Hyperscalers on Korea Invest Insights

AI Token Value Today and Tomorrow: Value Added for Memory Companies

Thu, 09 Jul 2026 09:00:00 +0900

Related context: this article follows our work on late July big tech earnings and the memory thesis, Micron’s FY3Q26 AI memory earnings, and the 1H26 AI infrastructure bottleneck review. Related hubs are the Exclusive Analysis Hub and the AI HBM Hub. The question is not simply whether AI demand is strong. The question is who keeps the economic value created by AI tokens.

TL;DR

If we decompose one current dollar of AI token revenue, the largest share of value added still sits at the model and application layer. The equity market, however, is being driven by data center investment that is being pulled forward before final application revenue has fully arrived. That is why today’s AI value chain looks less like a normal pyramid and more like an inverted one.

The current value-added split of one dollar of AI token revenue is roughly as follows.

Layer	Current value-added estimate
AI model company	45 to 55 cents
Cloud and data center	10 to 16 cents
GPU supplier	About 13 cents
HBM, server DRAM, SSD	About 2.3 cents
Other infrastructure software and storage	3 to 5 cents
Foundry	About 1 cent
Power	About 1.3 cents

The market capitalization and capex cycle cannot be explained by this table alone. Current AI service revenue is estimated around 60 to 100 billion dollars per year, while the leading GPU supplier’s data center revenue run rate has reached around 300 billion dollars, and the big four cloud companies’ 2026 capex is estimated near 725 billion dollars. The industry is building future demand before current revenue catches up.

Memory is a real bottleneck in this structure. But if memory captures only about 2 to 5 cents of value added per token dollar, the central question for memory stocks is not “is the P/E low?” The real question is whether the industry can sustain mid-cycle net income of roughly 100 to 140 trillion won.

Question	View
Where does AI token value accrue?	Near term: models and cloud. Long term: workflow applications and physical bottlenecks
Is memory a structural beneficiary?	Yes, but its value-added share is small and the stocks already price in a strong mid-cycle profit base
What should investors monitor?	HBM4 contract pricing, customer allocation, blended memory ASP, cloud backlog, and application revenue conversion

1. The Real Question: Who Keeps the Dollar?

The most common mistake in AI investing is mixing up revenue attribution with value-added attribution. If a customer pays one dollar for an AI service, that dollar may first appear as revenue at a model company. But inside that dollar are cloud costs, GPU depreciation, memory, foundry capacity, power, networking, servers, cooling, labor, software, taxes, and financing costs.

So “who books the revenue?” is not the same question as “who keeps the surplus?”

This article uses five rules.

Rule	Meaning
Revenue attribution	The first seller that invoices the customer
Value-added attribution	The economic value left after removing double counting
Token dollar	A dollar of API or model usage revenue
Application dollar	A dollar of final workflow or agent software spend
Depreciation basis	Data center capex is annualized over a five-year asset life

The capex treatment matters. AI data center spending does not disappear as an immediate cost. It is capitalized into GPUs, servers, power gear, buildings, cooling, and networks, then depreciated over several years. Therefore, the right cost in a token-dollar model is annual depreciation, not total capex.

Time also matters. Inference cost in 2025 is not the same as inference cost in 2028. Cost per unit of performance is falling quickly.

Period	Inference cost as share of revenue	Model gross margin
2025	50 to 67%	33 to 50%
2026	40 to 55%	45 to 60%
2027 to 2028	30 to 40%	60 to 70%

This leads to an important point. Token prices can fall while model gross margins rise, if inference costs fall faster than prices. Much of that margin is then reinvested into training, compute reservations, and customer acquisition. The value does not simply stay on the income statement. It flows back into GPUs, memory, power, and foundry capacity.

2. The Current Value Waterfall

The table below decomposes one current AI token dollar. Revenue attribution is the amount a layer effectively invoices or absorbs. Value added is the economic share after removing double counting.

Layer	Revenue attribution	Value-added estimate	Interpretation
AI model company	100 cents	45 to 55 cents	Direct customer layer. Falling inference cost is the largest lever
Cloud and data center	About 50 cents	10 to 16 cents	Operating layer that bundles GPUs, networks, power, and buildings
GPU supplier	About 18 cents	About 13 cents	Main physical bottleneck today
HBM, server DRAM, SSD	About 3 cents	About 2.3 cents	Critical to system performance, but small as a token-dollar share
Infrastructure software, storage, security, observability	About 5 cents	About 3 cents	More important as operations become more complex
Foundry	About 1.5 cents	About 1 cent	Common bottleneck across AI chip designs
Server OEM and ODM	About 3 cents	About 1.4 cents	High throughput, limited margins
Third-party networking	About 1.3 cents	About 0.7 cents	Important at cluster scale, but supplier power varies
Power	About 2.5 cents	About 1.3 cents	Small cost line, large site bottleneck
CPU and IP	About 0.5 cents	About 0.3 cents	Control and support compute
Construction, components, materials, residual	About 9 cents	Distributed	Lower layers of the data center stack

Two observations matter.

First, power is not the largest cost item inside a token dollar. But it is one of the largest physical bottlenecks. Power matters through grid access, substations, transmission, permits, and long-term power contracts.

Second, memory’s value-added share per token dollar is still small. This does not mean HBM is unimportant. HBM is indispensable. But the economic share retained by memory companies is far smaller than the share retained by GPUs or model/application layers. If the equity market ignores that difference, valuation errors appear.

3. The Inverted Pyramid

The current AI stack looks like this.

