Intel’s second quarter closed on June 30, 2026. On Monday, July 20, word got out that the Data Center and AI Group was about to be cut. On Tuesday, July 21, Intel confirmed it. On Thursday, July 23, after the closing bell, Intel reported the strongest revenue growth it has posted in more than fifteen years.
Three days. That is the gap between telling engineers they are surplus and telling shareholders those same engineers just delivered the best quarter of the decade.
And here is the detail that ends any argument about what management knew. The quarter closed three weeks before the layoffs were announced. Intel’s finance organization had the segment numbers. The CFO had them. The CEO had them. The decision to cut the Data Center and AI Group was made by people who already knew exactly what the Data Center and AI Group had produced.
What the group produced
Straight from the earnings release, no interpretation required. DCAI revenue was $6.262 billion, up 59 percent year over year from $3.939 billion. Operating income was $2.474 billion against $633 million a year earlier. That is not growth. That is a near-quadrupling. Operating margin went from 16.1 percent to 39.5 percent.
Now set it against the rest of the company. Intel Foundry did $5.765 billion in revenue and posted an operating loss of $2.089 billion. Corporate unallocated burned another $1.416 billion. The Data Center and AI Group, on its own, generated enough operating income to cover the entire foundry loss with $385 million left over.
The single most profitable thing happening inside Intel right now is the division being told to align its organization.
The language
Intel’s statement was that the data center group is aligning its organization to ensure it has the right roles and skills in place to position the business for long-term success.
Notice what that sentence does not say. It does not say too many people. It says wrong people. That distinction is deliberate and it is load-bearing, because right roles and skills is the formulation that lets a company cut and hire in the same quarter without anyone calling it a contradiction. Anyone who has sat on the receiving end of a reorg deck recognizes the phrase on sight. It is what gets written when the headcount number comes down from finance and the org has to reverse-engineer a story about capability.
Follow the money, because it did not evaporate
It moved, and you can trace it line by line through the filings.
Start in April. Intel paid Apollo $14.2 billion to buy back the 49 percent stake in the Fab 34 joint venture in Ireland, the same stake Intel sold in 2024 for net proceeds of roughly $11 billion. Apollo cleared about 29 percent in under two years. Intel funded it with cash on hand and a $6.5 billion bridge loan. It shows up in the six-month cash flow statement as $14.3 billion in partner distributions, against $91 million in the comparable prior-year period.
Then look at what management committed to on the call. Capital expenditure for 2026 moved from roughly $18 billion to more than $20 billion, with 2027 guided significantly higher. A five billion euro European manufacturing expansion. Expanded mask operations at the Bowers campus. Tool purchase orders locked in, clean room build-outs accelerated, substrate and memory supply secured.
Dave Zinsner said it plainly in the release: to support expected growth across products and foundry, Intel is meaningfully increasing its investments in equipment, clean room space, and substrates.
Equipment. Clean rooms. Substrates. He did not say engineers.
The balance sheet shows the strain. Adjusted free cash flow for the quarter was negative $8.4 billion. Total debt climbed from $44.1 billion to $48.5 billion in six months. Retained earnings fell from $49.0 billion to $33.8 billion.
So here is the mechanism, stated without euphemism. Payroll is being converted into fab capacity. A private equity firm was made whole at a three billion dollar premium in April. The engineers who produced a 59 percent growth quarter were told in July that the company needs to become more efficient.
They cut R&D during the best quarter in fifteen years
Research and development spending in the second quarter of 2026 was $3.368 billion. In the second quarter of 2025 it was $3.684 billion. Down 8.6 percent, in a quarter when revenue rose 25 percent.
Read that again. The strongest revenue growth in more than fifteen years, and R&D went down.
Meanwhile the full-year 2026 reconciliation shows $4.3 billion of restructuring and other charges baked into GAAP operating expenses. Intel has budgeted four and a third billion dollars for the cost of removing people this year, while cutting the budget for inventing things by three hundred million.
The eleven billion dollar fig leaf, and what sits underneath it
Every headline said Intel lost $11 billion. Technically true and almost entirely meaningless. GAAP net loss attributable to Intel was $11.033 billion, driven by a $12.529 billion mark-to-market charge on escrowed shares tied to the CHIPS Act Secure Enclave agreement with the federal government. Accounting treats those shares as a derivative liability, so when Intel’s stock price rises, Intel books a loss. The company lost eleven billion dollars on paper largely because its own equity got more valuable. GAAP operating income was actually positive $1.796 billion, an 11.1 percent margin against negative 24.7 percent a year earlier.
So the headline number is theater. Fine. But there is something underneath it that is not theater, and it is printed in Intel’s own earnings release: these non-GAAP financial measures are used in our performance-based RSUs and our cash bonus plans.
