Intel Foundry booked $5.8 billion of revenue in the second quarter. Two hundred ninety-three million of it came from customers other than Intel. That single ratio — ninety-five percent internal transfer pricing — is the argument the entire half-trillion-dollar market capitalization is resting on, and it barely moved this quarter.
Everything else in the print was extraordinary. Revenue of $16.1 billion arrived $1.8 billion above the guidance midpoint and roughly $1.7 billion above consensus, growing 25% year over year, the fastest quarter in more than fifteen years and the seventh consecutive beat against management’s own bar. Non-GAAP EPS of $0.42 doubled both the $0.20 guide and the $0.21 street number. Third quarter revenue is guided to $15.8 to $16.8 billion against a $15.1 billion consensus, with gross margin at 42% and EPS at $0.38. Data Center and AI delivered $6.3 billion, up 59% year over year and 24% sequentially, at a 40% operating margin — the strongest server growth in the company’s history. Client and Physical AI delivered $8.9 billion at 26% margins. Businesses Intel classifies as AI-driven grew over 70% and now constitute roughly 70% of total revenue.
The temptation is to read this as the turnaround finally compounding. The more accurate reading is that Intel is being paid, enormously, for owning capacity in a world that has run out of it. Management described the industry as facing one of the most severe supply constraints in its history — leading-edge logic, silicon wafers, memory, substrates — and said the shortages persist for the foreseeable future. Zinsner said wafer demand continues to outstrip growing supply and that relief is back-half loaded into Q3 and Q4. In that environment a merchant CPU vendor with owned fabs does not need to win a design competition. It needs only to have silicon when nobody else does.
This is where the moat question gets uncomfortable, because a shortage is not a moat. It is a market condition, and market conditions revert. Intel’s genuinely defensible assets are narrower and older than the growth rate suggests: the x86 instruction set and the enterprise software estate compiled against it, an installed base of hyperscaler and enterprise fleets where CPU replacement is a refresh decision rather than a re-architecture, and — increasingly the interesting one — advanced packaging capacity in EMIB-T that has a growing backlog and is scarce independent of who fabricates the die. The x86 franchise is a real moat and it is under real assault; Arm-based server silicon and AMD’s share trajectory did not stop existing because Intel had a good quarter. What the AI inference build-out has done is temporarily widen a moat that had been narrowing for a decade, by raising general-purpose CPU density per rack as agentic and multi-agent workloads proliferate around accelerators. Citi is underwriting 47% CPU share by 2030 on that logic. It is a defensible view. It is also a view about a cyclical tailwind being permanent.
The foundry is a different question entirely, and it is the one that determines whether Intel is a $60 stock or a $200 stock. Foundry moats are built from process density leadership, yield, packaging, and — most durably — customer switching costs, because porting a design to a new PDK is a multi-year commitment nobody makes twice. Intel has made real progress on the first three. 18A output exceeded internal targets by roughly 25% in the quarter and rose more than 50% sequentially, with yields reported around 85% against roughly 90% at TSMC’s N2. Panther Lake cost is down about 50% year to date. 18A-P entered risk production. 14A PDK 0.5 is complete, PDK 0.9 arrives in October, and defect density is tracking ahead of schedule. ASML confirmed Intel as the first company shipping high-volume logic on High-NA EUV. On the fourth pillar, switching costs, Intel has essentially nothing yet, because it has essentially no external customers. Two hundred ninety-three million dollars is roughly $1.2 billion annualized. TSMC’s moat is not its transistors, it is that four thousand design teams have already paid the cost of learning its flow. Intel cannot buy that with capex.
Which makes the capital plan the sharpest tension in the quarter. Operating cash flow was $7.0 billion. Adjusted free cash flow was negative $8.4 billion. Cash and short-term investments sit near $30 billion against roughly $45 billion of debt, with a $10 billion revolver and perhaps $10 billion of non-core assets in reserve. Against that base, 2026 capex was lifted to over $20 billion from $18 billion, tooling is up roughly 40% year over year, and 2027 is guided “significantly above” 2026 — Zinsner noting that U.S. tools-and-space spending from 2021 through 2026 approaches $100 billion, more than any other semiconductor company over the window. Customer prepays and long-term agreements are funding part of it, AMIC credits offset roughly thirty-five cents on the dollar with a lag, and Zinsner said plainly that Intel may tap capital markets if the ramp goes well. Read that carefully: on this balance sheet, success is the scenario that requires more capital, not less. Foundry minority partners take approximately $250 million per quarter in the second half and roughly $1.1 billion across 2027 and 2028 before common holders see anything.