Item	Estimated scale
AI model and application revenue	60 to 100 billion dollars per year
GPU data center revenue run rate	About 300 billion dollars
2026 big four cloud capex	About 725 billion dollars
AI revenue needed to normalize 700 billion dollars of capex	About 2.8 to 3.0 trillion dollars

If AI service revenue is currently 60 to 100 billion dollars, and memory revenue is 3 to 7 cents per token dollar, AI memory flow revenue should be only about 2 to 7 billion dollars per year. Yet the leading memory companies are producing revenue and profit numbers far above what current token revenue alone can explain.

For example, SK Hynix reported 97.1 trillion won of revenue in 2025 and 52.6 trillion won of revenue in the first quarter of 2026. That scale is hard to explain only from current AI token revenue. Much of today’s memory revenue is therefore tied to capex being pulled forward for future AI demand.

This is not automatically wrong. Data centers cannot be built instantly after demand appears. Power, land, substations, GPUs, servers, cooling, and networking must be secured years in advance. Pre-investment is necessary.

The question is speed. If application and cloud revenue catch up, today’s capex is normal growth investment. If final demand lags, the correction begins upstream: memory and server components first, then GPUs, then cloud capex plans.

4. Three Convergence Paths

AI revenue is not determined by token price alone.

AI revenue = token volume × token price

Token price per unit of performance is falling quickly. It may decline by roughly 10x per year through the early phase, then slow to 1.5 to 2x per year after 2027. But volume can explode at the same time. A simple question may use a few hundred tokens. A workflow, document process, coding task, or agent job can use tens of thousands, hundreds of thousands, or even more than a million tokens.

That means price declines do not automatically mean revenue declines. If price falls by 90% but volume rises 100x, revenue still rises 10x.

Scenario	Probability sense	Description	Market implication
Demand convergence	45%	Application revenue scales quickly and 1.0 to 1.6 trillion dollars of capex normalizes	Limited upstream drawdown, bottleneck premium remains
Digestion then reacceleration	40%	2027 to 2028 capex pauses by 20 to 30%, then workflow AI demand returns	Memory ASP could correct 40 to 60%, then recover
Structural oversupply	15%	Usage elasticity disappoints and data center buildout proves excessive	Capex cuts of 50% or more and upstream recession

On-device AI changes the distribution, not the whole demand picture. More inference on phones, PCs, cars, robots, and industrial devices can reduce some cloud token revenue. But silicon demand does not disappear. LPDDR, mobile SoCs, edge NPUs, embedded storage, and local servers may all benefit.

5. Long-Run Equilibrium: Workflows Capture the Dollar

Over the long run, tokens are unlikely to remain scarce. As token prices fall and model performance gaps narrow, value moves from token production toward workflow ownership.

On a pure API dollar, the long-run value split may look like this.

Layer	Long-run value-added estimate
Model and API company	45 to 55 cents
Cloud and data center	12 to 15 cents
GPU supplier	8 to 10 cents
Custom ASIC	4 to 5 cents
Memory	4 to 5 cents
Foundry	3 to 4 cents
Power and data center infrastructure	4 to 5 cents
CPU and IP	About 1.5 cents
Server OEM	About 1.5 cents
Other infrastructure software	4 to 6 cents
Residual	About 5 cents

But customers do not really buy tokens. They buy outcomes: customer support, coding productivity, claims processing, contract review, factory quality control, and logistics automation. On a final application dollar, the distribution changes.

Layer	Long-run revenue attribution	Long-run value added
Workflow applications, agents, distribution	100 cents	45 to 50 cents
Model and API	35 to 40 cents	18 to 24 cents
Cloud and data center	13 to 15 cents	4 to 6 cents
Total silicon	9 to 12 cents	6 to 8 cents
Power, land, cooling	1.5 to 2 cents	About 1 cent
Server OEM and generic infrastructure	About 1 cent	Low
Other software	3 to 5 cents	Medium

The long-run conclusion is simple: tokens commoditize, while workflow ownership and physical bottlenecks retain value.

6. Capture Tiers

The best AI businesses are not simply the ones with the most revenue. The best businesses have three traits.

Customers cannot switch easily.
Cost declines become margin expansion.
Supply cannot expand quickly when demand rises.

Tier	Layer	Why it matters
Tier 1	Foundry, workflow applications, integrated cloud platforms	Bottleneck supply, customer lock-in, full-stack control
Tier 2	GPU, platform AI suppliers, HBM leaders, custom ASIC	Current bottlenecks and high growth, but with price and competition risk
Tier 3	Server OEM, neocloud, commodity memory, regulated utilities	Volume grows, but rent capture is weaker

Foundry belongs in Tier 1 because AI wafer demand exists whether the winning chip is a GPU or a custom ASIC. Workflow applications belong there because model costs can fall while end-user value remains tied to business outcomes.

Integrated cloud platforms need case-by-case judgment. Pure GPU rental can become a capital-heavy utility. But a company that owns search, ads, productivity software, cloud, chips, models, and data centers can turn capex into margin.

7. Multiples: Cheap Looking Is Not the Same as Cheap

AI value-chain multiples should differ by layer.

Layer	Appropriate valuation lens
Foundry	EV/S 8 to 11x, P/E 20 to 26x if growth persists
GPU supplier	Normalized EV/S 9 to 13x, scarcity 15 to 18x, P/E 18 to 24x
Integrated cloud	EV/S 5 to 8x, full-stack 8 to 10x, P/E 20 to 28x
Platform AI supplier	Hypergrowth EV/ARR 10 to 18x, deceleration 5 to 8x
Workflow AI software	Simple tools 2 to 4x, workflow owners 8 to 12x, regulated data owners 12 to 20x
HBM and memory	EV/S 3 to 5x, leaders 5 to 7x, mid-cycle P/E 8 to 14x
Storage	EV/S 2 to 3.5x, platform 4 to 7x, normalized P/E 8 to 14x
CPU and IP	CPU 12 to 18x, IP conditionally 30 to 50x
Server OEM	EV/S 0.5 to 1.5x, P/E 8 to 15x
Power and data center infrastructure	Contracted assets at EV/EBITDA 10 to 18x or project IRR

Memory is the layer where investors should be most careful. A 5 to 7x P/E can signal cheapness. At a cycle peak, it can also simply mean that the earnings denominator is temporarily inflated.

