Credit spreads and refinancing risk for AI infrastructure customers spike sharply.
DRAM and NAND inventory builds and contract price declines occur simultaneously.
12-month forward EPS for Samsung Electronics, SK Hynix, and Micron are all revised downward together.
Semiconductor relative strength and market breadth (advance-decline) both deteriorate simultaneously.

13. Data to Track from Late July Onward

Timing	Data to Track	Judgment That Changes
Late July	U.S. Big-Tech earnings and conference calls	CAPEX, AI gross margin, free cash flow, memory price resistance
Late July	Additional Kimi K3 technical disclosures and independent evaluations	Which is larger — efficiency gains or usage growth
Late July	Samsung Electronics and SK Hynix detailed IR disclosures	HBM4, long-term contracts, DRAM and NAND ASP, shipments, CAPEX
August onward	3Q contract pricing and distribution inventory levels	Whether the pace of price gains has peaked and pre-buying is normalizing
Quarterly	ASML bookings, TSMC CAPEX, memory equipment orders	Speed of 2027–2028 supply response
Ongoing	CXMT and YMTC regulatory actions and customer qualifications	Chinese supply’s price contagion to global markets and market bifurcation

14. Final Judgment

This week served as the first comprehensive stress test of the memory supercycle narrative. Share prices moved sharply, but TSMC and ASML results, DRAM and NAND pricing, and server memory demand all tilted the weight of evidence toward 2026–2027 physical shortages remaining intact.

At the same time, the same data made 2028 risks more clearly defined. Equipment capacity and CAPEX are expanding, long-term HBM supply agreements slow the pass-through of price increases, customers must absorb elevated memory prices alongside data center financing costs, and Chinese commodity memory and open-model efficiency gains will exert pressure on 2028 pricing and multiples.

The choice between “the boom is over” and “this is a permanent supercycle” is therefore a false one. The current base-case path is as follows:

2026–2027: Physical shortage and high earnings
-> 2027–2028: Equipment, capacity, and yield response
-> Pace of price increases decelerates
-> Absolute earnings remain above historical levels, but multiples compress
-> Return dispersion across names driven by product mix, contract structure, and China exposure

From this point, the central task is not to re-prove that AI demand exists. It is to measure simultaneously the speed of the supply response, the monetization trajectory for customers, contract quality, 2028 earnings estimates, and how share prices react to positive news. Earnings today are speaking to 2026 and 2027. The market is already asking about 2028.

Key Sources

Limitations of Public Data

The long-term contract pricing formulas for HBM from Samsung Electronics, SK Hynix, and Micron — including floor and ceiling prices, minimum purchase volumes, cancellation clauses, and customer-specific credit terms — have not been disclosed publicly.
There are no reliable official figures for CXMT’s HBM yield or its progress toward Apple qualification.
Full unit cost by model and actual per-accelerator serving costs for Chinese AI API providers are not publicly reported.
The 26.7 EB HBM demand figure for 2030 is not a consensus forecast but a scenario constructed by combining several bullish assumptions.
The July 19 channel checks on server DRAM and HBM4 could not be independently sourced from primary documents and are used solely for directional reference.
The information cutoff date is July 19, 2026. Scenarios should be revisited following earnings releases from U.S. large-cap tech companies and memory makers expected in late July.

This article is an informational analysis based on publicly available sources and does not constitute a recommendation to buy or sell any specific security. Prices, supply-demand dynamics, earnings estimates, and policy conditions are subject to change; readers should verify the latest disclosures and assess findings against their own risk tolerance.

What China's Open-Model Convergence Actually Changes: Value-Chain Redistribution, Not Demand Collapse

Fri, 17 Jul 2026 22:00:00 +0900

Context This piece is a follow-up to Kimi K3 Resets the AI Price Curve. Where that piece verified the pricing and architecture of a single model, this one expands the lens to how the entire Chinese open-model ecosystem redistributes the semiconductor and Big Tech value chain, and how US policy and US-China tension reshape that reading. It pairs well with The Real Debate in Semiconductors, Are Semiconductors Cyclical, and What Is Fair Value?, and CXMT IPO And Memory Price Risk. Related hubs are the AI HBM Hub and the Exclusive Analysis Hub.

TL;DR

Chinese open models converging in performance looks more like value-chain redistribution than an AI demand collapse. The monopoly value of model APIs and leading-edge GPUs falls, while cheaper inference lifts usage and can raise the value of memory, storage, networking, power, and cloud distribution.
Relative preference splits this way. Within Korean memory, Samsung Electronics > SK Hynix; within US semiconductors, Micron and SanDisk > NVIDIA; within Big Tech, Meta and Amazon > Google and Microsoft > pure model vendors. [Inference: relative judgment]
Chinese open models have proven falling compute and memory cost per unit of intelligence. They have not proven a decline in total silicon spend. TSMC, if anything, raised its 2026 capex from $52 billion-$56 billion to $60 billion-$64 billion.
In HBM, CXMT trails the leading three by 1.5 to 2 product generations and by a 2- to 3-year commercialization gap. The 2028 threat is therefore a conditional option, not a present-day supply shock.
The technical diffusion of Chinese models and the revenue diffusion of Chinese API vendors are two different things. The most realistic path is rehosting Chinese models on AWS, Azure, or enterprise VPCs and selling them through Western security and contracting frameworks. [Analysis scope]

Key Framing

Chinese models can shake the pricing of US models, but they do not immediately collapse demand for AWS, Azure, US chips, or Korean memory. The direct casualty is the margin of closed-model APIs. What survives longest is usage-based infrastructure: the cloud distribution and security layer, along with memory, networking, and power.

1. Getting the Facts Straight First

The ecosystem direction is real, but not every new model has disclosed its training hardware. That distinction matters.

DeepSeek-V3 was trained on 2,048 NVIDIA H800 GPUs using 2.788 million GPU-hours. [Fact: DeepSeek V3 technical report] The H800 is export-restricted, but it is not a cheap consumer GPU. Kimi K3, by contrast, has disclosed 2.8 trillion parameters, a 1M-token context window, and 2.5x higher scaling efficiency versus K2, but the full technical report and training hardware are scheduled for release on July 27. [Fact: Kimi official announcement]

So the claim that “Kimi K3 was also trained on cheap NVIDIA GPUs” cannot yet be confirmed. [Blocked]

The Efficiency Gains Are Real

DeepSeek V4-Pro activates only 49 billion of its 1.6 trillion parameters. At the 1M-token range, per-token compute is 27% and KV cache is 10% of V3.2’s levels. [Fact: DeepSeek V4 model card] That is direct evidence that GPU and HBM use per token can fall sharply while performance holds.

