Intel Inside the AI Boom: The Old-Timer Crashes the Party

Intel posted its fastest revenue growth since 2011. Here’s what its comeback says about AI infrastructure, American manufacturing, and markets.

KEY TAKEAWAYS

• Intel reported $16.1 billion in second-quarter revenue, an increase of 25% from the prior year. Adjusted earnings of 42 cents per share were twice Wall Street’s expectation.

• Data Center and AI revenue jumped 59% to $6.3 billion, and Intel says it cannot keep up with server-chip orders. That suggests the AI infrastructure boom is broadening beyond GPUs.

• The shift from model training to inference and agentic AI changes the required mix of chips. Intel believes the ratio could move from one CPU per eight GPUs during training toward one-to-one in agentic workloads.

• Intel’s reported $11 billion GAAP loss came largely from a $12.5 billion non-cash charge tied to its rising share price. It is a useful reminder that headline earnings and underlying cash economics can tell very different stories.

• Intel’s newest chips are being manufactured in American fabs, but the surrounding supply chain remains global. Domestic production is progressing, although genuine semiconductor independence remains distant.

MY HOT TAKES

• Intel’s results are less interesting as a stock recommendation than as evidence about the direction of the AI economy. AI spending is not disappearing–it is spreading across a larger and more complicated technology stack.

• Agentic AI may revive demand for the decidedly unglamorous CPU because autonomous systems require constant fetching, checking, and coordination. The GPU performs the heavy math, but somebody still has to manage the traffic.

• Intel’s long-term customer agreements deserve a raised eyebrow. They provide valuable visibility, but sellers also lock in prices when they worry that today’s favorable conditions may not last.

• The enormous GAAP loss is not proof that Intel’s operating turnaround is imaginary. Investors should always determine whether a reported loss represents cash leaving the business or an accounting adjustment on paper.

• “Made in the USA” is a meaningful achievement, but the “mostly” matters. American fabs can reduce strategic vulnerability without eliminating dependence on international equipment, materials, packaging, and expertise.

• You can quote me: “What does it tell us when the people selling the shovels start hedging against the end of the gold rush?”

Made in the USA…mostly. There is apparently this trend on TikTok with…ehem, older folks dancing and going viral. Older–I think in this case means greater than or equal to Gen X. So, if you were born in 1980 or earlier–for the sake of this blogpost–I am talking about you, older-person. Anyway, being an older-person myself, it is nice to see the comeback. 😉 Another trend that seems to be taking hold–as of last night–older semiconductor companies announcing AI-generation growth!

Back in the 1990s, every PC on every desk on Wall Street wore a little badge of honor: "Intel Inside." Just for clarification, Intel released its first PC CPU in 1974–the 8080–putting the CPU, as we know it today, squarely in the Gen X category. Intel thrived! Then the AI gold rush arrived, the GPU became the belle of the ball, and Intel became the company your kids forgot existed. Well, last night the old-timer crashed the party, posting its fastest revenue growth since 2011. How? It turns out the AI everyone is racing to build needs more than just GPUs–it needs a conductor for the orchestra. And that conductor is made in the USA. Mostly. Let's dig in. There is a lot to learn here–please be patient.

Let's start with the numbers, because they've earned a moment of respect. Intel reported second-quarter revenue of $16.1 billion, up 25% from a year earlier, sailing past Wall Street's estimates. Adjusted earnings came in at 42 cents a share against an expected 21 cents–Intel didn't just beat the consensus, it doubled it. The engine room was the Data Center and AI segment, where revenue surged 59% to $6.3 billion, while the PC chip business chipped in a healthy 13% gain of its own. Management then raised the bar, guiding third-quarter revenue to a range of $15.8 to $16.8 billion, comfortably above what analysts had penciled in. And perhaps most tellingly, the company admitted it simply cannot keep up with data center orders. Does that sound familiar? Sit with that for a second. The company Wall Street spent the last few years writing obituaries for is now handing out deli-counter tickets. One note of sobriety before we go any further: the stock has risen more than 170% this year, which means a great deal of this “good news” is already probably baked into the price. This is not a note about whether to buy a stock. It's a note about what this stock is telling us about where the AI economy is heading.

So why is this happening now? Well, it helps to understand that AI is growing up, and like any adolescent, its appetites are changing. The first phase of the boom was all about training–teaching enormous models by feeding them, more or less, the entire recorded output of humanity. Training is a brute-force exercise, and the GPU, with its thousands of small cores crunching math in parallel, is the perfect…er, brute for the job. In that world, the CPU was a glorified stage manager. Data centers needed roughly one CPU to shepherd every eight GPUs, and that lopsided ratio explains how NVIDIA became the most valuable company on earth while Intel was left checking coats at the door.

But models don't stay in school forever. Eventually the industry stops teaching them and starts using them–that phase is called inference–and now we have entered a third phase the industry calls agentic AI, where models don't merely answer a question but go off and do multi-step work on their own. An AI agent retrieves documents, calls other software tools, checks the rules it is supposed to follow, coordinates with other agents, and reports back when the job is done. Ask yourself what all that fetching, checking, and coordinating actually amounts to. It isn't glamorous parallel mathematics. It's traffic management. And traffic management is what the humble CPU has been doing since the dawn of computing.