Useful valuation formulas include:

Formula	Use
EV/NOPAT = (1 - g/ROIC) / (WACC - g)	Normalized multiple with growth and capital efficiency
Justified PBR = (sustainable ROE - g) / (COE - g)	Long-run capital-intensive industry PBR
Mid-cycle EPS = (peak + trough + 2×normal) / 4	Cycle-aware earnings
Pre-profit value = terminal NOPAT × terminal multiple / (1 + discount rate)^t	Early-stage or loss-making firms

Capital costs also differ.

Layer	WACC or discount rate
Hyperscalers	8.5 to 9.5%
GPU and foundry	9 to 10.5%
Memory	10.5 to 12%
Vertical AI software	12 to 18%
Loss-making model layer	14 to 20%

8. The Memory Question: Can Mid-Cycle Net Income Reach 100 to 140 Trillion Won?

For memory companies, the key question is no longer whether HBM is good. The market already knows that. The key question is whether the current market cap requires a realistic or unrealistic mid-cycle profit base.

Suppose a leading memory company is valued near 1,400 trillion won. To sustain that value after one or two peak years, mid-cycle net income must be very large.

Item	Assumption
Peak net income	240 trillion won
Trough net income	50 trillion won
Normalized net income	60 trillion won
Mid-cycle net income	(240 + 50 + 2×60) / 4 = about 102 trillion won

Applying an 8 to 14x mid-cycle multiple gives a fair-value range of about 820 to 1,430 trillion won. If the current market cap is already near the top of that range, the stock is not automatically expensive, but the margin of safety is thin.

Further upside requires one or more of the following.

Requirement	Meaning
Evidence that mid-cycle net income can exceed 140 trillion won	HBM premium and high-margin mix remain after the peak
HBM4 custom premium is defended	Product customization protects pricing
ASP holds after Samsung’s supply entry	More supply does not crush pricing
General DRAM and NAND pricing also improves	The cycle broadens beyond HBM
Long-term customer agreements and volume commitments	Downcycle earnings risk falls

9. Samsung, SK Hynix, and Micron Are Not the Same Trade

Company	Core logic	Risk
Samsung Electronics	HBM share recovery and memory mix improvement. Expectations are lower than for the leader	HBM qualification, yield, foundry and system LSI losses
SK Hynix	Highest HBM profit quality and customer reference	Much of the good news is already in the price
Micron	Leading indicator for AI memory contracts, pricing, FCF, and capex	Volatility, short interest, and supply-cycle sensitivity

Samsung has more catch-up optionality. SK Hynix has the highest-quality memory profit pool, but expectations are embedded. Micron is the canary because its pricing, long-term agreements, data center revenue, and capex guide influence Korean memory estimates directly.

10. Cloud, Power, and Vertical AI

Cloud companies split into two groups: integrated platforms and capital-heavy rental platforms. The most important numbers are backlog, revenue conversion, operating margin, internal silicon mix, free cash flow, and bond issuance.

Power is a small token-dollar cost line but a large site bottleneck. Grid access, substations, transmission, PPAs, gas turbines, nuclear, SMRs, water, land, cooling, and permits matter more than the simple power bill.

Vertical AI companies can capture the largest long-run share if they own the workflow. The winners will have regulated or physical domains, proprietary data, action authority, measurable ROI, model replaceability, a path to 70%+ gross margin, and compliance moats. Generic chat wrappers, basic retrieval integration, API resale, data-less automation, GPU brokerage, and server assembly should be treated with caution.

11. What to Monitor

Metric	Why it matters
Combined cloud backlog	Whether capex is tied to future revenue
Self-built capex ratio	How much demand is moving outside public cloud
On-device inference share	Cloud token revenue substitution risk
HBM4 contract pricing	Durability of memory premium
Customer allocation by HBM supplier	Supplier bargaining power
Blended memory ASP	Whether the cycle is broadening
GPU lead times	Persistence of scarcity
Cloud operating margin	Whether capex turns into profit
AI application revenue growth	Whether final demand catches infrastructure spending

The thesis weakens if open-weight systems compress model pricing faster than expected, on-device AI reduces cloud demand more quickly, 2027 cloud capex slows sharply, HBM4 premiums compress, or AI application revenue fails to scale.

Final View

The AI token economy can be summarized in one sentence:

Tokens get cheaper, while value moves to workflow ownership and hard-to-expand physical bottlenecks.

Memory is one of those bottlenecks. HBM, high-performance DRAM, and SSDs are essential to AI servers. But memory’s value-added share per token dollar remains limited. That is why memory stocks require more than the statement that HBM is strong. Current prices already demand a strong mid-cycle profit base.

The four key questions are:

Does HBM4 preserve custom pricing?
Do leading suppliers hold ASP and margins after Samsung’s entry?
Do cloud backlog and AI application revenue catch up to capex?
Can memory companies sustain mid-cycle net income of 100 to 140 trillion won?

If those four conditions are met, today’s large memory market caps can be justified. If not, low P/E ratios may prove to be a cycle-peak illusion rather than a margin of safety.

Which NVIDIA Quarter Does Samsung's Earnings Selloff Resemble?