But So Is the Other Side

Huawei’s CloudMatrix384 pools 384 Ascend 910C chips to serve DeepSeek-R1. By JPMAM’s tally, CloudMatrix uses 49TB of HBM and 599kW of power, versus 21TB and 145kW for the comparison system, GB300 NVL72. [Fact: CloudMatrix paper, JPMAM comparison]

It is not an apples-to-apples performance comparison, but the direction is clear: China compensates for a weaker single chip with more chips, more memory, and more power. A cheaper individual chip does not necessarily mean less total silicon, networking, or power. [Inference: structural reading]

2. The Core Equation: What to Actually Watch

The answer to this debate is not in benchmark scores. It is in two equations.

Total compute demand = total tokens × compute per token

Total memory demand = total tokens × memory use per token
 + model, KV, and retrieval data storage

If open models cut token prices 70% and usage rises 5x, total demand increases even after the efficiency gain. Conversely, if usage merely doubles while per-token HBM falls 70%, HBM demand declines.

So the indicator to watch going forward compresses into one question: by how much does the token growth rate outpace the decline rate in memory per token?

The Verdict So Far

Total spend is determined by the following equation.

Total silicon spend = number of training runs × cost per run
 + inference tokens × cost per token

On its 2Q26 call, TSMC raised its 2026 capex from $52 billion-$56 billion to $60 billion-$64 billion. Microsoft, too, raised inference throughput 40%, yet large-customer token use still rose 30% quarter-over-quarter, and it held roughly $190 billion in 2026 capex. [Fact: TSMC 2Q26 call, Microsoft FY26 Q3 call]

So far, usage growth is beating efficiency gains. [Inference: data synthesis]

3. Semiconductor Impact: It Diverges by Stock

Stock / Group	Stock-Price Impact	Key Interpretation
Samsung Electronics	Positive	Broadest beneficiary, capturing not just HBM but server DRAM, NAND/eSSD, and general-purpose memory
SK Hynix	Mixed-to-positive	Token growth helps, but MoE and KV compression cut HBM per token, pressuring the scarcity multiple
Micron	Positive	Combined DRAM, HBM, and NAND exposure across the US supply chain; broad-memory upside similar to Samsung
SanDisk	Positive	Local deployment, model weights, and growing RAG/cache use flow into eSSD and NAND demand
NVIDIA	Negative near-term, mixed long-term	China share and top-tier GPU monopoly value decline; offset by rising total tokens and H200 shipments
AMD	Relatively positive	Open models make it easier to migrate across heterogeneous hardware; ROCm and actual serving share are the key variables
Broadcom / Marvell / Arista	Positive medium-term	Rising demand for Western custom ASICs, Ethernet, optical, and SerDes
TSMC	Neutral-to-positive	Training-chip demand from NVIDIA, AMD, and ASICs holds, but Chinese inference shifts to Ascend and SMIC

Why Samsung Has the Relative Edge Over Hynix

Among the same three memory makers, Samsung Electronics and SK Hynix diverge in direction. Cheap inference and open-model diffusion do not just lift HBM; they raise the total volume of server DRAM, NAND, and general-purpose memory. Samsung Electronics, with its broader exposure, captures more of that diffusion.

MoE and KV compression, conversely, reduce HBM per token. Hynix, with its concentrated HBM exposure, receives both the benefit of rising tokens and the burden of falling HBM per token at the same time. It is a structure where the scarcity multiple, not absolute demand, comes under pressure first. [Inference: exposure structure analysis]

Samsung Electronics does carry an offsetting risk, however: it is the first to be exposed to CXMT’s ramp-up of general-purpose DRAM output.

NVIDIA and H200

The US recently began shipping limited volumes of H200 to China, but the quantity is still small. It is positive for NVIDIA’s near-term revenue, but not enough to reverse the shift toward Huawei-centered domestic infrastructure. [Fact: July 2026 Reuters reporting] [Inference: impact judgment]

4. Big Tech Impact

Company	Verdict	Reason
Meta	Most positive	Open-ecosystem strategy is validated, and AI returns are captured through advertising and recommendation rather than APIs
Amazon	Positive	Sells AWS inference, storage, and networking regardless of which model wins
Google	Mixed-to-positive	TPU and cloud benefit, but Gemini API price premium comes under pressure
Microsoft	Mixed	Azure usage rises, but OpenAI model rent and capex payback are pressured
Apple	Positive	Cheaper small and open models lower on-device and Private Cloud costs
OpenAI / Anthropic	Negative	Shrinking performance gap and API price premium pressure high valuations and fundraising

Contrary to a common misconception, Meta is not the biggest casualty. It makes money from advertising and recommendation rather than model sales, and it uses open models to cut costs, so it stands to relatively benefit. The burden falls instead on capex and depreciation. [Inference: business model analysis]

The Lesson from the 2025 DeepSeek Shock

During the 2025 DeepSeek shock, NVIDIA lost 17% and roughly $593 billion in market capitalization in a single day. The initial reaction was a broad sell-off across GPUs, power, and data-center infrastructure, but it partly rebounded soon after. [Fact: 2025 market reporting]

This time, too, near-term stock prices are likely to react to efficiency fears, while medium-term earnings react to rising usage. [Inference: historical pattern]

5. Verdicts on the Chinese Open-Model Thesis

Judging the claims coming from both the bull and bear sides, one by one, looks like this.

Claim	Verdict
Frontier performance requires top-tier GPUs	Weakened, but not disproven
AI intelligence is a scarce resource	Largely collapsed at the level of raw model and token pricing
Capital scale is the moat	The pretraining moat weakens; the moat shifts to deployment, data, power, and distribution
Open source has zero marginal cost	Only the weight price approaches zero; inference cost is ongoing
A $1 billion training run has overtaken $100 billion of investment	An inaccurate claim comparing two different cost categories

The last item is especially often misused. Training cost and total infrastructure investment are not comparable categories.

6. How Far Has CXMT’s HBM Actually Gotten

This is the part of the China-threat discussion that gets exaggerated most often. Breaking it down gate by gate reveals the reality.