Intel's own CFO laid out the arithmetic on a recent earnings call, and it's worth committing to memory. In the training era, the ratio ran about one CPU for every eight GPUs. In the inference era, it tightened to one in four. In the agentic era, Intel believes it converges toward one-to-one–and in some workloads may even tip in the CPU's favor. If that math is even directionally correct, the market for server CPUs doesn't just recover–it multiplies. Right now, the demand is running so hot that Intel has been prioritizing data center chips over consumer processors just to keep the biggest customers fed. The Millers have certainly heard of NVIDIA–their grandson won't stop talking about it–but the company they had filed away with fax machines and flip phones turns out to make the one chip every AI data center suddenly can't get enough of. The industry has even coined a term for this emerging menagerie of silicon: the xPU era, in which CPUs, GPUs, ASICs, and custom chips each do what they do best, stitched together into one coordinated system. An orchestra needs more than a brass section, and somebody has to keep time. Intel's newest Xeon server chips–built on its most advanced manufacturing process, in American fabs–are its bid to hold the baton.

Now, to be fair and full disclosure, let's address the elephant sitting in the middle of the earnings release, because it may be the most educational thing in it. The very same report that announced this blowout quarter also disclosed a net loss of $11 billion under GAAP. How can both be true? GAAP, which stands for generally accepted accounting principles, is the strict rulebook that requires companies to record everything, including paper gains and losses on things they haven't sold. Non-GAAP figures strip out items management considers one-time or non-cash, so investors can see the underlying business. Neither lens is a lie; they are simply pointed at different things. In Intel's case, the culprit was a $12.5 billion mark-to-market charge on shares held in escrow as part of its CHIPS Act agreement with the U.S. government. And here is the delicious irony: that paper loss grew precisely because Intel's stock went up, making the government's escrowed stake worth more. Not one dollar of cash left the building. So the next time a scary loss headline crosses the screen, the question worth asking isn't whether the company is doomed–it's whether money actually left, or whether an accountant simply moved a number from one column to another. Half of reading Wall Street is knowing which lens the headline writer chose. 🧐

And now for the detail that made me sit up straight, because there is no free lunch hiding in even the best earnings report. Intel disclosed that it has begun writing long-term agreements with customers for its server chips–some with pricing locked in, others guaranteeing volumes. Server CPU prices have already climbed 10% to 20% since March, industry watchers expect another round of increases in the back half of the year, and big buyers are now contracting for chips years in advance, the way they have long done in the memory market. There are two honest ways to read this. The charitable reading: a company rebuilding itself should absolutely want multi-year revenue visibility, and customers desperate for supply are happy to provide it. The seasoned reading–the one from my Wall Street sayings notebook–is that sellers lock in today's prices when they want to preserve them in case tomorrow's prices are lower. That is not pessimism, it is simply what prudent sellers do when they suspect they may be standing somewhere near the top of a cycle. So here is a question worth turning over with the morning coffee: what does it tell us when the people selling the shovels start hedging against the end of the gold rush? Not a siren, mind you. Just a raised eyebrow.

Which brings us back to that tagline. Intel's newest chips are manufactured on its 18A process in American fabs, and thanks to the CHIPS Act stake, the taxpayer literally owns a piece of the outcome. Whatever one thinks of industrial policy–and my longtime followers know that I only do policy, not politics–this is what reindustrialization looks like when it actually works: advanced manufacturing capacity, on American soil, producing something the whole world suddenly wants. That is the "Made in the USA" part, and it is genuinely worth celebrating. Now for the "mostly." Intel's own CFO noted that the company is meaningfully increasing investments in equipment, clean room space, and substrates–and much of that surrounding ecosystem, from advanced packaging to specialized materials to the tools that make the tools, remains stubbornly global. A chip born in Arizona still takes a world tour before it lands in a server rack. Progress is real. Independence is not quite yet. And the gap between those two words is where supply chains, tariffs, and geopolitics will keep making themselves felt.

So where does this leave us? Not with a hot stock tip–the market has already thrown Intel quite a party, and arriving three hours late to a party rarely improves the experience. It leaves us with something more durable: evidence that the AI buildout is not fading but diversifying, and that as a technology matures, value migrates quietly from the flashiest component to the whole system around it. The winners list is getting longer and more complicated than "that GPU company," and for patient investors, that is a healthier market than the one we had. Somewhere in a data center this morning, a server rack is wearing the modern equivalent of that old badge from the 1990s, and for the first time in a long while, it means something again. Markets forget. And then, sometimes…well…they remember.

Yesterday's Markets

The Dow Jones Industrial Average fell 506 points, or 0.97%, to close at 51,711, while the S&P 500 dropped 1.21%, and the Nasdaq Composite declined 2.15%. The 10-year Treasury yield rose to 4.70%, its highest level since January 2025. Brent crude gained 7% to close at $100.69 a barrel, and WTI crude advanced 6% to settle at $92.19. Alphabet fell 7% and Tesla dropped 14% following their earnings reports.

NEXT UP

• S&P Global Flash Manufacturing PMI (July) may have ticked up to 54.4 from 53.9.

• S&P Global Flash Services PMI (July) probably climbed to 52.2 from 51.9.

• New Home Sales (June) are expected to have increased by 4.8% after sliding by -7.3% in the prior period.

• Next week: earnings announcements ramp up meaningfully. We will also get Durable Goods orders, more housing numbers, Conference Board Consumer Confidence, Personal Income, Personal Spending, GDP, and PCE Price Index. Next week will also bring us an FOMC meeting, mid-week. Next week will have something to say about where your portfolio spends the rest of its summer, so you had better show up right here for the very latest.