Wed, 08 Jul 2026 14:30:00 +0900

This is a follow-up to Samsung 2Q26 preview, Micron FY3Q26 earnings, NVIDIA elasticity and the Korea HBM cycle, late-July Big Tech calls and the memory thesis, and hyperscaler financing and memory bottlenecks. Related hubs: Exclusive Analysis and AI HBM.

TL;DR

Samsung Electronics’ 2Q26 preliminary earnings were strong on the surface. Revenue was about KRW 171 trillion and operating profit was about KRW 89.4 trillion. Sales rose 27.74% quarter over quarter and operating profit rose 56.21%. Operating profit was up 1,810.26% year over year. These figures were disclosed by Samsung Electronics.

Yet the stock fell 6.92% on July 7 and another 6.25% on July 8. The two-day decline was 12.7%.

That is not the same as the famous NVIDIA August 2024 sell-on reaction. NVIDIA fell 6.4% on the first reaction day in August 2024, but rebounded 1.5% the next day. Samsung did not bounce. It fell again.

Across 17 NVIDIA post-earnings reactions, the closest analogue is NVIDIA’s Q4 FY26 report on February 25, 2026. NVIDIA fell 5.5% on the first reaction day and 4.2% the next day, for a two-day decline of 9.4%. The reaction-vector distance from Samsung’s path is 2.55, the smallest in the sample.

The key point is not that a strong earnings print is useless. The point is that a strong print alone was not enough once the market started asking whether AI infrastructure growth was slowing. NVIDIA’s large post-print drawdowns eventually recovered, but the recovery came only after a new catalyst directly killed the fear that had caused the drawdown. For Samsung, the equivalent catalyst is late-July Big Tech CapEx guidance.

The practical conclusion is Wait. This is not a mechanical dip-buy setup yet. The next gates are late-July hyperscaler CapEx commentary, DRAM/NAND price momentum, and Samsung’s clean 3Q operating-profit path.

1. What was strange about Samsung’s earnings reaction

The numbers were strong.

Samsung Electronics announced 2Q26 preliminary earnings on July 7, 2026. The company disclosed the following consolidated figures.

Metric	2Q26 preliminary	1Q26	2Q25
Revenue	about KRW 171T	KRW 133.87T	KRW 74.57T
Operating profit	about KRW 89.4T	KRW 57.23T	KRW 4.68T
Revenue growth		+27.74% QoQ	+129.31% YoY
Operating profit growth		+56.21% QoQ	+1,810.26% YoY

Samsung also disclosed estimate ranges of KRW 170T to KRW 172T for sales and KRW 89.3T to KRW 89.5T for operating profit. Korean disclosure rules required the company to publish a single median figure.

The stock reaction was the problem. Samsung closed at KRW 318,000 on July 6, KRW 296,000 on July 7, and KRW 277,500 on July 8. Based on yfinance daily close data:

Date	Close	Daily return	Read-through
2026-07-06	KRW 318,000	+2.75%	pre-announcement base
2026-07-07	KRW 296,000	-6.92%	announcement-day selloff
2026-07-08	KRW 277,500	-6.25%	no bounce
Two-day cumulative		-12.74%	post-print path

Compared with the June 18 high of KRW 362,500, the July 8 close was already 23.45% lower. Samsung was not falling from a calm setup. It was already volatile before the earnings print, and the good number failed to stop the selling.

The key is that the market was no longer asking only how much Samsung earned. It was asking whether the pace of earnings improvement could keep accelerating.

2. Why compare Samsung with NVIDIA

Samsung is a memory company. NVIDIA is a GPU platform company. Their businesses are different. But both have been core price signals for the AI infrastructure cycle.

NVIDIA moved first when the AI infrastructure boom began. As bottlenecks moved downstream into HBM, DRAM, eSSD, substrates, and power infrastructure, Samsung and SK hynix began facing the same market question: is the next leg still accelerating?

Question	Meaning
Did earnings improve?	confirmed result
Did earnings beat the bar?	surprise versus expectations
Will the next quarter be better?	guidance and pricing
Is the growth rate still rising?	the part stocks care about most
Will hyperscalers keep spending?	the final payer for AI memory demand

Samsung’s two-day selloff came from that last layer. The print was strong, but expectations were already high. Some investors had expected operating profit above KRW 90T. At the same time, the market was worried about possible hyperscaler CapEx moderation. Korean single-stock leverage products, futures, and ETF flows amplified the move.

This analysis therefore does not compare Samsung’s business model with NVIDIA’s business model. It compares post-earnings price paths and the fear behind those paths.

3. Timing adjustment: NVIDIA reports after the close

NVIDIA usually reports after the US market close. The earnings information is not reflected in the announcement-day close. The real reaction appears in the next regular trading session.

Samsung’s guidance was reflected in the Korean session on the announcement day. So the timing has to be aligned as follows.

Company	Reporting style	First trading session with full reaction
Samsung Electronics	reflected in the Korean session	announcement day
NVIDIA	after-market release	next trading day

In this note, NVIDIA’s reaction D0 means the first trading day after the after-market release. NVIDIA’s D+1 means the following trading day. This aligns NVIDIA’s D0 with Samsung’s July 7 and NVIDIA’s D+1 with Samsung’s July 8.

Without this adjustment, the comparison would be wrong.

4. NVIDIA’s 17-quarter post-earnings reaction table

The table below combines NVIDIA quarterly revenue, growth rates, and post-earnings stock reactions. Revenue and report timing are sourced from NVIDIA IR and SEC 8-K releases. Stock returns are calculated from yfinance daily close data.