Gate	Current Verdict	Basis for Judgment
DRAM cell process	Commercialized, improving fast	Selling DDR5 and LPDDR5X; Apple is also testing DRAM for China-market products
TSV stacking	Small-volume HBM2, early HBM3	HBM2 production has been reported, but volume and yield are undisclosed
Packaging and base die	Under construction	The domestic ecosystem is forming, but mass-production yield, thermal, and reliability data are absent
Customer qualification	HBM unconfirmed	The Tencent contract and Apple testing are evidence of general DRAM capability, not HBM
Distance from the leaders	1.5-2 generations	The leading three are ramping HBM4; CXMT has not even verified HBM3 mass production

As of July 17, 2026, CXMT’s official published product lineup includes only DDR5, LPDDR5/5X, DDR4, and LPDDR4X, and no HBM. TrendForce likewise classifies CXMT’s HBM3 as still in early verification, and assesses that technical barriers and domestic-equipment requirements are delaying mass production. [Fact: CXMT official materials, TrendForce]

Capacity estimates also diverge. A plateau near 240,000 wafers per month conflicts with a year-end forecast of 350,000, and because the 350,000 figure comes from a private model rather than company guidance, it is hard to treat as a confirmed number. [Blocked]

How the Legacy-DRAM Hypothesis Needs to Be Revised

2026-2027: CXMT’s HBM investment delays the easing of Chinese legacy-DRAM supply.
2028: If HBM3/3E secures Chinese accelerator customer qualification and meaningful volume, it could begin displacing the older-generation HBM market within China first.
2029 and beyond: Only once HBM4, base die, and packaging yield catch up does it directly pressure the global leading market and Hynix’s margins.

In other words, rather than “CXMT is making legacy supply tighter right now,” the more valid framing is “CXMT is not freeing up as much legacy supply as expected." [Inference: stage-by-stage judgment]

7. How US Policy Reshapes This Reading

The US is treating AI less as a commercial technology and more as core infrastructure for the allied bloc. That is why a purely technical analysis cannot supply the full answer.

The US AI Action Plan and Executive Order 14320 state explicitly that the US intends to export a full-stack American AI system, bundling hardware, cloud, networking, models, and applications, to allied nations, while reducing technological dependence on adversary nations. Even if a Chinese model is superior on performance and price, the US stack retains a policy advantage in allied-nation government procurement and critical industries. [Fact: White House AI Action Plan, EO 14320]

Yet This Is Not a Full Decoupling

In January 2026, the US BIS moved to review H200 and MI325X exports to China case by case, under approved-customer and security conditions. It is a compromise that keeps China partly inside the US chip ecosystem while retaining control. [Fact: BIS 2026-01-13]

Nor is the use of Chinese models by US private companies fully banned. The No DeepSeek on Government Devices Act is still only at the bill-introduction stage. Australia’s government, however, has ordered DeepSeek removed from government systems, and Italy’s privacy authority has restricted processing of user data. [Fact: official actions by country]

De facto bans from procurement and security departments are likely to take effect before legislation does. [Inference: policy sequencing]

8. Can Western Enterprises Actually Use Chinese Model APIs?

Diffusion potential differs completely by pathway. This table is the answer to the question.

Adoption Method	Diffusion Potential	Judgment
Direct calls to mainland China-hosted APIs	Low	Data, jurisdiction, and procurement risk
Qwen’s US, EU, Japan, and Singapore APIs	Medium	Data localization is possible, but Chinese-vendor risk remains
Chinese models rehosted by AWS or Azure	Medium-high to high	Western cloud providers hold the contracting, security, and data control
Enterprise VPC or on-premises open weights	High	No need to send data to Chinese servers
Government, defense, and critical infrastructure	Very low	Procurement restrictions can extend even to model lineage

The most realistic pathway looks like this.

Chinese model developed
→ Rehosted on AWS, Azure, or enterprise VPC
→ Sold through Western security, contracting, and audit frameworks

In other words, the technical diffusion of Chinese models and the revenue diffusion of Chinese API vendors are separate matters. [Inference: pathway analysis]

The Limits of Direct APIs

DeepSeek’s Privacy Policy states that it collects, processes, and stores personal data, including input data, directly in China, and may use it to improve its service and models. [Fact: DeepSeek Privacy Policy] That is a disqualifying condition for companies handling source code, customer information, healthcare or financial data, or export-controlled technology.

Qwen is somewhat different. Alibaba Cloud Model Studio offers US, German, Japanese, and Singapore regions, can restrict data and inference to a specific region, and states explicitly that it does not use customer data for model training. [Fact: Alibaba Cloud region and security policy] Direct API adoption could therefore grow in non-regulated industries and in Southeast Asia, the Middle East, and Latin America.

Data localization and SOC 2, however, do not eliminate the risk of service disruption, sanctions, or procurement exposure arising from US-China tension. [Inference: residual risk]

9. Revising the Thesis: What Changes and What Stays

First, the case for a US hyperscaler collapse weakens. AWS and Azure directly host DeepSeek, providing data isolation, SLAs, and security assessments. US clouds can absorb the cost innovation of Chinese models and turn it into revenue. [Fact: AWS and Azure official announcements]

Second, pricing pressure hits closed-model APIs first. It burdens the token prices and margins of OpenAI, Anthropic, and Google, but inference volume and AWS/Azure usage can still rise. It is relatively favorable for Meta’s open-model strategy.

Third, the US-China split supports total infrastructure investment. Both blocs build out duplicate accelerators, memory, networking, and power grids. That lowers the likelihood that efficiency gains translate directly into a decline in global silicon spend.

Fourth, SK Hynix’s HBM premium can persist longer within the allied bloc. CXMT’s HBM is more likely to penetrate Chinese accelerators and Chinese data centers first. Adoption by Western CSPs requires policy and supply-chain certification on top of technical qualification. That said, export controls accelerate China’s push for self-sufficiency, which is a risk to Hynix’s China-market share from 2028 onward.

Fifth, Samsung Electronics is a relative hedge. Cheap inference and open-model diffusion can raise the total volume of server DRAM, NAND, and general-purpose memory, not just HBM. On the other side is the risk of being first exposed to CXMT’s ramp-up of general-purpose DRAM output.

10. The Order of Casualties and Beneficiaries

Greatest casualty: the excess profit of closed-model APIs, and highly leveraged GPU-leasing operators
Intermediate risk: heavily front-invested, customer-concentrated operators such as Oracle
Meta: not the biggest casualty. It uses open models to cut costs, so relative benefit is possible. The burden falls on capex and depreciation
NVIDIA: the multiple and product mix come under pressure before near-term EPS does. Displacement inside China is a risk, but the CUDA, networking, power-efficiency, and development-timeline moats remain
Memory: the base case is not a decline in absolute HBM demand, but a narrowing HBM scarcity premium plus tier expansion into DDR/CXL/eSSD

11. Scenario Update

It makes sense to provisionally revise the probabilities used in Are Semiconductors Cyclical, and What Is Fair Value?

Scenario	Prior	Revised
Excess demand persists	30%	30%
Asset re-concentration	40%	40%
Supply/efficiency normalization	20%	25%
System demand short-circuit	10%	5%

Chinese model performance lowers the probability that AI demand itself disappears (the short-circuit scenario, from 10% to 5%). In exchange, efficiency gains and Chinese-made hardware lower the scarcity premium of NVIDIA and HBM (the normalization scenario, from 20% to 25%).

