Intel leans into the AI upturn – but supply ceilings dull the shine

A cyclical recovery meets structural constraint

Intel closed 2025 with the air of a company finally catching the AI updraft—only to find its own fabs acting as the governor on growth.

Revenue for the fourth quarter came in at $13.7 billion, at the top of guidance and marking a fifth consecutive beat. Non-GAAP gross margin of 37.9% was roughly 140 basis points ahead of management’s October forecast, helped by higher volumes and lower inventory reserves. Earnings per share of $0.15 nearly doubled the $0.08 guided.

For the full year, revenue was $52.9 billion, “down slightly” year-on-year – a notable underperformance relative to the AI-heavy peers benefiting from the same demand tide. Management was explicit about the culprit: constraints in Intel’s own manufacturing network and at external suppliers that pinched growth, particularly in the second half.

The operational clean-up is more visible on the income statement than the top line. Non-GAAP gross margin ticked up to 36.7% in 2025, 70 basis points higher than 2024, while non-GAAP EPS rose by $0.55 to $0.42. The swing was driven less by pricing power and more by classic cost surgery: operating expenses fell 15% to $16.5 billion as the new management team simplified structures and “rightsized” the organisation.

Cash flow tells the same story of a balance sheet being fortified for a long campaign. Operating cash flow reached $9.7 billion for the year; hefty gross capex of $17.7 billion, offset by $6.5 billion of external funding, left adjusted free cash flow at negative $1.6 billion. But the trajectory improved sharply: $3.1 billion of adjusted free cash emerged in the second half alone.

By year-end, Intel was sitting on $37.4 billion of cash and short-term investments, bolstered by the partial monetisation of Mobileye, the Altera stake sale to Silver Lake, accelerated U.S. government funding and strategic injections from SoftBank and NVIDIA. Debt was trimmed by $3.7 billion. In a capital-intensive, geopolitically tinged business, Intel is arming itself with liquidity and partners.

PCs: AI rebrands a mature franchise

The Client Computing Group (CCG) remains the economic anchor, even as growth shifts elsewhere. Q4 CCG revenue was $8.2 billion, down 4% sequentially, as the company deliberately ceded low-end volume to feed higher-margin markets. Yet beneath the headline decline, the AI PC narrative is taking form.

Intel estimates 2025 client CPU consumption surpassed 290 million units, the second consecutive year of recovery from the post-Covid trough and the fastest growth since 2021. Within that market, “AI PC” units grew 16% in Q4.

The more important signpost for investors is execution on Intel 18A silicon in client. Management had promised one Core Ultra Series 3 (“Panther Lake”) SKU shipping on 18A by year-end; it delivered three. At CES, the new lineup underpinned over 200 notebook designs, with battery life claims up to 27 hours, a 70% generational graphics uplift and benchmark gains of 50–100% versus peers, according to the company.

These claims, if borne out in independent testing, matter less for one product cycle than for the credibility of Intel’s once-ambitious, now more sober, process road map. 18A is the first proving ground for the “new Intel” manufacturing story – and it is being tested in the most brutally price-sensitive of segments.

For now, however, client is also the shock absorber for Intel’s supply shortages. Management is actively tilting scarce internal wafer capacity toward data center CPUs, supplementing PC needs with an increased mix of outsourced wafers and focusing client output on mid- and high-end SKUs. Low-end PCs are a clear lower priority.

The trade-off is visible in near-term guidance: Q1 2026 will see a “more pronounced” decline in CCG revenue than in data center. Rising DRAM, NAND and substrate prices—symptoms of the industry-wide AI build-out—are another potential headwind for PC demand and margins.

Data center: CPUs reassert their role in AI infrastructure

If 2023–24 were about GPUs, Intel’s 2025 narrative is about the quiet resurgence of the CPU in AI infrastructure – and its own attempt to exploit that.

Data Center and AI (DCAI) revenue reached $4.7 billion in Q4, up 15% sequentially and described by the CFO as the fastest sequential growth “this decade.” Management noted the figure would have been “meaningfully higher” absent supply constraints. The demand is coming from both hyperscale and enterprise customers, as power-efficient CPUs drive a refresh cycle across traditional servers and sit at the orchestration heart of AI systems.

The broader framing is that AI usage is shifting from sporadic, human-prompted queries to “persistent and recursive” machine-to-machine interactions. That, Intel argues, elevates the CPU’s role in coordinating and securing data and model flows, rather than relegating it to a mere host processor behind the GPU.

The company’s ASIC business, quietly nested under this umbrella, is beginning to show scaling characteristics. Custom ASIC revenue grew more than 50% in 2025, 26% sequentially in Q4, reaching an annualized run-rate above $1 billion. With networking at the core of AI infrastructure build-outs, Intel is positioning design services, IP libraries and manufacturing as a vertically integrated answer to a $100 billion addressable market.

