After three years of "it's coming," Intel finally brought numbers instead

Here's the deal: Intel reported Q2 2026 on July 23. Revenue came in at $16.1 billion, $1.8 billion above the midpoint of its own guidance. Non-GAAP gross margin hit 41.8 percent, 280 basis points better than guided. Non-GAAP EPS was $0.42 against $0.20 guided — more than double. Shares rose over 13 percent in after-hours trading.

The process story matters more than the P&L, though. CEO Lip-Bu Tan said Intel Foundry exceeded internal volume targets on 18A and 18A-P. Q2 18A output ran roughly 25 percent above internal targets and grew more than 50 percent quarter over quarter. Panther Lake manufacturing cost came down about 50 percent year to date.

A week earlier, on July 15, ASML confirmed on its own earnings call that Intel is the first company to ship high-volume logic chips manufactured with High-NA EUV. Select 18A layers on a subset of Intel Core Ultra Series 3 processors — Panther Lake — are now dual-qualified for 0.55 NA scanners, and products are shipping to customers at yields matched to ASML's established platform. Reporting around the same date put Intel Foundry's 18A yields at roughly 85 percent, up from about 65 percent the prior quarter.

One sentence: for the first time in three years of foundry rebuild roadmaps, Intel showed shipments rather than plans. And the price of admission was a capex line that jumped right along with it.

The cast: Lip-Bu Tan's Intel, ASML, and Panther Lake

Since Lip-Bu Tan took over, Intel's story has narrowed to three tasks. Win back process leadership, turn the foundry into a business that sells to outside customers, and generate enough cash in data center to fund both. The difficulty is that none of the three moves quickly. Under prior management, Intel said some version of "the next node will be different" repeatedly, and the market discounted the claim a little further each time.

18A is the node meant to reverse that discount. It combines gate-all-around transistors (RibbonFET) with backside power delivery (PowerVia), and it is the generation that squares off directly against TSMC's N2. Panther Lake is the first high-volume consumer product built on it. Weak yields would have relegated 18A to a technology demonstration; strong yields earn Intel the right to court external customers.

ASML played referee in this story. High-NA EUV is ASML's next-generation lithography platform, priced somewhere around $350 million per tool, and until now it had been used for development rather than production. "It's actually in high-volume manufacturing" is the kind of claim that only carries weight when the tool vendor says it, which is why ASML naming Intel on its earnings call mattered. It is good news for ASML too — proving High-NA works in production is what unlocks orders from everyone else.

TSMC has been far more cautious about High-NA, arguing that existing 0.33 NA EUV plus multi-patterning gets it to N2 economically. That judgment has been correct so far. High-NA halves the exposure field, so large dies require stitching, and the tools cost nearly twice as much. Intel jumping first reads less as technical confidence than as strategic necessity. The company that's behind adopting the new tool first is a recurring pattern in semiconductor history.

The quarter in numbers

Breaking it out by segment sharpens the picture.

Metric Q2 2026 Change
Total revenue $16.1B +25% YoY
Non-GAAP gross margin 41.8% +280bp vs guidance
Non-GAAP EPS $0.42 vs $0.20 guided
Data Center & AI (DCAI) $6.3B +59% YoY
Client & Physical AI (CCPG) $8.9B +13% YoY
Intel Foundry revenue $5.8B +31% YoY
Foundry external revenue $293M —
Foundry operating loss $2.1B improved $348M QoQ

The last two rows are the ones that matter. Of $5.8 billion in foundry revenue, external customers accounted for $293 million. The rest is Intel selling to Intel. For the foundry to become a real business, that external number has to change order of magnitude. A $2.1 billion operating loss that improved by $348 million sequentially says the direction is right, but the burn rate still exceeds $8 billion annualized.

DCAI's 59 percent growth is the eye-catching line. It signals that AI server demand is pulling CPU volume along with it — host CPUs sitting next to Nvidia accelerators. Don't misread it as evidence that Intel's own accelerator business is competitive. That business remains small.

The thing the market actually reacted to came elsewhere in the call. Intel raised 2026 capital expenditure to over $20 billion and said 2027 would be "significantly above" 2026. Spending more because yields improved is coherent logic, but for investors waiting on free cash flow, it reads as another invoice.

What each party gets out of it

Intel gets negotiating leverage. What wins foundry customers is not a roadmap deck but shipment history. "Chips built on 18A are going to customers today, and High-NA-patterned layers are inside them" is a materially different sentence in a conversation with Apple or Microsoft. This is the quarter that made the sentence true for the first time.

ASML gets commercial validation for High-NA. The standing industry skepticism was always about when the tool would reach real production. With Intel as a reference, ASML can point other customers at a working case, which feeds directly into the 2027-and-beyond order pipeline. For equipment vendors, the first commercial deployment isn't marketing material — it's the basis for revenue forecasts.

The US government gets a data point too. The political case for the CHIPS Act rests on whether America regained leading-edge logic manufacturing, and 18A progress answers that question favorably. For anyone who treats the concentration of leading-edge capacity in Taiwan as a strategic risk, Intel becoming a functioning alternative carries policy value beyond its revenue.