The timeline scenarios also hold: usage beats efficiency through 2027 at 70%, mix and pricing normalize from 2028 onward at 25%, and total capex contracts at 5%. However, the driver of the 70% scenario shifts from a single global ecosystem to dual investment across the US bloc and the China bloc. [Inference: scenario recalibration]

12. What Would Change This Judgment

The Kimi K3 technical report (July 27): once training hardware is disclosed, the truth of the “frontier training on cheap GPUs” claim will be settled
Token growth rate versus the decline rate in memory per token: the gap between these two values is the real answer to this debate
CXMT’s HBM3E mass-production qualification and actual packaging volume: if confirmed, both Hynix’s 2028 EPS and its multiple need to come down together
Slowing HBM pricing and content growth: the first signal of a narrowing scarcity premium
Expanding rehosting of Chinese models by Western CSPs: evidence that cloud-distribution value is rising
Actual enforcement of US procurement and security regulation: the scope of the de facto ban that operates ahead of legislation

This is not the stage to conflate terminal risk with current-period earnings. Hynix’s 2028-and-beyond outlook can be adjusted once CXMT’s HBM3E mass-production qualification is confirmed. [Inference: sequencing of judgment]

Closing

Returning to the question, the answer is this.

It is a fact that Chinese open models are converging on the US frontier, and it is a fact that cost per unit of intelligence has fallen sharply. But that does not mean total silicon spend is declining. TSMC’s capex increase and Microsoft’s 30% rise in tokens are the answer so far.

US policy substantially reshapes this reading. The US-China split produces duplicate investment across both blocs, supporting total infrastructure demand, and within the allied bloc, it protects the position of the US stack and Korean memory by policy.

Western enterprise adoption of Chinese model APIs must be viewed by separating the model from the vendor. The technology spreads as open weight, but the party selling it is more likely to be AWS, Azure, and enterprise VPCs than the Chinese API vendors themselves.

So the conclusion is redistribution. What collapses is the excess profit of closed-model APIs. What remains is usage-based infrastructure.

This post synthesizes public papers (DeepSeek V3/V4, Huawei CloudMatrix384), official company announcements (Kimi, the TSMC 2Q26 call, the Microsoft FY26 Q3 call, CXMT, Alibaba Cloud, DeepSeek’s Privacy Policy), US government materials (the AI Action Plan, EO 14320, BIS), regulatory actions by various countries, and market research (TrendForce, JPMAM). Kimi K3’s training hardware remains unconfirmed until its technical report is released on July 27, and CXMT’s capacity estimates and the scenario probabilities are author estimates as of the time of writing, not company guidance. The stocks mentioned are examples used to illustrate value-chain structure and are not a recommendation to buy or sell any specific security. Investment decisions and responsibility for them rest with the individual investor.

Kimi K3 Resets the AI Price Curve: From Kimi Linear to HBM and Big Tech Strategy

Fri, 17 Jul 2026 12:31:36 +0900

Kimi K3 launched on July 16, 2026 at $3 per million uncached input tokens and $15 per million output tokens. That is not the familiar ultra-low-price positioning of a Chinese challenger. It matches Claude Sonnet 5’s standard price and sits 40% below GPT-5.6 Sol on input and 50% below it on output. Moonshot AI is positioning K3 as a primary enterprise model, not a budget fallback.

The architecture is designed to support that claim. K3 has 2.8 trillion total parameters but effectively activates 16 of 896 experts per token. Kimi Delta Attention controls the cost of long context, Attention Residuals selectively retrieve earlier representations across depth, and quantization-aware training uses MXFP4 weights with MXFP8 activations. Moonshot recommends a supernode with at least 64 accelerators.

That combination creates a two-sided semiconductor outcome. Memory and compute per token can fall while the number of institutions able to deploy a frontier-class model, and the amount of work they run, can rise. K3 is both an efficiency technology and an infrastructure workload.

Related reading: AI token value and memory value capture / AI token futures and cost per token / US-China divergence in agentic inference infrastructure / Cross-checking the 2030 HBM shortage model

Executive Summary

K3 combines 2.8T parameters, native vision and a 1M-token context window. Its products and API are live, but as of July 17 the full weights, technical report and license have not been released. The open-weight thesis must be verified after the July 27 deadline.
Moonshot’s official GDPval-AA v2 score is 1,668, not 1,687. AA-Briefcase is 1,548. BrowseComp 91.2 uses context compaction; the no-compaction 1M-context result is 90.4. The results are strong, but mixed harnesses and company-run evaluations require independent replication.
Kimi Linear’s claims of up to 75% lower KV cache and up to 6.3x theoretical decoding throughput at 1M context come from a 48B-total, 3B-active research model. They should not be presented as measured K3 API performance.
K3 is not a low-cost model. It matches Sonnet 5’s standard price, is 50% more expensive than Sonnet’s temporary launch promotion, and is cheaper than GPT-5.6 Sol. Because only max reasoning is currently available and independent tests show high output-token use, cost per completed task may be less attractive than the rate card implies.
The semiconductor impact pits lower compute and KV cache per request against more self-hosted deployments and higher total workload. Server DRAM and enterprise SSDs are the cleanest second-order beneficiaries because long-lived agent context spills below HBM into disaggregated cache tiers.
The model and cloud layers experience opposite economics. OpenAI, Anthropic and Gemini API pricing face pressure, while Azure, AWS and Google Cloud can monetize K3 and other models through compute, storage and networking. Meta must defend US open-weight leadership.
The July 27 proof points are the license, full weights, external evaluation, throughput on NVIDIA and AMD, support on Chinese accelerators, actual output tokens per task, vLLM compatibility and cloud catalog adoption.

1. Correcting the Numbers Before Drawing Conclusions

Several numbers changed as the launch circulated through social media. The official values matter because some differences alter the investment interpretation.

Item	Official value	Investor interpretation
Total parameters	2.8T	Total model size, not per-token active compute
Expert routing	16 of 896	Extremely sparse MoE; routing and communication become first-order constraints
Context	1M tokens	Useful for repositories and research, but task cost depends on output length and cache hits
GDPval-AA v2	1,668	The official table does not show 1,687
AA-Briefcase	1,548	Above GPT-5.6 Sol at 1,495, below Claude Fable 5 at 1,583
BrowseComp	91.2	Uses compaction starting at 300K tokens
BrowseComp without compaction	90.4	The cleaner result for the native 1M-context claim
Product availability	Web, Work, Code and API live	Commercial testing can begin now
Weights	Promised by July 27	License and complete artifacts remain unverified as of July 17
Reasoning effort	Max only	Low-cost modes are not yet available

Moonshot explicitly states that K3 still trails Claude Fable 5 and GPT-5.6 Sol in overall user experience. At the same time, it reports frontier-level performance across coding, knowledge work and multimodal benchmarks. The credible interpretation is not that China has conclusively taken the lead. It is that a Chinese model has entered the performance band immediately below the strongest proprietary systems while promising a 1M context and downloadable weights.