Quarter	Report date	Revenue	QoQ	YoY	Reaction D0	D+1	Two-day	T+20
Q1 FY23	2022-05-25	$8.3B	n/a	n/a	+5.2%	+5.4%	+10.8%	+0.9%
Q2 FY23	2022-08-24	$6.7B	-19.3%	n/a	+4.0%	-9.2%	-5.6%	-27.1%
Q3 FY23	2022-11-16	$5.9B	-11.9%	n/a	-1.5%	-1.7%	-3.1%	+6.5%
Q4 FY23	2023-02-22	$6.1B	+3.4%	n/a	+14.0%	-1.6%	+12.2%	+27.5%
Q1 FY24	2023-05-24	$7.2B	+18.0%	-13.3%	+24.4%	+2.5%	+27.5%	+38.2%
Q2 FY24	2023-08-23	$13.5B	+87.5%	+101.5%	+0.1%	-2.4%	-2.3%	-12.9%
Q3 FY24	2023-11-21	$18.1B	+34.1%	+206.8%	-2.5%	-1.9%	-4.3%	-3.7%
Q4 FY24	2024-02-21	$22.1B	+22.1%	+262.3%	+16.4%	+0.4%	+16.8%	+33.9%
Q1 FY25	2024-05-22	$26.0B	+17.6%	+261.1%	+9.3%	+2.6%	+12.1%	+33.3%
Q2 FY25	2024-08-28	$30.0B	+15.4%	+122.2%	-6.4%	+1.5%	-5.0%	-1.2%
Q3 FY25	2024-11-20	$35.1B	+17.0%	+93.9%	+0.5%	-3.2%	-2.7%	-10.4%
Q4 FY25	2025-02-26	$39.3B	+12.0%	+77.8%	-8.5%	+4.0%	-4.8%	-13.3%
Q1 FY26	2025-05-28	$44.1B	+12.2%	+69.6%	+3.2%	-2.9%	+0.2%	+15.0%
Q2 FY26	2025-08-27	$46.7B	+5.9%	+55.7%	-0.8%	-3.3%	-4.1%	-2.2%
Q3 FY26	2025-11-19	$57.0B	+22.1%	+62.4%	-3.2%	-1.0%	-4.1%	-6.6%
Q4 FY26	2026-02-25	$68.1B	+19.5%	+73.3%	-5.5%	-4.2%	-9.4%	-8.6%
Q1 FY27	2026-05-20	$81.6B	+19.8%	+85.0%	-1.8%	-1.9%	-3.6%	-5.7%

The key message is clear. Strong results did not always lift the stock. As the AI cycle matured, the market became more sensitive to the change in growth rate than to the level of revenue.

5. The closest analogue is NVIDIA Q4 FY26

Samsung’s path was:

Company/event	D0	D+1	Two-day
Samsung 2Q26 preliminary earnings	-6.92%	-6.25%	-12.74%

We compare this two-day vector with all 17 NVIDIA quarters.

Rank	NVIDIA quarter	Report date	D0	D+1	Two-day	Distance from Samsung
1	Q4 FY26	2026-02-25	-5.46%	-4.16%	-9.39%	2.55
2	Q3 FY24	2023-11-21	-2.46%	-1.93%	-4.34%	6.21
3	Q3 FY26	2025-11-19	-3.15%	-0.97%	-4.10%	6.48
4	Q1 FY27	2026-05-20	-1.77%	-1.90%	-3.64%	6.74
5	Q2 FY26	2025-08-27	-0.79%	-3.32%	-4.09%	6.79

Q4 FY26 is the clear match. It is not just the numbers. The drivers were similar.

NVIDIA’s Q4 FY26 revenue was $68.1B, up 73% year over year and 20% quarter over quarter. The next-quarter revenue guide was $78B. The print was strong. But the stock fell 5.5% on the first reaction day and 4.2% on the next day.

The market was asking whether the next step could still get better. Customer concentration, hyperscaler spending durability, AI CapEx peak risk, and competition were all in focus.

Samsung faced the same structure. KRW 89.4T of operating profit was strong, but investors had already moved the bar higher. The market wanted proof that the growth rate was still accelerating, not only proof that the current profit level was high.

6. Why the August 2024 NVIDIA sell-on is the wrong template

The intuitive comparison is NVIDIA Q2 FY25 in August 2024. NVIDIA fell 6.4% on the first reaction day. That resembles Samsung’s July 7 decline.

But the next day changes the conclusion.

Case	D0	D+1	Read-through
NVIDIA Q2 FY25, August 2024	-6.4%	+1.5%	one-day selloff, then bounce
NVIDIA Q4 FY25, February 2025	-8.5%	+4.0%	one-day selloff, then strong bounce
NVIDIA Q4 FY26, February 2026	-5.5%	-4.2%	two-day selloff
Samsung 2Q26	-6.92%	-6.25%	two-day selloff

The August 2024 and February 2025 NVIDIA reactions were dip-and-bounce. Samsung was dip-and-continue. That distinction matters. A one-day sell-on can be a dip-buy setup. A two-day selloff can mean the market has not finished digesting the fear.

7. Did NVIDIA recover immediately after large post-print selloffs?

No.

There were three NVIDIA quarters with a first-reaction-day decline worse than 5%: Q2 FY25, Q4 FY25, and Q4 FY26. All three were still below the pre-announcement price after 20 trading days.

NVIDIA quarter	D0	T+5	T+10	T+20
Q2 FY25, August 2024	-6.4%	-14.6%	-5.1%	-1.2%
Q4 FY25, February 2025	-8.5%	-10.6%	-11.8%	-13.3%
Q4 FY26, February 2026	-5.5%	-6.4%	-4.9%	-8.6%

All three fell further by T+5. All three failed to recover the pre-announcement price by T+20.

The sample is small, so this is not a deterministic rule. But it is enough to reject the simple idea that strong earnings plus a large drop automatically equals an immediate buying opportunity.

