The Comeback Nobody Believed

Intel's Q1 earnings tonight were supposed to be quiet. Analysts had the company earning about a penny a share. Revenue was expected to be flat. Nobody was looking for a surprise — because the last surprise from Intel, in Q4, was the bad kind: the company warned it was running low on wafers and guided investors toward disappointment.

Instead: earnings of 29 cents a share. Revenue of $13.6 billion, more than a billion above expectations. Q2 guidance so strong that it made the Q1 numbers look modest. The stock jumped 20% in after-hours trading. It has now more than tripled from its 2024 lows. The company that Silicon Valley had quietly written off as the permanent loser of the AI era just posted one of the most surprising quarters in its history.

To understand what happened tonight, you have to go back eighteen months — and follow three separate threads that all converged at once.

The Brief

• The numbers tonight were genuinely surprising — even to Intel. Q1 revenue of $13.6 billion beat consensus by more than 9%. Adjusted EPS of $0.29 was 29 times what analysts expected. The data center and AI division grew 22% to $5.1 billion. Q2 guidance of $13.8–14.8 billion was far above Wall Street's $13.07 billion forecast. After the Q4 call, where Intel said it was running short on wafers, a result this strong raises a legitimate question: if things were this good, why was guidance so pessimistic three months ago? ✓ CNBC · Apr 23 / Sherwood News · Apr 23

• The CPU market woke up — and Intel happened to own it. For the last three years, the AI story was all about Nvidia GPUs and the massive data centers built around them. But there is a second wave: as AI moves from training large models to actually running them at scale — answering queries, running agents, doing inference — CPUs are becoming load-bearing again. Intel makes most of the world's server CPUs. Google has committed to using Intel's Xeon processors and custom chips for AI inference on Google Cloud. These are not exploratory conversations. They are production commitments. ✓ CNBC · Apr 23 / Intel earnings call

• Elon Musk just handed Intel a landmark customer announcement. On Tesla's Q1 earnings call Wednesday, Musk said Tesla, xAI, and SpaceX plan to use Intel's 14A manufacturing process at the Terafab chip complex being built in Austin, Texas. This is the first major public commitment to Intel's next-generation 14A node from a marquee name. If Musk follows through, it validates Intel's foundry roadmap in a way that years of management presentations could not. ✓ CNBC · Apr 23 / Tesla Q1 earnings call · Apr 22

• The US government bought in at $20.47. Tonight the stock trades around $78. In August 2025, the White House negotiated an $8.9 billion stake in Intel at $20.47 per share under the finalized CHIPS Act terms. In January, Trump called CEO Lip-Bu Tan "very successful" on Truth Social. The stock jumped 10% that day. The government's return tonight is roughly 3.8 times what it paid. That is not a reason to buy Intel stock. But it is a reason why the political support behind Intel's foundry is not going away. ✓ CNBC · Jan 9, 2026 / Trefis

• Intel bought back its Irish fab — and that tells you something about confidence in its margins. In 2024, Intel sold a 49% stake in its Leixlip, Ireland fab to Apollo Global Management for $11.2 billion, as the company was burning cash. In April 2026, Intel bought that stake back for $14.2 billion. Paying more to get back what you sold cheap is only a good move if you expect the asset to keep growing in value. Intel's board made that call before tonight's earnings, suggesting internal confidence in the trajectory was already high. ✓ Intel earnings release / Reuters · Apr 2026

How Intel Got Here

In 2024, Intel was in genuine trouble. The company had promised for years that it would catch up to TSMC and Samsung on manufacturing. It kept missing. Engineers left. Customers — including Apple, which had relied on Intel chips for its Macs for fifteen years — switched to TSMC. AMD ate into Intel's server market share. When Nvidia's AI chip boom happened, Intel was not in it. Its GPU business was too small. Its data center CPUs were losing share. The stock lost 60% in a year.

The turnaround started with a new CEO. Lip-Bu Tan took over in March 2024. He had run Cadence Design Systems for twelve years, knew the chip industry inside out, and came in with a mandate to cut layers of management and fix the manufacturing. Within a year, 18A — Intel's most advanced chip process — entered high-volume production. The yield has been improving 7% per month.

Then the AI market shifted in a way that happened to benefit Intel specifically. The first phase of AI — training giant foundation models — was almost entirely about Nvidia GPUs. Intel had essentially no part in that. But the second phase — inference, the act of running a trained model to actually answer questions and do things — puts a different kind of load on data centers. Inference needs CPUs. Lots of them, running constantly, handling the coordination of AI workloads. Intel's Xeon server CPUs were already in most data centers. They went from a commodity to a critical component almost overnight.

"The CPU is reinserting itself as the indispensable foundation of the AI era."

— Lip-Bu Tan, CEO of Intel · Q1 2026 earnings call, April 23, 2026 ✓ Intel Q1 2026 earnings call

Three Things That Have to Keep Going Right

Intel's stock has priced in a lot of good news. At roughly $78 after hours, it trades at nearly 100 times forward earnings — a valuation that leaves no room for setbacks.

First, the 18A yield has to keep improving. The process is in production, but yields — the percentage of chips that come off the line working correctly — are not yet where they need to be for Intel to compete with TSMC on cost. Some wafers have had defect issues. The monthly 7% improvement is the right direction, but Intel is currently the only major customer of its own 18A fabs. Winning a second one — a company that does not already have a manufacturing relationship with Intel — is the milestone that changes the story from "promising" to "proven."

Second, the Musk commitment needs to hold. Tesla's announcement that it will use Intel's 14A process at Terafab is the most significant foundry customer signal Intel has had in years. But 14A is not expected to start production until 2027 or 2028. Musk has a history of moving quickly on things that excite him and quietly deprioritizing things that don't. The announcement is credible. It is not yet a contract with money behind it.

Third, the CPU demand has to be real, not a one-time restocking. If data centers simply under-bought CPUs during the GPU boom and are now catching up, the 22% growth disappears once they have what they need. If AI inference keeps expanding — more queries, more agents, more products running models in the background — then Intel's data center business has years of demand ahead that nobody had in their forecast two years ago. Tonight's earnings do not answer this question. They make it more urgent.

The Read

The Intel story of 2024 and 2025 was about a company deciding whether it wanted to exist at the frontier of chip manufacturing. Most of the industry had already written its obituary. TSMC was untouchable. Samsung was struggling. Intel was supposed to be somewhere between irrelevant and bankrupt. The decision to go all in on 18A, to bring in a new CEO, and to take the US government's money at $20.47 a share was a bet on a thesis that almost nobody held: that America would eventually decide it could not afford to have all its advanced chips made in Taiwan, and that Intel was the only candidate to change that.

That thesis is now being validated one quarter at a time. It is not complete. Intel's foundry still has no major external customer signed and shipping. Its GAAP earnings are still negative. The stock at 100x forward earnings prices perfection. But the direction is right, and the timing turned out to be exactly what the country needed — a domestic chip manufacturer that could absorb government investment, get the manufacturing right, and be ready when the AI inference demand arrived. Tonight, all three of those things are true simultaneously.

EP18 said the war damaged the economy's fuel line but did not touch its brain — and TSMC was the brain. Tonight's Intel earnings add a second layer to that story: the brain has a spine now too. TSMC makes the GPU that trains the model. Intel makes the CPU that runs it. Both are sold out. Both are building more. The AI infrastructure buildout is not a single horse race — it is the whole stable running at once, and Intel just reminded everyone it is still in the race. ~ Framework

Market Truths · 財經真言 · Published Tuesday, Thursday, Saturday · markettruthspod.com