The benchmark table is not a single controlled tournament. K3 runs at maximum reasoning effort. Depending on the task, models use Kimi Code, Claude Code or Codex, and some competitor scores are the best results across harnesses or are imported from external leaderboards. Independent reproduction remains necessary.

Still, Terminal-Bench 2.1 at 88.3, FrontierSWE at 81.2, SWE Marathon at 42.0, AutomationBench at 30.8, GPQA-Diamond at 93.5 and MMMU-Pro at 81.6 indicate that K3 is designed for long-horizon tool use and coding, not merely chat. The more important signal is the package: near-frontier capability, 1M context and promised open weights.

2. How a 2.8T Model Becomes Deployable

2.1 Total parameters are not active parameters

Computing all 2.8T parameters for every token would be prohibitively expensive. Stable LatentMoE effectively activates 16 of 896 experts, or about 1.8% of the expert pool. The technical report is needed to establish exact active parameter counts, but total parameters clearly do not equal per-token compute.

Sparse MoE shifts rather than eliminates bottlenecks.

What improves	What becomes harder
Active compute per token	Router quality and expert balance
Compute needed for a quality target	Communication across accelerators
Some inference costs	Avoiding idle accelerators and hot experts
Scaling model capacity	Keeping the full weight set accessible

This is why Moonshot recommends a supernode with 64 or more accelerators. Even when only a small subset of experts is active, the selected expert is not known in advance and the full model must remain accessible. At four bits, 2.8T weights imply a theoretical minimum of about 1.4 TB before scales, metadata, buffers, cache and replication. Sparse activation reduces arithmetic but does not remove model-residency memory or fabric requirements.

2.2 Kimi Linear addresses long-context cost

The Kimi Linear paper predates K3 and evaluates a 48B-total, 3B-active research model, not K3 itself. It combines Kimi Delta Attention with full Multi-head Latent Attention in a 3:1 ratio.

Full attention is strong at exact copying and fine-grained retrieval, but KV cache grows with context. Linear attention compresses history into a fixed-size state, reducing sequence-length dependence, but can lose exact detail. Kimi Linear uses three KDA layers followed by one full-attention layer to balance efficiency and expressiveness.

Component	Role	Trade-off
KDA	Compress long history into fixed-size state	Can weaken exact copying and fine retrieval
Full MLA	Restores precise token-to-token recall	KV cache still grows with context
3:1 hybrid	Balances efficiency and quality	Requires more complex kernels and serving software

The paper reports up to 75% lower KV cache and up to 6.3x theoretical decoding throughput at 1M context. A measured comparison in the complexity analysis reports 2.3x acceleration. These results establish direction, not guaranteed K3 API performance.

For investors, lower KV cache means more concurrent requests per accelerator. It also makes longer repositories, more documents and persistent agents economical. Memory per token can fall while total tokens rise.

2.3 Attention Residuals improve depth efficiency

Standard residual connections keep adding earlier layer outputs. In very deep networks, useful representations can be diluted. Attention Residuals allow a layer to select the earlier representations it needs. Block AttnRes groups layers and retains block-level representations, reducing memory overhead.

Moonshot’s research says roughly eight blocks recover most of the gains, and Block AttnRes can match a baseline trained with about 1.25x compute. This improves training capital efficiency, but it does not automatically reduce data-center demand. Better efficiency can be reinvested into larger models and longer tasks.

2.4 Quantization is a hardware-portability strategy

K3 applies quantization-aware training from supervised fine-tuning onward, using MXFP4 weights and MXFP8 activations. This should control accuracy loss better than aggressive post-training quantization and make deployment across multiple hardware platforms easier.

Moonshot’s kernel arena included NVIDIA H200 and a general-purpose GPU from an alternative vendor. The official post does not name a Chinese chip or claim H200-equivalent economics. What is verified is the strategic intent to optimize outside NVIDIA as well as on NVIDIA.

That matters for both China and global buyers. China needs frontier-level models that can survive constrained access to top NVIDIA systems. Other buyers want bargaining power across NVIDIA, AMD and custom accelerators.

3. What the Price Card Reveals

3.1 K3 is priced as a primary model

Model	Cached input per 1M	Standard input	Output	Current positioning
Kimi K3	$0.30	$3	$15	Frontier primary-model pricing
Claude Sonnet 5 launch promo	Varies	$2	$10	Temporary through August 31
Claude Sonnet 5 standard	Varies	$3	$15	Same headline price as K3
GPT-5.6 Sol	$0.50	$5	$30	67% higher input and 100% higher output than K3

K3 is more expensive than Sonnet 5’s current promotion. It is inaccurate to describe it as universally half-price. The clearer strategy is to match Sonnet’s standard tier while undercutting the most expensive frontier API.

The price is both a confidence signal and a monetization decision. Moonshot is no longer accepting a deep discount simply to acquire usage. It is claiming a position in the enterprise default-model tier.

3.2 Cost per completed task matters more than cost per token

Rate cards assume equal token use. Agents differ in planning length, tool calls, retries and verbosity. K3 currently exposes only max reasoning. Artificial Analysis reports that K3 used about 130 million output tokens in its Intelligence Index evaluation, more than twice the peer median of roughly 63 million. A cheaper output token can still lead to a costly completed task.

Task cost = input tokens × input rate + output tokens × output rate + tool and retry cost

Enterprise buyers will monitor task success, output tokens, time to first token, throughput, tool reliability, cache hit rate and session stability. Moonshot says Mooncake delivers more than 90% cache hits on coding workloads, making cached input one-tenth the uncached price. If that rate holds on enterprise traffic, K3’s effective economics improve materially.

3.3 Mooncake moves memory demand down the hierarchy

Mooncake separates prefill and decode clusters and disaggregates KV cache across CPU, DRAM and SSD rather than keeping everything in GPU HBM. Its paper reports up to 525% higher throughput in simulation and 75% more requests on production workloads.

This explains why AI memory demand extends beyond HBM.