8. The real low and recovery took weeks

The second attached analysis goes further. It asks when NVIDIA actually bottomed and when it recovered.

Anchor	Pre-print base	Low	Max drawdown	Time to low	Breakeven	New high	Low-to-later-high
Q2 FY25, August 2024	$125.55	$102.78 on 2024-09-06	-18.1%	6 sessions	2024-10-07, 5.5 weeks	2024-10-14, 6.5 weeks	+45.3% to 2025-01-06
Q4 FY25, February 2025	$131.24	$94.29 around 2025-04-04 to 2025-04-08	-28.2%	27 sessions	2025-05-14, 11 weeks	2025-06-25, 16 weeks	+119.6% to 2025-10-29
Q4 FY26, February 2026	$195.55	$165.17 on 2026-03-30	-15.5%	23 sessions	2026-04-14, 6.5 weeks	2026-04-24, 8 weeks	+42.7% to 2026-05-14

The pattern is important.

First, the low did not arrive immediately. Even the fastest case took six trading sessions. The slower cases took more than a month.

Second, breakeven took 5.5 to 11 weeks.

Third, the rebound was eventually large. But it did not come from cheapness alone. It came after a catalyst removed the fear behind the selloff.

9. The rebound trigger had to kill the exact fear

9-1. August 2024: the Fed addressed macro fear

The August 2024 drawdown was about macro fear. Weak labor-market data raised recession concerns. Regulatory noise also hurt sentiment.

The rebound trigger was the Fed’s 50bp cut on September 18, 2024. The fear was rates and recession. The Fed directly addressed that fear.

9-2. February 2025: tariff fear required a policy reversal

The February 2025 post-print decline later deepened because of tariff shock. NVIDIA moved toward a 12-month low in early April.

The rebound trigger was the 90-day tariff pause. A policy shock created the fear; a policy reversal removed it.

9-3. February 2026: GTC addressed AI CapEx peak fear

This is the most relevant anchor for Samsung.

NVIDIA’s Q4 FY26 print was strong, but the market worried that AI investment might peak in 2026. Hyperscaler concentration, AI return-on-investment questions, and customer spending durability all mattered.

The rebound catalyst was GTC 2026. Jensen Huang said NVIDIA had visibility into at least $1T of demand or orders through 2027, compared with the prior year’s $500B figure through 2026. Data Center Dynamics and Constellation Research reported the comments.

This directly addressed the fear. The market worried that AI CapEx was peaking; NVIDIA answered that demand through 2027 was much larger.

Even then, the bottom came after GTC, not on GTC day. The stock bottomed on March 30, about two weeks later.

10. Applying the lesson to Samsung

Samsung’s selloff fear has two parts.

First, memory pricing and HBM demand may have been priced too quickly.

Second, if hyperscalers slow AI infrastructure spending, memory pricing and volume could both weaken.

Therefore, the rebound does not need another reminder that 2Q earnings were good. The market already knows that. The rebound needs evidence that the fear is wrong.

Samsung selloff fear	Required catalyst	Timing
Big Tech AI CapEx slowdown	Alphabet, Microsoft, Meta, and Amazon maintaining or raising 2026 to 2027 CapEx	late-July 2026 earnings calls
Memory price-growth slowdown	DRAM/NAND spot and contract-price momentum staying firm	weekly checks and 3Q negotiations
HBM catch-up doubt	HBM4/HBM4E qualification, production shipments, NVIDIA mix	2H26 IR and earnings calls
2Q one-off cost confusion	clean 3Q operating-profit path after bonus/provision adjustments	3Q preview and 3Q earnings
flow-driven liquidation	foreign and program-selling slowdown, lower leverage-product volatility	daily flow checks

The most important gate is late-July Big Tech earnings. If Alphabet, Microsoft, Meta, and Amazon keep or raise AI infrastructure spending plans, Samsung’s selloff can be reclassified as a flow correction rather than demand damage.

If two or more hyperscalers talk about slowing the pace of AI infrastructure investment, the NVIDIA Q4 FY26 template becomes more relevant and the correction can last longer.

11. Why the Fed put is weaker this time

Two of NVIDIA’s prior recoveries had macro or policy support.

In August 2024, the Fed cut rates by 50bp.

In early 2025, a tariff pause reversed a policy shock.

In February 2026, GTC provided a company-driven demand reset.

Samsung now has less help from the first two channels. Since June 2026, strong US labor data and inflation concern have made rates a constraint again. The market is not confidently pricing a near-term Fed rescue.

That makes the third channel more important: direct demand confirmation. Late-July hyperscaler CapEx commentary can play the same role for Samsung that GTC played for NVIDIA in March 2026.

12. Practical view: wait for proof, then scale in

The conclusion is not a strong sell call. It is also not an automatic buy call.

The stance is Wait.

A strong print followed by a two-day selloff can be a signal that the market has moved from current profit level to next growth rate. That is a different regime.

The entry gates are:

Gate	Meaning	If confirmed
late-July Big Tech CapEx maintained or raised	final demand for AI memory remains intact	selloff fear weakens
DRAM/NAND price momentum holds	2Q pricing logic continues into 3Q	3Q earnings support improves
Samsung clean 3Q QoQ remains healthy	underlying earnings path survives provision noise	quality of earnings improves

If two of these three gates pass, the case for staged entry improves. If all three weaken, the correction can extend.

The main mistake would be to assume that record earnings automatically lead to a bounce. NVIDIA’s history says otherwise. The rebound required a catalyst that killed the specific fear behind the selloff.

13. Samsung versus SK hynix

This note focuses on Samsung, but the read-through also matters for SK hynix.

