Why IBM's Earnings Miss Is Evidence of Memory Strength: Reading IBM and Ericsson for Semiconductor Sentiment

Tue, 14 Jul 2026 20:00:00 +0900

Context This piece is the demand-side counterpart to the supplier-side story covered in SK Hynix’s Q2 Profit Cut and Unchanged Target Price: that the spot-price spike does not fully flow into blended ASP. It is the counterpart because IBM’s own customers pulled purchases forward precisely out of fear of that same spot-price spike. It pairs well with Who Pays for the 2027 Semiconductor Consensus, The Supply Gap Spreading Beyond HBM, and Late-July Big Tech Earnings Calls and the Memory Thesis. Related hubs are the AI HBM Hub and the Exclusive Analysis Hub.

TL;DR

IBM disclosed preliminary Q2 results on July 14 that missed consensus. Revenue came in at $17.2 billion (consensus $17.86 billion) and adjusted EPS at $2.93 (consensus $3.01), and the stock plunged more than 17% in premarket trading. The formal earnings call is still scheduled for July 22.
One of the three stated reasons for the miss is decisive for semiconductor investors. CEO Arvind Krishna wrote that “in the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases."
The same day, Ericsson reported revenue of SEK 52.7 billion, down 6%, and warned of margin pressure from component-cost increases tied to AI demand. It said the pressure would build further in coming quarters, requiring price increases and cost cuts.
Lay the two prints on top of each other and one picture emerges. IBM shows it from the demand side and Ericsson from the cost side: memory price increases have begun showing up in real corporate earnings.
There are two sides to what this means for semiconductor sentiment. That memory tightness has now been confirmed by end-customer behavior rather than supplier claims is a bullish argument. But the cost is erosion of other IT spending, and the late-June pull-in leaves a Q3 payback risk hanging over the numbers. [Analysis scope]

Key Framing

If the reason IBM could not sell software is that its customers were hoarding memory, that is bad news for IBM but demand evidence for memory. Yet the same sentence also proves the zero-sum: memory's gain is other IT spending's loss.

1. IBM: What the Preliminary Earnings Warning Actually Said

Let’s establish the facts first. What came out on July 14 was not the formal results but a preliminary result and a CEO investor letter. IBM said the final figures may differ somewhat from the preliminary numbers, and the formal earnings call remains scheduled for July 22 as planned. [Fact: IBM newsroom]

The Numbers

These are the preliminary figures from CEO Arvind Krishna’s investor letter. [Fact: IBM investor letter]

Item	Result	Note
Revenue	$17.2 billion	+1%, below consensus of $17.86 billion
Adjusted (non-GAAP) EPS	$2.93	+5%, below consensus of $3.01
GAAP diluted EPS	$2.27	-2%
GAAP gross margin	57.7%	-100bp
Operating gross margin	59.4%	-70bp
YTD operating cash flow	$7.8 billion	-
YTD free cash flow	$4.8 billion	-

By segment, Software rose 5%, Consulting rose 1% at constant currency, and Infrastructure fell 7%. The stock dropped more than 17% in premarket trading. [Fact: CNBC]

The Three Reasons for the Miss

CEO Krishna cited three reasons directly.

First, softness in the z17 mainframe. He called it “the strongest start to a mainframe program in history,” yet said Transaction Processing and the related software stack fell short of expectations. The z17 program itself is actually running at 130% of the prior program’s pace.

Second, customers reprioritizing capex. This is the sentence semiconductor investors should focus on.

“In the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases.”

Third, an execution gap. He acknowledged: “We did not adapt and move quickly enough, and numerous large deals failed to close on the timelines we expected.”