The road map, however, is still being rebuilt in flight. Under new DCAI leadership, Intel is simplifying its Xeon portfolio and concentrating resources on a 16-channel Diamond Rapids and on accelerating Coral Rapids, where symmetric multithreading returns to the server line-up. The company is also collaborating with NVIDIA on a custom Xeon integrated with NVLink, signalling a tactical willingness to be a CPU arms supplier even in GPU-dominated racks.

Timing here matters enormously for market share. Management acknowledges that Q1 will be the supply trough, with data center and client both down sequentially but server falling less sharply as capacity is redirected. Internal inventories that flattered Q3 and Q4 shipments have been drawn down to roughly 40% of peak. From Q2 onwards, wafer starts on Intel 7, Intel 3 and 18A are slated to increase each quarter, but 2026 will be a year of catching up rather than running ahead.

Investors will need to judge whether Intel’s argument—that this portion of the AI wave is “largely an x86 phenomenon”—can hold against both its main x86 rival and ARM-based server incursions, once supply no longer obscures the signal.

Foundry: technology milestones ahead of economics

If client PCs are the proof point and data center the revenue engine, Intel Foundry is the wager on strategic relevance.

Fourth-quarter foundry revenue was $4.5 billion, up 6.4% sequentially on a higher mix of EUV wafers. EUV wafer revenue has grown from less than 1% of output in 2023 to over 10% in 2025—a rapid pivot toward leading-edge manufacturing. External foundry revenue was a modest $222 million, largely U.S. government projects and the effect of Altera’s deconsolidation.

The P&L, however, is still deep in the red. Q4 operating loss for Intel Foundry widened to $2.5 billion, $188 million worse than Q3, primarily due to the early ramp of 18A. This is the economic reality of a node being qualified for revenue before it reaches mature yields.

The technical markers are more encouraging than the near-term margins. With the launch of Core Ultra Series 3, Intel claims to be the only manufacturer shipping gate-all-around transistors with backside power into revenue production, from fabs in Oregon and Arizona. 18A’s derivative, 18AP, is progressing with a v1.0 PDK in customer hands.

More strategically significant is 14A. Intel has released a 0.5 PDK and is in active engineering engagements with “a couple of key customers” who are already considering test chips and early product planning. Management is explicit that it will not deploy 14A capacity spend beyond technology development until it has firm customer commitments. In a capital cycle where competitors are underwriting multi-year fabs on spec, that restraint is notable.

High-NA EUV will feature in 14A, with multiple variants planned. Risk production is currently expected in late 2027, with volume ramps in 2028—deliberately aligned with the timing of the “leading foundry.”

The more immediate commercial wedge may be advanced packaging rather than wafers. EMIB and EMIB-T are highlighted as differentiated technologies, with management saying several customers are willing to prepay for scarce substrate capacity and that many advanced packaging deals are now measured “well north of $1 billion” each, not merely in the hundreds of millions they had previously expected.

For equity holders, the foundry story remains long-dated: substantive external 14A wafer revenue lies beyond the visible horizon, while losses are here today. What has changed in 2025 is that technology execution on 18A has finally moved from slideware to shipping products, giving Intel a firmer argument that it can meet both its own CPU road map and prospective foundry customers on credible process nodes.

Near-term guidance: a disciplined squeeze through the bottleneck

The paradox of Intel’s set-up entering 2026 is that the demand picture is the cleanest it has looked in years, even as the company braces investors for a weak first quarter.

Q1 revenue is guided to $11.7–$12.7 billion, with the $12.2 billion midpoint at the low end of seasonal patterns. Non-GAAP gross margin is expected to retreat to around 34.5%, with breakeven EPS. The margin compression stems from both lower revenue over a largely fixed-cost base and a higher mix of still-dilutive 18A products as Panther Lake ramps.

Management characterizes Q1 as the nadir of internal supply constraints: buffer inventory is depleted, wafer mix was shifted aggressively toward server in Q3/Q4, and those dies will only exit fab in late Q1. From Q2 onwards, they expect sequential supply improvements each quarter as yields and throughput improve and additional tools are installed.

Operating expenses for 2026 are targeted at $16 billion, continuing the cost discipline theme. Capex, previously flagged as down year-on-year, is now guided to a range of “flat to down slightly,” with more of the spend weighted to the first half and skewed toward tools rather than new space. The company still expects positive adjusted free cash flow for 2026 and plans to retire all $2.5 billion of debt maturities this year.

In effect, Intel is attempting a delicate threading: investing enough to relieve capacity bottlenecks in the near term and secure longer-term AI and foundry relevance, while keeping a hard line on 14A capacity until customers sign up and maintaining free cash flow discipline.

For investors, the numbers in 2025 show a company with improving operational grip and a fortified balance sheet, but one whose growth is capped more by its own fabs than by end-market appetite. The crucial question over the next year will be whether the incremental wafers that emerge from those fabs translate into renewed share gains in data center and a credible foothold as a second-source foundry, or merely allow Intel to participate in the AI cycle at a safe, but unremarkable, distance behind the leaders.