Now go look at what non-GAAP adds back. Share-based compensation, $687 million, added back. Restructuring and other charges, $170 million in the quarter, added back. Acquisition-related adjustments, added back. The escrowed shares charge, added back.
And restructuring charges are, in Intel’s own description, primarily related to employee severance and benefit arrangements.
Follow that all the way down. The severance paid to the people being cut is explicitly excluded from the earnings measure that determines executive bonuses and performance stock units. In the number that pays the people making the decision, firing people is free. It costs nothing. It is scrubbed out before the calculation runs.
That is not a conspiracy theory and it is not an inference. It is disclosed, in the same document that reported the record quarter, four pages after the segment results.
The man at the top
Lip-Bu Tan was hired in March 2025 on a package valued at roughly $69 million: one million in base salary, up to two million in bonus, and about sixty-six million in equity awards and options including a twenty-five million dollar new-hire option grant. He also agreed to buy $25 million of Intel stock with his own money. Compensation analysts estimated at the time that the package could exceed $400 million if he tripled the stock and beat the S&P 500. His reported total compensation for fiscal 2025 was $93.0 million. His 2026 annual package is $27 million, which the compensation committee describes as 96 percent at risk.
The stock has gone from roughly $23 to north of $100. It is up well over 300 percent in a year. He is deep in the money.
There is more history worth knowing. Tan was CEO of Cadence Design Systems from 2009 to 2021. In July 2025 Cadence pleaded guilty and agreed to pay more than $140 million to resolve charges that it violated U.S. export controls by selling chip design tools to China’s National University of Defense Technology, a military-linked institution, over a period running from 2015 to 2021 that overlaps his tenure. Tan was not named in the indictment. Senator Tom Cotton wrote to Intel’s chairman citing Tan’s reported control of dozens of Chinese companies and stakes in hundreds of Chinese chip and advanced manufacturing firms. On August 7, 2025, the President publicly called for his immediate resignation.
Tan kept the job. Days later he was at the White House. Two weeks after that, the federal government converted $8.9 billion of unpaid CHIPS Act and Secure Enclave money into 433.3 million Intel shares at $20.47 apiece, a 9.9 percent stake priced below market, with the clawback and profit-sharing provisions attached to the earlier $2.2 billion in grants removed. At current prices that position is worth roughly five times what was paid for it.
I want to be precise about what I am and am not claiming. I am not saying Tan committed a crime; he was not charged. I am not saying the turnaround is fake; the segment numbers are real and 18A appears to be working. I am saying something narrower and much harder to argue with.
Every actor in this story has upside except the people doing the work.
The CEO’s options are worth a fortune. Apollo cleared three billion. The federal government is sitting on a multibillion dollar paper gain. Nvidia and SoftBank are up on their stakes. The engineer in Hillsboro who shipped Xeon 6 gets thirteen weeks of severance plus time served.
The ledger
The employee count is in the earnings release and it is not ambiguous. Intel proper, excluding subsidiaries: 96,400 people in June 2025, 78,500 in March 2026, 77,600 in June 2026. Total Intel including Mobileye and other subsidiaries went from 101,400 to 82,300.
That is roughly nineteen thousand fewer people in twelve months. About three thousand of that is the Altera deconsolidation rather than layoffs, so call the real number something close to sixteen thousand. Nine hundred left in the second quarter alone, the quarter Intel just called its best in fifteen years.
Widen the frame and it gets worse. Intel employed around 132,000 people in 2022. It cut 12,000 that year, roughly 15,000 in 2024, and more than 20,000 across 2025. In Oregon alone, where Intel is the largest for-profit employer with about twenty thousand workers, it cut nearly 2,400 in July 2025 and another 669 that November, more than 3,100 in a single year in a single county. Among the July cuts were over three hundred module development engineers, the people whose job is developing next-generation chips.
Severance in prior rounds has run at least thirteen weeks of pay plus additional weeks based on years of service. For a fifteen-year Intel veteran in Washington County, that is what a career converts into.
The silence
Here is the detail I cannot get past.
The Q2 2026 earnings call ran over an hour. Nine sell-side analysts from Melius, Morgan Stanley, Bernstein, UBS, Bank of America, Cantor Fitzgerald, Wells Fargo and others asked questions. The CEO spoke. The CFO spoke. They covered 18A yields, High-NA EUV, Clearwater Forest, Diamond Rapids, Coral Rapids, ASIC run rates, substrate constraints, memory pricing, PC seasonality, capex phasing, and a possible move into memory.
Nobody mentioned the layoffs. Not Tan. Not Zinsner. Not one analyst.
The company had announced seventy-two hours earlier that it was cutting the division that grew 59 percent, and across an entire public earnings call, not one professional whose entire job is interrogating management thought it was worth a question.