Samsung has a huge earnings base, a lower multiple, and HBM4/HBM4E catch-up optionality. But investors also have to deal with HBM purity, foundry and S.LSI losses, bonus provisions, and foreign-flow pressure.

SK hynix has cleaner HBM exposure and a stronger position in NVIDIA’s supply chain. But it is already priced as the winner. If Big Tech CapEx remains strong, SK hynix is the trend-reacceleration candidate, while Samsung is the re-rating candidate where the Micron/NVIDIA read-through has been partly erased.

Lens	Samsung Electronics	SK hynix
Strength	earnings level, low multiple, HBM catch-up option	HBM purity, customer trust, leadership position
Burden	HBM certification, foundry/S.LSI, provision interpretation	high expectations, crowded-winner risk
Rebound condition	clean 3Q earnings and HBM customer proof	Big Tech CapEx durability and HBM allocation
Style	under-reflected alpha candidate	trend re-acceleration candidate

Both depend on the same final question: will hyperscalers keep spending?

14. Conclusion: Samsung needs better disproof, not just better earnings

Samsung’s 2Q26 preliminary earnings were strong. But the stock fell sharply for two straight days. Compared with 17 NVIDIA post-earnings reactions, Samsung’s path looks much closer to NVIDIA Q4 FY26 in February 2026 than to the August 2024 one-day sell-on.

In that NVIDIA case, the company posted strong numbers but failed to recover the pre-announcement price for about a month. The recovery began only after GTC 2026 directly challenged the AI CapEx peak fear with a $1T through-2027 demand statement.

Samsung’s equivalent proof point is not another reminder that 2Q earnings were strong. It is hyperscaler confirmation that AI infrastructure spending remains firm into 2027.

That is why the late-July Big Tech earnings calls matter. Until then, the right discipline is to wait for evidence rather than buying only because the stock has fallen after a record print.

Evidence classification

Fact

Item	Evidence
Samsung 2Q26 preliminary earnings	Revenue about KRW 171T and operating profit about KRW 89.4T, Samsung official newsroom, 2026-07-07
Samsung comparison periods	1Q26 revenue KRW 133.87T and operating profit KRW 57.23T; 2Q25 revenue KRW 74.57T and operating profit KRW 4.68T, Samsung official newsroom
Samsung price path	July 6 KRW 318,000, July 7 KRW 296,000, July 8 KRW 277,500, yfinance daily close calculation
Samsung two-day reaction	July 7 -6.92%, July 8 -6.25%, two-day -12.74%
NVIDIA Q4 FY26 event	February 25, 2026 event confirmed on NVIDIA IR
NVIDIA Q1 FY27	Revenue $81.6B, up 85% YoY and 20% QoQ, NVIDIA IR financial reports
GTC 2026 $1T comment	Demand/order visibility through 2027 reported by Data Center Dynamics and Constellation Research
NVIDIA price-path calculations	yfinance daily closes, 17 quarters aligned with next-day reaction for after-market releases

Inference

Item	Interpretation
Samsung/NVIDIA comparison	This is a price-path and fear comparison, not a business-model comparison.
Q4 FY26 similarity	Two-day path is the closest, and the fear set also overlaps: expectations, CapEx durability, and customer concentration.
Late-July Big Tech calls	If AI CapEx concern drove the Samsung selloff, hyperscaler guidance is the most direct disproof.
Wait stance	The print was strong, but the stock is asking about growth-rate durability. Confirmation should come before aggressive buying.

Speculation

Item	Possible path
Positive path	Big Tech CapEx holds or rises, DRAM/NAND price momentum holds, and Samsung’s clean 3Q earnings path improves.
Negative path	Two or more hyperscalers slow AI infrastructure spending or memory price momentum weakens.
Recovery timing	NVIDIA analogues suggest breakeven can take weeks, not days. For Samsung, mid-August to September is more realistic if catalysts turn positive.

Blocked

Item	Why it is blocked
Samsung 2Q26 segment detail	Preliminary earnings do not yet disclose DS, foundry, or S.LSI detail.
Late-July hyperscaler CapEx guidance	Not yet reported.
Samsung clean 3Q QoQ	Requires later earnings and IR detail after provision adjustments.
Exact consensus beat for all 17 NVIDIA quarters	Some quarters are verified, but a uniform consensus data set for every quarter was not fully obtained.

Sources and calculation notes

Samsung Electronics, Earnings Guidance for Second Quarter 2026
NVIDIA Investor Relations, Financial Reports
NVIDIA Investor Relations, 4th Quarter FY26 Financial Results
Data Center Dynamics, Jensen Huang says NVIDIA has one trillion dollars in orders through 2027
Constellation Research, Nvidia GTC 2026 and $1 trillion order flow
Stock calculations: yfinance daily closes for 005930.KS and NVDA, from 2022-05-20 to 2026-07-08. For NVIDIA, after-market releases are aligned so the next regular trading day is reaction D0.

The Big Tech Financing Relay: AI Capex Has Not Been Cut, And The Memory Bottleneck Looks Stronger

Tue, 07 Jul 2026 21:45:00 +0900

This is a follow-up to the late-July Big Tech earnings-call memory-thesis preview, the Micron FY3Q26 review, the 2027 semiconductor consensus-payability work, and the AI data-center capex bottleneck series.

TL;DR

The Big Tech financing relay says one thing clearly: AI infrastructure capex has not been cut. If anything, it has outgrown the old assumption that elite hyperscalers could fund everything internally.

Top hyperscaler 2026 capex guidance now sits around $695bn to $750bn. That compares with about $410bn in 2025 and about $245bn in 2024, implying roughly three consecutive years of 60% plus growth. The more important change is funding mix. The path has moved from internal cash to bonds, multi-currency bonds, off-balance-sheet structures, and now equity issuance.

