Intel Foundry Bleeds Three Billion Dollars in Fiscal Q3
Intel remains a company split by its own ambitions. The Q3 2025 earnings report, released yesterday, confirms the internal product group is subsidizing a massive, capital intensive bet on external manufacturing. Total revenue hit 13.1 billion dollars, a figure that sits 200 million dollars below the consensus estimate on Wall Street. The stock price reacted predictably, dropping 4.2 percent in after hours trading to 31.42 dollars.
The most alarming data point lies within the Intel Foundry segment. This division reported an operating loss of 3.2 billion dollars for the quarter. This is an expansion from the 2.8 billion dollar loss reported in the same period last year. Per the latest SEC filings, Intel is struggling with the high startup costs of High-NA EUV lithography machines. Each machine costs roughly 380 million dollars. Intel now owns five of them, but they are not yet generating significant external revenue. The utilization rates at the Oregon and Arizona fabs remain stuck below 65 percent, a level that makes profitability impossible in the semiconductor world.
Segment Performance Breakdown Q3 2025
| Segment | Revenue (Billions) | Operating Margin | Year-over-Year Change |
|---|---|---|---|
| Client Computing (CCG) | $7.2 | 28% | -2% |
| Data Center & AI (DCAI) | $3.9 | 11% | -8% |
| Intel Foundry | $4.4 | -73% | -12% |
| Altera / Mobileye | $1.1 | 4% | -15% |
The 18A Technical Gambit and the PowerVia Hurdle
Intel CEO Pat Gelsinger has staked the entire company on the 18A process node. This node is designed to leapfrog TSMC by introducing RibbonFET (Gate-All-Around) and PowerVia (Backside Power Delivery) simultaneously. While Intel claims 18A is healthy, independent reports suggest that defect densities are still too high for high volume mobile chips. According to Bloomberg market data, major potential customers like Apple and Qualcomm have not yet committed to 18A for their 2026 flagship products, preferring the proven stability of TSMC N2.
The technical mechanism of PowerVia is the core differentiator. By moving power delivery to the back of the wafer, Intel reduces IR drop (voltage sag) and allows for higher clock speeds. However, this adds significant complexity to the manufacturing process, specifically in wafer thinning and through-silicon via alignment. If Intel cannot achieve a yield of at least 60 percent on 18A by the end of this year, the foundry business will likely require another massive capital infusion or a government bridge loan beyond the existing CHIPS Act allocations.
Client Computing and the AI PC Saturation
Intel’s primary cash cow, the Client Computing Group, is under siege. While the company shipped 25 million Core Ultra processors this year, the average selling price (ASP) is falling. PC manufacturers like Dell and HP are leveraging Intel against AMD’s Ryzen 9000 series to drive down component costs. The AI PC narrative has failed to trigger the massive refresh cycle Intel promised in 2024. Consumers are finding that local NPU (Neural Processing Unit) tasks do not yet justify a 1,200 dollar hardware upgrade.
Furthermore, the data center segment is a graveyard for Intel’s margins. Reuters reports that AMD’s EPYC Turin chips have captured 34 percent of the x86 server market as of October 2025. Intel’s Granite Rapids is a competitive product, but it arrived twelve months too late to stop the momentum shift in enterprise cloud spending. Hyperscalers like Amazon and Google are increasingly moving workloads to their own ARM-based silicon, further shrinking the total addressable market for Intel’s Xeon lineup.
The Nvidia Shadow and Gaudi 3 Failure
In the artificial intelligence space, Intel is functionally invisible. The Gaudi 3 accelerator was marketed as a cost-effective alternative to Nvidia’s H200, but software compatibility issues remain a barrier to entry. While Nvidia reports 80 percent gross margins on its Blackwell architecture, Intel is forced to discount Gaudi 3 by as much as 50 percent to secure even minor contracts with secondary tier cloud providers. The lack of a unified software stack comparable to Nvidia’s CUDA means that developers are unwilling to port their LLM training workloads to Intel hardware.
Intel’s internal product roadmap now relies heavily on Lunar Lake and Arrow Lake to stabilize the ship. These chips utilize external TSMC tiles for the compute logic, a humiliating admission that Intel’s own fabs are not yet capable of producing its most advanced designs efficiently. This outsourcing strategy protects market share but decimates gross margins, as Intel must pay TSMC a premium for the 3nm capacity. The company is effectively paying its chief competitor to keep its own product line alive.
The next critical milestone occurs on January 15, 2026, with the scheduled volume release of Clearwater Forest. This will be the first major server product built entirely on the 18A node using the final version of the RibbonFET architecture. If Clearwater Forest hits its performance targets without the stability issues that plagued the 13th and 14th generation desktop chips, Intel may finally find a floor for its declining server market share. Investors should watch the 18A defect density reports coming out of the D1X fab in Hillsboro over the next ninety days.
