Did Intel’s Execution Story Just Outrun Its Own Factories?

By the Numbers

• Revenue: $16.5B, +21% YoY, in line with consensus above the $15.8B to $16.8B guide midpoint.
• Non-GAAP EPS: $0.52, against $0.38 guidance and roughly $0.40 consensus. That is +126% YoY from $0.23.
• GAAP EPS: $0.60, down 33% YoY. It includes a roughly $1.0B mark-to-market gain on government escrowed shares, which is why GAAP sits above non-GAAP.
• Non-GAAP gross margin: 44.1%, +4.1 ppts YoY and roughly 200 bps above guidance.
• Non-GAAP operating margin: 19.6%, up from 11.2% a year ago.
• Data Center and AI (DCAI): $7.3B, +78% YoY and +17% QoQ. Segment operating income was $3.0B, a 42% margin.
• Client Computing and Physical AI Group (CCPG, formerly CCG): $8.1B, -5% YoY and -9% QoQ, below management expectations. AI PC revenue grew 35% YoY.
• Intel Foundry: $6.7B including intersegment revenue, +58% YoY and +16% QoQ. The operating loss was $1.9B, about $236M better QoQ.
• External foundry revenue: $411M, +118M QoQ, mostly from a non-recurring customer program.
• All Other: $0.7B, -30% YoY, reflecting the Altera deconsolidation.
• Cash flow: operating cash flow of $5.1B and adjusted free cash flow of $1.6B. Gross capex was $4.2B in the quarter.
• Balance sheet: roughly $54B in cash and short-term investments after a roughly $23B August equity offering. Total debt is roughly $50B.
• Q4 guide: revenue of $15.0B to $16.0B. Analysts on the call framed the $15.5B midpoint as roughly $1.8B below the Street.
• Q4 guide, margins and EPS: non-GAAP gross margin of 38%, non-GAAP EPS of $0.25, and GAAP EPS of $0.10.
• Q4 Fab 28 impact: roughly $300M of revenue and roughly 200 bps of gross margin, with most wafer scrap taken in Q4. The CFO's early assessment puts a similar revenue impact in Q1 2027, roughly $600M across the two quarters.
• Q4 segment color: DCAI up mid-teens percent QoQ, CCPG down more than 20% QoQ.
• Capex: 2026 gross capex of about $20B, with 2027 expected to rise sharply. Management does not expect positive free cash flow in 2027.

The News

Intel reported third-quarter 2026 revenue of $16.5 billion, up 21% year over year, with non-GAAP EPS of $0.52 against guidance of $0.38. Revenue was in line with consensus, while EPS and gross margin beat both guidance and the Street; management counts this as its eighth straight quarter of beating its own outlook. What moved the stock was the fourth-quarter guide, which absorbs a client inventory correction and a roughly $300 million hit from the Fab 28 shutdown in Israel and cuts non-GAAP EPS roughly in half sequentially.

Analyst Take

Read closely, on the product side this print describes two businesses sharing a ticker. One is a data center business that, by management's account, posted its strongest growth in fifteen years. It is sold out, and customers are asking for long-term supply agreements. The other is a PC business absorbing an OEM inventory flush driven by memory costs it does not control. The after-hours reaction suggests investors priced the second business.

The bear case deserves its full weight. Q4 non-GAAP EPS roughly halves. Management told investors 2027 will be an investment year without positive free cash flow. Foundry still loses close to $2 billion a quarter. The roughly $23 billion August equity raise was sized to carry Intel through that stretch, which shifts the question from solvency to execution.

Our read diverges on one point. The constraint that limited the beat sat inside Intel's own fabs and substrate supply chain, not in its order book. If that holds, server share next year looks like a manufacturing question rather than a product question, and the turnaround thesis gets tested on the factory floor.

What the Numbers Mean

The most revealing color came in Q&A, when the CFO said 2027 server market share would depend on supply rather than roadmap. Coming from a CFO, that framing deserves weight. It also shows how far the factories have already stretched. Management now expects DCAI to finish 2026 nearly $10 billion above where it expected when the year began.

The competitive scorecard for next year will be kept in Ireland, Oregon, and the Japanese substrate houses the CEO visited to press for allocation. Granite Rapids on Intel 3 is now Intel's largest server family and adds to gross margin. Yet Intel 3 remains constrained, and converting some Oregon space to Intel 3 output does not reach finished product until the second half of 2027.

Some relief may come from the node shift rather than the Intel 3 expansion. Clearwater Forest, Intel's first 18A server part, is shipping for revenue. Diamond Rapids on 18A-P is sampling for a 2027 ramp, on a node where management suggests capacity is less tight. Being sold out is a pleasant problem until it is the only problem you have.

Allocation is where the friction sits. Intel is freeing server capacity partly by repurposing Oregon 18A wafer capacity that would otherwise build client parts such as Panther Lake, and substrates for its EMIB packaging are short across the board. Today's PC correction makes that trade cheap, because OEMs are cutting orders anyway.

The harder test comes if client demand recovers in the second half of 2027, alongside the Nova Lake ramp on 18A-P and the arrival of the substrate capacity Intel has secured. At that point, PC OEMs, merchant Xeon buyers, purpose-built silicon customers, and the external packaging customers Intel expects to announce would all compete for the same wafers and substrates. Not all of them will get what they want. Our read is that cloud providers holding long-term agreements stand at the front of that line, and buyers on shorter commitments stand further back.