For memory, the implication is cleaner than for hyperscaler equities. Microsoft has tied roughly $25bn of its 2026 capex to component-price inflation. Meta has cited component pricing, especially memory, as one reason for higher 2026 capex. Amazon has also warned about component and memory-cost pressure. The buyers are absorbing the price burden. Memory suppliers receive it.

The public-equity read-through is therefore not “own hyperscalers simply because they are spending.” It is more specific: watch the memory and HBM bottleneck nodes that can turn that spending into ASP, margin, and long-term contract visibility.

The Financing Timeline

Timing	Issuer	Type	Size	Interpretation
2025-10-30	Meta	Bonds	$30bn	Long-term AI and data-center funding
2025-11	Amazon	Bonds	about $15bn	First major return to US dollar bonds in years
2025-11	Alphabet	Bonds	about $17.5bn	Pre-funding AI infrastructure
2026-02	Alphabet	Bonds	about $32bn	Multi-currency deal including a 100-year tech bond
2026-02	Oracle	Debt-led funding	$25bn	Part of a larger cloud-infrastructure funding plan
2026-03-10/11	Amazon	Bonds	about $54bn	Dollar plus euro record-scale funding
2026-04-30	Meta	Bonds	$25bn	Issued just after Q1 results and higher capex guide
2026-05-12	Amazon	CHF bonds	CHF 2.82bn	Currency diversification
2026-06-01/03	Alphabet	Equity	$84.75bn	The decisive signal: dilution accepted for AI compute
2026-07-07	Amazon	Reported bond launch	$25bn plus	Reported eight-tranche US IG deal. Final terms not yet confirmed

The July 7 Amazon deal should not be mixed with the March record financing. As of publication, the final July 7 size, tranche spreads, new-issue concession and order book were not confirmed in public sources. It is therefore treated as a reported latest financing step, not a fully settled fact.

Capex Is Still Rising

Company	2026E capex	YoY	Key point
Amazon	about $200bn	about +50%	AWS, custom chips and data-center buildout
Microsoft	about $190bn	+61%	Roughly $25bn attributed to component-price inflation
Alphabet	$180bn to $190bn	about 2x	2027 expected to rise substantially again
Meta	$125bn to $145bn	about 2x	Memory and component pricing cited as pressure points
Oracle	volatile	high	Cloud buildout is highly capital-intensive
Top 5	$695bn to $750bn	about +67%	Versus about $410bn in 2025 and $245bn in 2024

The absolute dollar amount is still rising. The risk is the second derivative: even if 2027 capex reaches $1tn, the growth rate will likely slow from the 2025-2026 pace. That matters more for hyperscaler equity multiples than for memory suppliers. If memory keeps taking a larger share of the server bill of materials, memory revenue can still grow faster than total capex.

Why Funding Mix Matters

The escalation path is visible:

Internal cash.
Straight bonds.
Multi-currency bonds.
SPVs, private credit and off-balance-sheet structures.
Equity issuance.

Alphabet’s June 2026 equity raise is the key tell. This is not a weak company. It is one of the strongest cash-flow machines in the world. If it chooses dilution to fund AI compute, the capital intensity of the AI buildout has moved beyond the old internal-cash model.

Microsoft looks different. Its visible public benchmark funding has been less aggressive than Alphabet’s or Amazon’s, which may point to stronger internal funding capacity rather than lower AI capex ambition.

The Memory Read-Through

The key change is not simply that memory prices are up. It is that hyperscalers are not cutting demand in response. They are increasing budgets, signing long-term contracts and raising capital.

Signal	Memory implication
Microsoft component-inflation comment	More dollars are needed for similar capacity
Meta memory-price comment	Memory is moving the whole capex budget
Amazon component-cost warning	Costs are higher, but capex continues
Micron SCAs	Customers are locking in long-term supply
HBM4/HBM4E transition	Higher value per accelerator and more wafer intensity

This is why hyperscaler equities and memory suppliers should be separated. Hyperscalers bear the cost, FCF drag and payback risk. Memory suppliers receive the pricing and allocation benefit.

Public Equity Checklist

For Samsung Electronics, the questions are DS core profit, HBM4/HBM4E customer qualification, 3Q DRAM and NAND pricing, and the scale of foundry/System LSI losses.

For SK Hynix, the questions are HBM4 long-term contract structure, 2027 allocation, and whether US-listed access through the SKHY ADR broadens the investor base without distorting common-share pricing.

For Micron, the question is whether the SCA structure becomes the benchmark for how investors value memory-cycle durability.

For hyperscalers, the clean stance is more cautious: demand is real, but FCF, debt, dilution and payback period risk are also real.

Scenarios

Scenario	Condition	Memory read-through
Bull	Amazon order book strong, late-July calls maintain or lift 2027 capex, memory costs remain absorbed	HBM and server DRAM pricing durability strengthens
Base	2026 capex maintained, 2027 growth continues but slows, memory-price increases moderate	Memory thesis remains intact but stock reactions stay selective
Bear	Capex guides cut, payback concerns rise, component inflation causes deployment delays	ASP thesis weakens until contract pricing is confirmed
Structural damage	Customers reduce memory content per server, HBM contract prices fall, bond demand weakens	Memory thesis needs full re-underwriting

Final View

The Big Tech financing relay confirms that AI infrastructure demand remains real. But it is not a simple hyperscaler-equity bullish signal. The more precise read-through is that memory is one of the bottlenecks taking a larger share of AI infrastructure dollars.

The next tests are the final terms of Amazon’s July 7 deal, the late-July Big Tech calls, Samsung’s July 30 earnings call, SK Hynix’s 2Q call, and 3Q/4Q memory contract-price data.