What Wasn’t Bad

Not everything was bad. Red Hat growth accelerated to 11%, and Distributed Infrastructure surged 37% while building a backlog of roughly $500 million. [Fact: IBM investor letter] In other words, the picture is that AI infrastructure and open hybrid cloud are alive and well, while the traditional mainframe and the software stack sitting on top of it lost ground. [Inference: reading of the segment data]

2. Ericsson: Same-Day Confirmation From the Cost Side

On that same July 14, Ericsson also reported Q2 results. This one is a cost story. [Fact: Ericsson results release]

Item	2Q26	Prior Year	Change
Revenue	SEK 52.7 billion	SEK 56.1 billion	-6%
Adjusted EBITA	SEK 6.9 billion	-	-7%
Net income	SEK 4.1 billion	-	-12%
Networks revenue	-	-	-8%

The revenue decline reflects weaker IPR licensing income and currency headwinds offsetting growth in several regions. That said, some assessments note that earnings themselves came in above market expectations. [Fact: Reuters/RTE reporting]

What matters for investors is CEO Börje Ekholm’s warning. He said Ericsson is under pressure from component-cost increases tied to AI demand, that the company has taken measures in response, and that the effects will show up more strongly in coming quarters. The company said cost pressure will keep building and that it needs to offset the margin impact through price increases and cost cuts. In Q3, Networks’ adjusted gross margin is expected to see some pressure from higher volumes of network rollout projects. [Fact: Ericsson results release]

3. Put the Two Together: Memory Pricing Has Reached the Real Economy

IBM and Ericsson differ in business, geography, and customer base. Yet on the same day, they show different faces of the same cause.

Company	Where It Showed Up	Specific Symptom
IBM	Demand side	Customers pulled forward memory, server, and storage purchases, delaying software and mainframe spending
Ericsson	Cost side	AI-demand-driven component cost increases pressure margins, requiring price pass-through

In other words, memory price increases are no longer a story confined to the earnings tables of the three memory makers. They have begun showing up on the income statements of the buyers.

This pattern did not appear out of nowhere. Semiconductors also sold off on June 26, and the sell logic back then was exactly this: concern that rising memory prices would ripple across the broader tech industry, pressuring margins and choking off demand. That day, Micron and SanDisk fell more than 6% and Western Digital fell more than 7%, and Apple’s and Microsoft’s product-price increases to offset rising memory costs were cited as evidence. [Fact: June 26, 2026 reporting]

The scenario the market feared in June has now obtained real-world evidence through the two companies’ July 14 results. [Inference: connecting the two events]

4. The Impact on Semiconductor Sentiment: Separating the Two Sides

Whether you read this news as bullish or bearish for semiconductors depends on which sentence you focus on. Let’s honestly separate the two sides.

The Bull Case: Demand Confirmed by Customer Behavior, Not Supplier Claims

The biggest weakness of the memory bull case has always been that “the demand talk comes from the supplier.” Even when Samsung, SK Hynix, and Micron say the market is tight, the market has discounted it as sell-side optimism. As covered in The Worst Case Is Already the Price, the market has been pricing in roughly half of consensus and refusing to re-rate.

But IBM’s statement is a different kind of evidence. The CEO of a company that buys memory, not one that sells it, said his customers hoarded memory, and he said it to explain why his own results were weak. That makes it credible precisely because it is testimony against his own interest. The phrase “to secure supply-constrained infrastructure ahead of expected price increases” is direct evidence that memory shortage and price-increase expectations have penetrated all the way to the end-customer level. [Inference: the nature of the testimony]

Ericsson’s warning points the same way. That rising component costs require price increases means memory and component suppliers’ pricing power is real.

The Bear Case: The Cost Is Other IT Spending and Margins

The flip side of the same sentence also has to be seen. IBM customers’ capex is not unlimited. Whatever shifted toward memory, servers, and storage came out of software and mainframe spending. Ericsson’s margin was likewise squeezed by component costs. This is the real-world version of the payability problem covered in Who Pays for the 2027 Semiconductor Consensus.

Memory’s gain is not free; it is a loss to some other IT budget. If this zero-sum dynamic persists, two things follow. First, set makers and software vendors raise prices, which then suppresses end demand (Apple and Microsoft have already done this). Second, rising memory prices eventually eat into their own demand. [Inference: the zero-sum structure]

The Most Practical Risk: Payback From the Pull-in

There is a third axis investors can easily miss. The pull-in Krishna described was concentrated in the last few weeks of June. That means Q2 memory demand may have been inflated.