There are losers. TSMC takes no immediate revenue hit, but for the first time in roughly two decades it faces the sentence "we were not the first to deploy the most advanced lithography." For Samsung Foundry the pressure is more direct. Samsung adopted gate-all-around first and then struggled with yields; if Intel demonstrates 85 percent-class yields on 18A, Samsung risks losing the "only alternative to TSMC" position to Intel.

Precedents: what worked and what didn't

Process leadership has changed hands before, and the pattern is instructive. The canonical successful reversal is TSMC in the mid-2010s. Intel was dominant, but TSMC produced 16nm FinFET reliably, won Apple's volume, and the gap inverted from there. The decisive factor wasn't adopting new technology first — it was stabilizing yield first. Customers buy predictability over bleeding edge.

The canonical failed reversal is Intel's own 10nm. Intel pushed 10nm with multi-patterning instead of EUV and lost years. The ambition was technically real but lost to manufacturing reality, and TSMC stabilized 7nm and widened the gap in the meantime. The failure mode wasn't overconfidence in new technology; it was betting on a single path with no fallback.

Samsung Foundry's 3nm GAA is worth holding up next to this too. Samsung introduced gate-all-around ahead of TSMC but couldn't convert it into major customers because early yields were low. Technical leadership that failed to become commercial results, in textbook form. Whether Intel's High-NA lead follows that path comes down to yield and cost, same as always.

Measured against those cases, Intel's current position looks cautiously positive. Moving 18A yields from roughly 65 to 85 percent is a different trajectory than early Samsung 3nm, and halving Panther Lake cost in six months indicates the manufacturing learning curve is behaving normally. What remains unproven is external volume, and $293 million says this story is still in the demonstration phase.

How the competition plays this

TSMC will most likely respond with quiet indifference. N2 is already in production, an additional $100 billion is committed to Arizona, and above all TSMC controls the advanced packaging bottleneck — CoWoS keeps its leverage in the AI accelerator market intact regardless of lithography generation. Most observers expect TSMC to adopt High-NA around the A14 generation, and until then the argument is simply "we reach the same result for less."

Samsung is under more urgency. How much of the current wave of US big-tech contracts Samsung Foundry captures is the open question, and Intel establishing itself as the alternative destabilizes Samsung's number-two slot specifically. Samsung's emphasis on long-term HBM supply at the Korea-US AI summit in Silicon Valley reads sensibly in that light: consolidate an unambiguous lead in memory, buy time in foundry.

China's response runs on a separate track. SMIC has no EUV access at all and cannot participate in the High-NA race, so resources are being funneled toward mature nodes and a domestic GPU ecosystem — the same week's confidential Hong Kong IPO filing from MetaX is part of that program. Intel's progress paradoxically intensifies the pressure on Beijing to go it alone.

The market itself is the wild card. The Philadelphia Semiconductor Index entered a bear market on July 17, down more than 20 percent from its late-June peak, triggered when Moonshot AI's open-weight Kimi K3 undermined the premise that expensive compute would stay scarce. Intel's beat is being cited as counter-evidence, but one company's quarter rarely reverses a sector-level narrative.

So what actually changes

If you design chips, the practical change is that there are now three leading-edge foundry options rather than one and a half. Leading-edge logic has effectively meant TSMC, with Samsung as a conditional alternative. If Intel's 18A holds up on yield and shipments, you can genuinely get three quotes. That is not a small shift in leverage. Actually migrating still depends on PDK maturity and IP ecosystem, where TSMC remains far ahead.

For investors, three numbers to track. First, foundry external revenue — whether $293 million grows materially next quarter determines whether the narrative is real. Second, free cash flow against capex; over $20 billion in 2026 and "significantly above" that in 2027 absorbs a large share of the earnings improvement. Third, the pace of foundry loss reduction, because a sustained $350 million per quarter of improvement lets you back out a breakeven date.

For ordinary users, the visible effect is laptops. Panther Lake cost falling by half in six months means either cheaper machines at the same performance or better silicon at the same price. Expect that to surface as Core Ultra Series 3 designs proliferate through the second half of the year.

At the industry level, the meaning of this quarter is a modest reduction in single-supply-chain risk. Leading-edge logic concentrated in one island has been a widely flagged geopolitical vulnerability, and Intel becoming a working alternative shrinks it. There is a long way to go, but this is the first quarter in three years that produced evidence the direction is correct.

🥄 Three Things You're Probably Wondering

— Has Intel caught TSMC? No. Leading on a lithography generation and winning the foundry business are different problems, and $293 million in external revenue is not comparable to TSMC at any scale. What Intel has done is return to the category of companies that plausibly could — which, after three years of not managing that, is a real change.

— Does being first on High-NA actually matter? Getting yield out of it matters more than having it. Intel's 10nm failure and Samsung's 3nm both featured fast technology adoption that collapsed on yield. The difference this time is ASML confirming yields matched to its established platform. Whether High-NA wins on cost per wafer is still genuinely contested.

— How should I read chip stocks right now? The sector entered a bear market on July 17, triggered by an open-weight model undermining the compute-scarcity premise. Individual fundamentals, Intel's quarter included, remain solid — so index and companies are moving separately. Which one is right won't be clear until we see two more quarters of capex follow-through.

Sources

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!