The gross margin bridge tells a subtler story. Management credited the Q3 lift to falling Intel 3 and 18A costs and a better mix. By its account, 18A costs are down roughly 40% since Q1. Learning on yield and cycle time is turning into dollars, which is the whole foundry thesis compressed into a single quarter.

The Q4 step-down is mostly fixed-cost absorption as client wafer starts get cut, plus scrap from Fab 28 in Kiryat Gat. The October 14 explosion there took the lives of two workers, and that human cost belongs ahead of any financial one. Fab 28 runs Intel 7, the company's 10-nanometer-class process, and was reportedly considered for closure in mid-2025. Yet nearly all of the Q4 revenue impact falls on 10-nanometer-class server products. A fab Intel reportedly weighed closing turns out to carry meaningful server supply in a sold-out market. The CFO said production should restart this week.

Purpose-built silicon, the custom ASIC business inside DCAI, involves a deliberate trade. It is Intel's fastest grower, having roughly tripled over three years, and management expects it to triple again. Its margins sit below the corporate average. Management said it will chase gross profit dollars over margin percentage. We think that is the right instinct for a manufacturer that needs to load its fabs, though it will make reported margins look noisier than the underlying economics.

Management's client commentary turned noticeably more cautious. It expects the inventory correction to run through Q1 and possibly into Q2, and it assumes no memory relief in 2027. Operators who have run PC supply chains will recognize the pattern: OEMs cut weeks of inventory faster than end demand falls, which exaggerates the downturn at the chip supplier. Wildcat Lake selling out in a shrinking market suggests the cost-optimized part arrived at the right moment. Management still argues PCs and edge systems remain central to agentic and physical AI. The edge business holding up better than PCs this quarter is the only early evidence, and 2027 will not settle the argument.

The planned acquisition of Classiq is a longer-dated signal of how Intel intends to compete in quantum. It pairs Intel's spin-qubit manufacturing with hardware-agnostic quantum software, a different go-to-market than superconducting-qubit players have chosen. Once the deal closes, Intel gains a quantum software revenue line alongside its hardware research. We would view that as an option, not a 2027 driver.

Market Analysis

Same-quarter peer prints suggest Intel's split is a sector pattern, not a company quirk. TSMC reported high-performance computing at 68% of revenue and guided to another large sequential step in Q4. AI infrastructure is absorbing leading-edge capacity. In that environment, Intel's positioning as a U.S.-domiciled advanced-node alternative should resonate, if 18A-P and 14A customer conversations convert.

Management again declined to name wafer customers. It did point to a couple of advanced-packaging customer announcements over the next two quarters. Given the substrate bottleneck, packaging may turn into revenue before external wafers do. Intel is adding packaging capacity in New Mexico and Malaysia for 2027 customer ramps.

The memory side is where the PC damage starts. SK hynix posted an 80% operating margin and expects demand to outrun supply through 2030. It has also locked long-term agreements covering more than half of its 2027 to 2028 capacity. Memory has become the apex predator of the PC bill of materials. Every dollar of that margin is a dollar a PC buyer does not spend on a better processor. Intel's assumption of no memory relief in 2027 matches what the memory makers themselves are signaling.

Texas Instruments shows the same divergence on a different part of the board. Its data center revenue nearly doubled year over year while personal electronics sat roughly flat. AI is pulling infrastructure demand upward while consumer devices stall.

The print that matters most has not happened yet. AMD reports on November 3. In its prior quarter, AMD guided to server growth above 80% in the second half. If AMD's server number lands near that, it should help show whether Intel's supply ceiling is costing it share now, or whether the whole x86 server market is simply supply-gated.

The rivalry has limits. During the quarter, Intel and AMD brought their x86 PyTorch optimizations together in oneDNN. We read that as a shared defense of the instruction set against Arm, even as the two compete for the same constrained server sockets.

Looking Ahead

The key trend we'll be monitoring is whether Intel's supply curve bends fast enough in 2027 to meet the demand it described. Management now sizes the server CPU opportunity, driven largely by agentic workloads, at up to five times today's level within five years. The estimate rests on an energy argument: agent tasks on deterministic CPUs consume joules, while GPU reasoning queries can run to watt-hours. That thesis is worth testing against Arm-based hyperscaler silicon such as Google's Axion, which makes a similar efficiency pitch. It remains a management estimate.

The nearer-term tests are concrete:

1. AMD's November 3 print will show how much share moves while Intel is constrained.
2. Intel released the 14A 0.9 PDK on schedule in October. The 1.0 PDK, targeted for Q1 2027, and the promised packaging announcements will show whether external foundry revenue can grow beyond one-off programs.
3. At CES in January, Nova Lake debuts, and management guides to roughly double Panther Lake's design wins at launch.

For CIOs, the practical implication arrives sooner. The planned multiyear collaboration with Oracle Cloud Infrastructure extends to future OCI X14 offerings built on Diamond Rapids. It looks like an early template for how cloud providers will lock in x86 supply, and management says several other customers want similar long-term agreements. Server CPU allocation is returning as a procurement reality. Buyers planning 2027 server refreshes should lock allocation in Q1 negotiations rather than count on spot availability.