Accelerator HBM → server DRAM → enterprise SSD → remote cache tier

KDA can reduce KV cache per request, yet persistent 1M-context agents increase total cache volume and retention time. Hot data remains in HBM and DRAM; colder context moves to SSD. Efficiency can soften an HBM-only thesis while strengthening the broader memory hierarchy.

4. Chinese Open Weights Move Up the Enterprise Stack

OpenRouter reports that Chinese models surpassed US models in token share on its platform in early June 2026, based on more than 450 trillion tokens from January through June 14. DeepSeek’s share rose from roughly 9% to 18% and it became the leading provider by mid-May.

The adoption path has three stages:

Low-cost open models take classification, translation and testing workloads.
Better coding and agent performance take repetitive enterprise workflows.
Near-frontier quality and million-token context compete for primary routing.

K3’s price is designed for the third stage. It is not following the most aggressive low-price strategy. It is charging a frontier-tier API price while promising weights that clouds, governments and enterprises can deploy themselves.

Proprietary APIs and open weights accumulate different strategic assets.

Proprietary frontier API	Near-frontier open weights
Controls the best UX and capability	Multiplies deployment routes and hardware options
Centralizes usage data	Lets enterprises retain data and operations
Changes pricing and policy centrally	Released files are difficult to withdraw
Model provider captures gross margin	Cloud, chip and application vendors share value

The strongest proprietary model can remain number one while losing paid token share. Enterprises can route repetitive work to K3-class models and reserve the most difficult legal, design and research tasks for the top closed model. Paid token mix and blended price matter more than leaderboard rank.

5. Semiconductor Demand: Efficiency and Diffusion Arrive Together

5.1 NVIDIA: near-term efficiency risk, medium-term deployment elasticity

Sparse MoE, KDA, quantization and autonomous kernel optimization reduce GPU time per task. Hardware portability also weakens the assumption that every frontier workload must stay on CUDA. Those are negative valuation signals for NVIDIA.

The positive side is equally material. Moonshot recommends at least 64 accelerators for self-hosted K3. Released weights could drive clouds, governments, laboratories and large enterprises to build their own clusters. Demand concentrated behind one API becomes hardware demand across many data centers.

Total accelerator demand = lower GPU time per task × higher total workload × more self-hosting institutions

The first term is negative; the last two are positive. K3 alone is not enough to cut NVIDIA earnings estimates, but neither is it enough to assume that every efficiency gain creates more GPU demand. Workload elasticity is the deciding variable.

5.2 AMD: optionality from hardware choice

AMD benefits if MXFP4, MXFP8 and vLLM portability allow enterprises to separate model choice from accelerator choice. A near-frontier open model gives buyers a realistic workload on which to test NVIDIA alternatives.

The proof must come after the weights. ROCm kernels, expert-parallel communication, 1M-context throughput and 64-accelerator stability need to be measured. AMD’s upside grows only if K3 demonstrates superior cost per successful task on MI systems.

5.3 HBM: lower per-request cache, larger model residency

Demand layer	K3 efficiency impact	Direction
Weight residency	A 2.8T model must remain quickly accessible	More accelerators and high-bandwidth memory
Per-request KV cache	KDA lowers cache size and growth	Less HBM capacity per request
Concurrent users and agents	Lower cost and open deployment can expand workloads	More total HBM and DRAM

The headlines can be negative for SK hynix, Micron and Samsung because 75% lower KV cache and 6.3x throughput sound like lower memory intensity. Those numbers belong to a Kimi Linear research model, not measured K3 production. HBM also stores weights, activations, communication buffers and batches, not only KV cache.

The medium-term balance is neutral to positive if self-hosted clusters and agent workloads grow faster than efficiency. It turns negative if workload elasticity is weak and compression improves faster than usage.

5.4 Server DRAM and enterprise SSDs are the clearest second-order beneficiaries

Long context and cache reuse cannot remain entirely in HBM. Once serving separates prefill and decode and spills cache across CPU, DRAM and SSD, server DRAM and enterprise SSD become operating assets for AI inference.

Samsung can sell HBM, server DRAM, high-capacity SSDs, foundry and packaging.
SK hynix combines HBM and server DRAM with Solidigm enterprise SSDs.
Micron supplies HBM, data-center DRAM and SSDs.

K3 does not show that AI needs less memory. It shows that AI memory is becoming hierarchical. The hottest data stays in HBM, retained context sits in server DRAM, and colder cache moves to enterprise SSD. Vendors with a full memory stack have greater resilience than a single-product HBM thesis.

5.5 Networking and custom silicon are the hidden MoE bottlenecks

Spreading 896 experts across accelerators creates variable communication depending on token routing. This is why Moonshot emphasizes balanced expert-parallel training, static shapes and removal of host synchronization from the critical path. Efficient operation across 64 or more accelerators requires high-bandwidth scale-up and scale-out fabric.

That is structurally positive for Broadcom, Marvell, NVIDIA networking, optical interconnect and switch silicon. It also encourages inference ASICs optimized for open models. K3’s 48-hour small-chip design exercise is not a commercial product, but it illustrates faster model-hardware co-design.

6. US Big Tech Strategy and Stock Transmission

6.1 Microsoft: pressure on OpenAI economics, more Azure usage

Microsoft owns exposure to OpenAI IP and economics as well as Azure infrastructure. The April 2026 partnership update keeps Microsoft as a primary cloud partner and extends a non-exclusive IP license through 2032.

K3 pressures the model layer because repetitive work can move away from OpenAI APIs. It can support the cloud layer if Azure hosts K3 as managed or self-hosted infrastructure.

Microsoft layer	K3 impact
OpenAI revenue share	Negative through price and mix
Copilot	Positive if routing costs fall
Azure AI	Positive if multi-model usage rises
Custom silicon and data centers	Positive if open-weight optimization expands

The near-term stock impact is neutral. The medium-term question is whether Azure converts model price deflation into usage and Copilot margin.

6.2 Amazon: Anthropic and AWS have different economics

Amazon is a major Anthropic investor and the provider of AWS, Bedrock, Trainium and Inferentia. K3 can reduce Claude pricing power and the value of Amazon’s Anthropic stake. If enterprises run K3 on AWS, however, EC2, Bedrock, storage, networking and Trainium usage can rise.

Amazon’s optimal strategy is to keep the workload on AWS regardless of which model wins. If K3 arrives with a permissive license and efficient Trainium support, AWS can monetize a competitor’s diffusion. The stock impact is neutral to modestly positive, although inference price competition could lower cloud margins even as revenue grows.

6.3 Alphabet: Gemini pricing pressure, TPU and Vertex defense

Google controls a model, a chip, a deployment platform and final demand through Search, Ads and Workspace. Vertex Model Garden supports first-party, third-party and open models.