The $11 billion GAAP net loss reported alongside $2.2 billion of non-GAAP net income is the strangest line item in large-cap equities right now, and it deserves to be understood rather than dismissed as noise, because it recurs. A $12,529 million mark-to-market charge on Escrowed Shares from the CHIPS Act Secure Enclave agreement drove it. The government bought 433.3 million shares at $20.47 in August 2025 for $8.9 billion; 158.7 million sit in escrow, released as Secure Enclave funds disburse, and that obligation is carried as a derivative liability remeasured at the prevailing share price every quarter. Intel’s stock roughly tripled. Intel therefore books a loss. The taxpayer position is now worth north of $45 billion against an $8.9 billion cost, and there is an unexercised warrant for another 5% at $20 should Intel ever fall below 51% ownership of the foundry. The mechanical consequence for shareholders is that GAAP earnings will look worse precisely when the equity performs best, and the share count keeps climbing — 5.03 billion shares outstanding, up about 9.5% year over year, with escrow releases still ahead of it.
Margins carry a subtler warning. The 41.8% non-GAAP gross margin beat guidance by 280 basis points on factory yield and cycle time rather than price. But Q3 is guided flat at 42% even though Q2’s inventory write-downs do not repeat, which means underlying mix is absorbing two to three points of headwind. The cause is Panther Lake: early-life 18A product runs below the corporate average, and it ramps directly into the guide. Foundry lost $2.1 billion at the operating line, improving $348 million sequentially. Straight-line that and breakeven lands around 2028, which is also when 14A reaches high-volume manufacturing. The two dates are not a coincidence; they are the same bet.
On the stock, the setup is more interesting than the beat. INTC closed Thursday at $100.23, down 2.33% on the day, then traded to roughly $112 after hours with a print high of $113.55 — a 12% move against options pricing that had implied 12.5% to 15%. The 52-week range is $18.97 to $142.35. The stock is up roughly 180% year to date and roughly 320% over twelve months, one of the best large-cap performances of the cycle, and market capitalization sat near $504 billion at the close. The detail that matters: shares had already fallen roughly 26% from the June high going into this print, on AI-capex sentiment unwinding and profit taking rather than anything Intel did. The Q1 surprise of comparable magnitude produced a 23.6% day-of move. This one produced half that. The market is beginning to discount beats on this name, which is what happens when a stock has already priced several years of execution.
Street positioning reflects the confusion rather than resolving it. Consensus is a Hold at an average target between $108 and $115, but the dispersion is the actual information: HSBC’s Frank Lee at $200, KeyBanc’s John Vinh at $155, Citi’s Atif Malik at $130, UBS’s Timothy Arcuri at $121 Neutral, Susquehanna’s Christopher Rolland at $115 Neutral, TD Cowen at $115 Hold, Morgan Stanley’s Joseph Moore at $75 Equal-Weight, Rosenblatt’s Kevin Cassidy at $65 Sell, with a $45 low on the tape. A forty-five to two-hundred range is not a valuation disagreement. It is two incompatible descriptions of what the company is.
The base case sits at $105 to $130. First three quarters of non-GAAP EPS now track to roughly $1.09, putting fiscal 2026 near $1.50 against the $1.11 consensus standing into the print, with 2027 revisions likely landing between $2.50 and $3.00 — call it 35 to 42 times forward. The catalysts are the October PDK 0.9 milestone and Q4 supply relief converting visible backlog into revenue. The bull case at $150 to $200 requires external foundry revenue to inflect off $293 million with a named top-five fabless customer attached to 14A, plus the ASIC and design services line holding its 3x growth off a run rate approaching $2 billion; HSBC’s target explicitly underwrites Intel as a credible second source to TSMC, which nothing in this quarter proves but the yield trajectory makes arguable. The bear case at $60 to $80 is not a company-specific break at all — it is cohort derating. Intel lost a quarter of its market value in the month before printing its best quarter in fifteen years, which tells you the multiple is being set by AI infrastructure sentiment rather than by Intel’s execution. Compress toward 25 times the 2027 number and the stock is $65, exactly where the two bears already sit. The company-specific accelerant would be a second half PC market worse than the low-double-digit unit decline management guided, driven by memory pricing Intel does not control and cannot hedge — though hiring Seok-hee Lee, formerly of SK hynix, and opening collaboration with three major memory vendors suggests they know precisely how exposed that flank is.
One under-discussed read-through: Applied Materials, KLA, and Lam Research all traded up more than 2% after hours on the capex raise. Those businesses monetize Intel’s spending without carrying Intel’s foundry operating loss, its expanding share count, or its escrow derivative. If the thesis is that Intel’s capital plan is real, the equipment names express it with less structural drag.
For the equity itself, demand is not the disqualifying risk. Demand is visibly ahead of supply and server unit growth is guided double-digit through 2028. The risk is that Intel spends over $20 billion this year and significantly more next year building capacity whose external buyer has not signed. Watch one line when Q3 prints. External foundry revenue was $293 million. If it is not materially higher by the time 14A PDK 0.9 ships in October, then the half-trillion-dollar valuation is being paid for a CPU shortage rather than a foundry, and shortages end.