A pull-in is, by definition, borrowing future demand. That means Q3 could see a corresponding gap. This risk should be read not as a memory-short argument but as a timing-distortion warning. Demand is not disappearing; rather, an optical illusion can form where Q2 looks strong and Q3 looks weak. [Inference: the structure of the pull-in]

This view meshes precisely with the SK Hynix Q2 profit cut analysis. On the supplier side, LTAs meant the spot-price surge did not fully flow into blended ASP, lowering Q2 earnings. On the demand side, customers who feared that very spot-price surge pulled purchases forward. These are two faces of the same phenomenon, and both need to be confirmed in Q3.

5. Market Reaction and Implications for the Korean Market

Confirmed Reactions

IBM shares plunged more than 17% in premarket trading. [Fact: CNBC]
Reports emerged that IBM’s warning spread panic across software and consulting names broadly. [Fact: market reporting]
Ericsson shares also fell.

How This Lands on the Korean Market

From the perspective of the Korean market, the character of this news is delicate. Yesterday (July 13), SK Hynix plunged 15.37% as broker earnings cuts and outflows coincided, then rebounded this morning. The IBM news that broke that same night carries both evidence that “memory demand is real” and a warning that “IT budgets are being eroded as the cost” at once.

So how the Korean market reacts tomorrow depends on which sentence the market reads first. Reading the demand evidence favors memory; reading the zero-sum dynamic and the pull-in payback weighs on it. [Inference: divergence in market interpretation]

One thing, though, is clear. This event has shifted the center of gravity in the debate over whether memory demand is real. The question is no longer “does demand exist,” but has moved to “how sustainable is that demand, what is real demand once the pull-in is stripped out, and when does the eroded IT budget push back.” [Inference: shift in the point of contention]

6. What to Watch From Here

This preliminary result is not final. A sequence of milestones lies ahead that need confirmation.

IBM’s formal earnings call on July 22: final figures, full-year guidance, and commentary on whether the pull-in was a one-off or a trend
Big Tech earnings calls on July 28-30: the slope of 2027 capex and whether memory price pressure gets absorbed. This is the judgment event covered in Late-July Big Tech Earnings Calls and the Memory Thesis
Samsung Electronics’ Q2 final results on July 30: DS core-business profit and Q3 pricing commentary
Q3 memory shipments: whether the payback from the late-June pull-in actually materializes
Ericsson’s Q3: whether component-cost pressure is offset by price pass-through, or whether margins keep getting squeezed

Once these five are confirmed in sequence, it will become clear whether this IBM statement was evidence of memory strength or a signal of the cycle’s peak.

Closing

IBM’s earnings miss is unambiguously bad news for IBM. But reading the reason behind the miss turns it into different information for semiconductor investors. The sentence “customers could not buy software because they were hoarding memory” carries weight precisely because it is testimony left by the buyer of memory, not the seller.

At the same time, that same sentence is also a warning. The money memory is taking does not fall from the sky. It comes out of IBM’s software revenue and out of Ericsson’s margin. How long that zero-sum dynamic can persist is the real question of this cycle.

And the pull-in concentrated at the end of June may make Q2 look good and Q3 look bad. Before reading the numbers, this distortion needs to be stripped out first.

This post is an analysis based on IBM’s preliminary Q2 results and CEO investor letter released on July 14, 2026, Ericsson’s Q2 results release, and public reporting. IBM’s figures are preliminary and final results may differ; the formal earnings call is scheduled for July 22. The names mentioned are examples for analysis, not a recommendation to buy or sell any specific stock. The interpretation of market reaction and the impact on semiconductor sentiment is inference and requires verification against future earnings and guidance. Investment decisions and their outcomes are the sole responsibility of the investor.

Ericsson on Korea Invest Insights