K3 pressures Gemini API pricing but can create TPU and Google Cloud demand. Lower model cost also reduces the expense of AI Overviews, ad generation and Workspace agents. The stock impact is neutral to positive because Google is not solely a model vendor. The risk is that Gemini loses both performance and price leadership, raising cloud customer-acquisition cost.

6.4 Meta: from open-weight beneficiary to defender

Meta benefited from open-weight diffusion by commoditizing competitors’ APIs and lowering its own recommendation, advertising and content costs. If Chinese labs release near-frontier capability and longer context first, Meta risks losing its position as the benchmark US open-weight ecosystem.

K3 forces two responses: the next Llama must compete on long context, agent tools and deployment cost, and Meta must provide a trusted US alternative for enterprises and governments unwilling to deploy Chinese weights. The near-term earnings impact is limited because advertising drives cash flow. The strategic risk is lower returns on AI infrastructure and developer ecosystem investment if Llama falls behind.

6.5 Existing market positioning matters more than one launch

From June 22 through July 16, before most of the K3 evidence could affect trading, the US AI complex had already diverged sharply.

Company	Return	Drawdown from period high	Positioning signal
Meta	+17.9%	-3.1%	Strong ad cash flow and AI utilization expectations
Microsoft	+9.2%	-1.2%	Cloud and software resilience
Amazon	+7.3%	-3.2%	AWS and consumer recovery expectations
Alphabet	+1.4%	-5.5%	Mixed search and cloud positioning
NVIDIA	-0.6%	-3.1%	Relatively resilient
Broadcom	-4.5%	-9.7%	Custom AI optimism meets valuation pressure
AMD	-9.2%	-14.3%	Alternative accelerator optionality with high volatility
Micron	-29.6%	-32.0%	Correction after elevated memory expectations
Oracle	-29.1%	-32.7%	Tension between AI infrastructure growth and financing
Marvell	-38.8%	-40.1%	Severe derating in networking and custom silicon

These are not K3-caused returns. They show the positions into which K3 arrived. A relatively resilient NVIDIA may be more sensitive to efficiency headlines, while already-corrected Micron and Marvell could react more strongly if the released weights create verifiable infrastructure demand.

7. Company Impact Map

Company or layer	Near-term signal	Medium-term path	What to do now
NVIDIA	Valuation pressure from efficiency and portability	Self-hosted 64+ accelerator clusters offset efficiency	Watch workload elasticity, do not change estimates on the launch alone
AMD	Alternative deployment optionality	Share upside if ROCm economics are proven	Wait for post-July 27 throughput
Broadcom and Marvell	Networking complex already derating	MoE expert parallelism raises fabric demand	Verify orders and actual K3 cluster topology
SK hynix	Sensitive to KV-cache efficiency headlines	HBM, server DRAM and Solidigm SSD hierarchy exposure	Value the full memory stack, not HBM alone
Samsung Electronics	HBM efficiency risk plus catch-up position	HBM, server DRAM, SSD and foundry optionality	Broadest portfolio, execution still required
Micron	High expectations already corrected	Integrated US AI memory and storage exposure	Watch deployment volume versus pricing
Microsoft	OpenAI price and mix pressure	Azure and Copilot cost leverage	Cloud usage matters more than model margin
Amazon	Pressure on Anthropic asset value	AWS, Bedrock and Trainium benefit from model choice	The cleanest internal offset
Alphabet	Gemini price pressure	TPU, Vertex and Search cost benefits	Application-layer defense remains strong
Meta	Pressure on US open-weight leadership	Llama acceleration or broader open ecosystem	Strategic impact exceeds near-term earnings impact

There is not enough evidence for a new buy or sell call from this launch alone. Weights, license and external hardware throughput remain unavailable. The order of proof matters more than the direction of the narrative.

8. Three Scenarios

Base case: strong model, limited re-rating

Subjective probability: 55%. Full weights and a reasonably permissive license arrive by July 27. External results reproduce 85% to 95% of official performance. K3 runs on NVIDIA and AMD, but 64+ accelerators, max reasoning and high output-token use keep deployment expensive.

Proprietary APIs face more price pressure on repetitive work.
Clouds gain K3 hosting and self-deployment demand.
Per-token efficiency offsets higher workload.
HBM is neutral to modestly positive.
Server DRAM, enterprise SSD and networking are positive.

Bull case: open weights enter primary enterprise routing

Subjective probability: 25%. The license is close to MIT, vLLM and SGLang support is stable, and NVIDIA, AMD and Chinese accelerators show strong throughput. External evaluation confirms performance immediately below the top proprietary models. Lower reasoning modes reduce task cost. Major clouds and enterprise gateways add K3 to default routing.

Proprietary model blended price and token share fall.
Hyperscaler multi-model usage rises.
Self-serving clusters increase accelerator demand.
HBM, networking, server DRAM and enterprise SSD benefit from deployment volume.
Meta and US open-model programs accelerate investment.

Bear case: weights reveal a cost and quality gap

Subjective probability: 20%. Weights are delayed or restricted, external scores fall below the official table, preserved-thinking-history requirements and excessive proactiveness create enterprise failures, and task cost exceeds Sonnet 5 because of max reasoning and verbosity. The 64-accelerator requirement limits self-hosting.

Moonshot may have to retreat from frontier pricing.
Proprietary APIs retain a UX and reliability premium.
Incremental semiconductor demand remains limited.
Existing earnings and capex regain control of stock prices.

9. The July 27 Verification Checklist

Proof point	Strong signal	Weak signal
Weights and license	Full release on time with commercial modification rights	Delay, restrictions or missing components
External evaluation	Official scores broadly reproduced under one harness	Large drop versus company table
Cost per task	Lower reasoning modes and fewer output tokens	Max-only reasoning and persistent verbosity
NVIDIA throughput	Stable expert parallelism on 64-accelerator nodes	Fabric bottlenecks and low utilization
AMD throughput	Cost advantage under ROCm and vLLM	Immature kernels or accuracy loss
Chinese accelerators	Named hardware and measured throughput	Only “alternative GPU” language
Caching	Near-90% hits on enterprise traffic	Sharp decline outside coding
Cloud adoption	AWS, Azure, Google Cloud or Oracle listings	Limited to a few Chinese platforms
Competitive pricing	Standard price cuts or wider cache discounts	Temporary promotions only
Memory orders	Upward revisions to HBM, server DRAM and SSD volumes	Efficiency rises while volumes stagnate

10. The Strongest Counterargument

The strongest bear case is straightforward: semiconductor demand falls if efficiency improves faster than usage. Model compression, linear attention, quantization, cache reuse and inference ASICs could process the same number of useful tasks with far fewer GPUs and less HBM. Open-weight competition can reduce API prices without producing enough incremental paid work to justify AI capex.