That is the real story. Not that Intel cut. That cutting the best performing group in the company during its best quarter in fifteen years was so unremarkable to everyone in the room that it never came up.
And one more thing buried in the same release
Under the heading Culture and Leadership, sandwiched between executive appointments, Intel disclosed that it and Google Cloud expanded a multi-year collaboration to scale AI capabilities across Intel’s workforce as part of Intel’s AI-powered transformation. The underlying announcement, made days before the layoff news broke, describes integrating Gemini-powered generative AI across Intel’s global workforce and introducing custom agentic workflows to speed the chip design lifecycle.
I build agentic systems for a living. I am not going to tell you Gemini Enterprise replaced a data center architect this quarter, because it did not, and anyone claiming otherwise is selling something. But I will tell you what that sequence looks like from the inside, because I have watched it happen at close range more than once. You deploy agents into the design lifecycle. You declare an AI-powered transformation. Then you discover the org needs different roles and skills. The tooling arrives first, the headcount rationale arrives second, and the causality never has to be stated out loud because it never has to be defended.
The honest counter-argument
I owe you the other side, because it is not nothing.
DCAI was staffed for an Intel that owned nearly the entire server CPU market at fat margins, and that Intel is gone. AMD’s EPYC has been taking share on performance per watt for years. Arm-based server silicon from Ampere and others is real. The hyperscalers are building their own: Graviton and Trainium at Amazon, Axion and TPU at Google, Cobalt and Maia at Microsoft. Intel’s revenue per employee is still poor next to AMD and embarrassing next to Nvidia. Some of the roles being cut supported cancelled programs. Some of this is genuine redeployment toward 18A and advanced packaging, where the actual constraint lives. And Intel is supply constrained right now, with customers wanting more chips than it can make. When the bottleneck is wafers, adding architects does not ship one additional CPU this year, while capital does.
All of that is true. And none of it survives contact with the timeline.
If this is real portfolio surgery, you announce it on Thursday alongside the beat, with a number, with a plan, and with a stated destination for the people affected. You say how many. You say which teams. You say how many are being offered internal transfers into the 18A and packaging organizations you are simultaneously funding with twenty billion dollars. You take the question from Stacy Rasgon and you answer it.
What you do not do is let it leak on a Monday, decline to disclose the count, go silent for three days, and let the people you cut find out from CNBC that the division they were cut from carried the company.
So why do they do it
Because in the accounting that actually governs the decision, the people who produced the number never appear as the people who produced the number. They appear as operating expense. And the cost of removing them is added back before anyone’s bonus is calculated.
That is the whole answer. Everything else is press release.
Intel’s stock has quadrupled through eighteen months of continuous layoffs. The market has learned that cuts precede rallies at this company. Management has learned it too. No executive at Intel has suffered a single adverse consequence for cutting, and several have made generational money doing it. Until that incentive changes, the beat and the cut will keep arriving in the same week, in that order, and the cut will keep coming first.
If you are in DCAI right now
The record quarter is not your protection. It is your funding.
Understand the position you are actually in. You are the profitable division inside a company spending twenty billion dollars a year trying to become something else. Your margin is the subsidy. That makes you valuable to the enterprise and expendable to the plan at the same time, which is a genuinely strange thing to be, and nobody is going to explain it to you in your next one-on-one.
So document what you shipped, with numbers, this week. Find out what your work is worth on the open market before somebody else tells you what it is worth. And internalize the one durable lesson here, which has nothing to do with Intel: the value you create does not accrue to you automatically. It accrues to whoever holds the equity, unless someone deliberately builds a mechanism to route it back to the people who made it. Almost nobody builds that mechanism. That is not cynicism. That is just the org chart.
Intel just told 77,600 people that their best quarter in fifteen years was a cost problem.
Believe them.
About the Author
Gal Ratner is the founder and CTO of Inverted Software and WhiteStar Labs, and Chief Architect at Prana Entertainment. He has spent nearly thirty years shipping production software on the Microsoft and .NET stack for clients including Microsoft, Sony, Rockstar Games, 2K Games, Best Buy, and Allegiant Air. He was employee number six at Break.com and a Los Angeles Business Journal CTO of the Year finalist.
His current work is production agentic AI: MCP servers, the Microsoft Agent Framework, RAG pipelines, SQL Server 2025 vector search, and the PLogger observability framework. He also runs ShopSnap, a zero-transaction-fee e-commerce platform built for merchants the incumbents will not serve. He is the author of the novel The Archive of Lost Suns, trains Brazilian jiu-jitsu under Sergio Penha in Las Vegas, rides motorcycles, and co-hosts Edge Grip Podcast.
He is the .NET AI guy that ships.