Evidence for that outcome would include slower inference revenue despite falling prices, higher accelerator utilization but fewer new clusters, HBM bit growth below model-efficiency gains, weak conversion of cloud AI backlog into revenue, and enterprises using open models only to cut costs rather than create new workflows.

The bull case requires price declines to unlock new work: repository-scale coding, research automation, video editing, chip design and workflows that were previously uneconomic. Workload elasticity relative to model efficiency is the common proof point for both accelerator and HBM investing.

11. Conclusion

Kimi K3 does not prove that China has surpassed the strongest US proprietary models. Moonshot acknowledges the remaining UX gap. As of July 17, the full weights and technical report are not available, and a mixed-harness benchmark table cannot declare a winner.

It does change market structure. A 2.8T, 1M-context, near-frontier model is live at Sonnet’s standard price, with weights promised within ten days. Chinese open weights are moving from cheap second-tier substitutes toward primary enterprise workloads.

For semiconductors, the event is more demand reallocation than demand destruction. KDA and quantization reduce HBM and GPU time per request. Sparse MoE and 64-accelerator supernodes increase model-residency memory and fabric. Mooncake pushes cache into server DRAM and enterprise SSDs. The HBM-only story becomes less simple, while the full memory and data-center hierarchy remains exposed to wider deployment.

US Big Tech faces the same split. Model APIs absorb price pressure; clouds and applications absorb lower model cost. Microsoft, Amazon and Alphabet can host K3 even if their preferred models lose share. Meta must defend its strategic role as the trusted US open-weight standard.

On July 27, the important evidence is not the existence of a weight file. It is a permissive license, reproducible evaluation, multi-hardware throughput and competitive cost per completed task. If those conditions hold, K3 becomes an event that changes both the AI price curve and the semiconductor demand path. If they do not, it remains an impressive product launch rather than an industry reset.

Sources and Limitations

Primary materials: Kimi K3 launch, Kimi K3 API documentation, Kimi Linear paper, Kimi Linear GitHub, Attention Residuals paper, Mooncake paper, Claude Sonnet 5 pricing, GPT-5.6 Sol pricing, OpenRouter model adoption analysis, Microsoft-OpenAI partnership, Google Vertex Model Garden, and Amazon-Anthropic partnership.

The semiconductor and stock transmission analysis is an inference from public architecture, serving and market data. Exact active parameters, weight size, license terms, AMD and Chinese-accelerator throughput, and enterprise cost per completed task remain blocked as of July 17. This article is for research and information purposes only and is not investment advice.

Kimi K3 on Korea Invest Insights

Weekly Semiconductor Deep Dive: Strong Earnings, Compressed Multiples, and the 2028 Inflection

TL;DR

1. Scope and Evidence Rules

2. A Map of the Week’s Events

3. Evidence Strengthened This Week

3.1 TSMC and ASML Showed That AI Orders Have Translated into Real Activity

3.2 Memory Prices Remain Elevated

3.3 HBM Ramp Cannibalizes Commodity DRAM Supply

3.4 2027 Shortage Is a Strong Conclusion; 2.5× Shortage by 2030 Is a Scenario

4. Risks Strengthened This Week

4.1 Earnings Growth Rate Has Become More Important Than Strong Absolute Earnings

4.2 Long-Term Contracts Create Both an Earnings Floor and a Price Ceiling

4.3 Equipment Orders Are Both Current Demand and Future Supply

4.4 High Memory Prices Can Erode Customer Demand

4.5 AI Capex Bottleneck Is Shifting to Cost of Capital

5. Was the Share Price Plunge a Supply-Demand Shock or a Fundamental Repricing?

6. China Is Three Variables at Once

6.1 China Is a Major Demand Market

6.2 China Is a Commodity Memory Competitor

6.3 China Can Be Segmented by Policy

7. Kimi K3 and Efficiency: Demand Shifting, Not Demand Destruction

8. Five Clocks — When They Are Watched Together, the Contradictions Resolve

9. Scenario Update

10. Roles by Stock and Value Chain

Samsung Electronics: Risk-Adjusted Core

SK Hynix: Pure Beta on Sustained Scarcity

Micron: Intersection of U.S. Policy and Pure-Play Memory

SanDisk: Tactical Beta on NAND and eSSD

TSMC and ASML: Demand Validators and Leading Indicators of Future Supply

Korean Equipment and Materials: Actual Orders and Recurring Revenue Over Published Plans

11. Counter-Arguments: Where This Thesis Could Be Wrong

Demand Evidence May Have Been Double-Counted

The 2030 Demand Model May Underestimate Efficiency Gains

CAPEX May Grow Faster Than Returns

China Policy May Not Be Implemented

Open-Model Company Disclosures May Not Be Independently Reproducible

The Entire Price Decline May Have Been Attributed to Forced Liquidation

12. Observable Conditions That Would Change the Thesis

Conditions That Raise the Probability of a Bullish Outcome

Conditions That Raise the Probability of P3 Normalization

P4 Warning Conditions

13. Data to Track from Late July Onward

14. Final Judgment

Key Sources

Limitations of Public Data

What China's Open-Model Convergence Actually Changes: Value-Chain Redistribution, Not Demand Collapse

TL;DR

1. Getting the Facts Straight First

The Efficiency Gains Are Real

But So Is the Other Side

2. The Core Equation: What to Actually Watch

The Verdict So Far

3. Semiconductor Impact: It Diverges by Stock

Why Samsung Has the Relative Edge Over Hynix

NVIDIA and H200

4. Big Tech Impact

The Lesson from the 2025 DeepSeek Shock

5. Verdicts on the Chinese Open-Model Thesis

6. How Far Has CXMT’s HBM Actually Gotten

How the Legacy-DRAM Hypothesis Needs to Be Revised

7. How US Policy Reshapes This Reading

Yet This Is Not a Full Decoupling

8. Can Western Enterprises Actually Use Chinese Model APIs?

The Limits of Direct APIs

9. Revising the Thesis: What Changes and What Stays

10. The Order of Casualties and Beneficiaries

11. Scenario Update

12. What Would Change This Judgment

Closing

Related Posts

Kimi K3 Resets the AI Price Curve: From Kimi Linear to HBM and Big Tech Strategy

Executive Summary

1. Correcting the Numbers Before Drawing Conclusions

2. How a 2.8T Model Becomes Deployable

2.1 Total parameters are not active parameters

2.2 Kimi Linear addresses long-context cost

2.3 Attention Residuals improve depth efficiency

2.4 Quantization is a hardware-portability strategy

3. What the Price Card Reveals