Intel (INTC): The $10.3B Foundry Loss That May Actually Be $1.5B, and the One Question That Still Matters

Twelve signals across seven lenses

The overall committee posture is HIGHER_SCRUTINY. Every signal below carries a single-binary conditional structure: the bull branch depends on something the bear branch specifically doubts, and most paths route through the same customer-commitment window.

Finding 1: The $10.3B headline loss is roughly $1.5B in cash

This is the single most important reframe in the analysis, and three lenses sit at different points on it. Stress Scanner uses the $10.3 billion GAAP figure as its anchor. Atomic Auditor explicitly strips out depreciation and estimates the cash operating loss at roughly $1.3 to $1.8 billion per year. Roadkill Radar sits in a wider $4 to $6 billion per year band. Both extremes are internally correct under different framings. The implication diverges enormously.

FY25 total depreciation was $10,757 million against property, plant, and equipment of approximately $105 billion, overwhelmingly concentrated in Foundry fabs. Estimating 80 to 85 percent of depreciation sitting in Foundry yields Foundry depreciation of roughly $8.5 to $9.0 billion per year. Subtract that from the $10.3 billion segment operating loss and you get a cash operating loss in the $1.3 to $1.8 billion band, manageable against $37.4 billion cash for many years.

Finding 2: Five of seven lenses converge on a single binary window

Stress Scanner, Moat Mapper, Gravy Gauge, Atomic Auditor, and Roadkill Radar independently named the 2H 2026 to 1H 2027 Intel 14A external customer commitment window as their dominant monitoring trigger. This kind of cross-lens convergence is unusual. In most equity analyses, lenses each find their own forcing functions and the synthesizer has to weight them. Intel produces a single binary: either a named commercial customer commits to 14A in the window, or the moat thesis is materially impaired.

CEO Tan has been explicit on this framework: "I do not subscribe to the belief that if you build it, they will come." 14A capacity is gated on customer commitments. Twelve months into the customer-gated era, zero publicly named commercial external customers have signed. The $222 million of Q4 2025 external foundry revenue is, per CFO Dave Zinsner's commentary, driven by U.S. government projects plus one-time Altera deconsolidation accounting. True commercial external revenue is a rounding error against the $52.9 billion consolidated base.

What the window actually looks like

• Q2 2026:First hard independent test of 18A yield progression through the P&L. A gross margin print above 36 percent validates the yield-to-margin conversion; at or below 34.5 percent suggests the yield claims are not translating.
• 2H 2026:The window opens. Tan has indicated that 14A customer engagements are active and the commitment decision timeline tracks to this period. Industry conferences through the second half typically produce directional customer commentary.
• Q4 2026 earnings:The committee's explicit threshold. Zero named commits by this print would materially impair the thesis and likely trigger the Foundry Stranding Cascade compound-risk scenario.
• 1H 2027:The window closes. Practical wafer-planning lead times for 2027-2028 custom silicon programs mean that customers who have not committed by this point are effectively committed to a competing foundry.

Finding 3: Per-share economics are permanently impaired even in the bull case

Intel issued 664 million new shares in FY25, a 15.3 percent single-year dilution on a 4,330 million share base. The $5 billion NVIDIA strategic investment, the $2 billion SoftBank investment, the U.S. Government CHIPS warrant structure, plus restricted stock unit issuance drove the count. The issuance is done. The dilution is realized and not reversible.

A less visible mechanical drag sits below the dilution line. The FY25 balance sheet carries a new $12.1 billion noncontrolling interest line attributable primarily to SoftBank, NVIDIA, and USG-linked structures sitting at subsidiary levels. In FY25, that NCI line attributed $293 million of earnings to minority holders. In FY26, management guided the attribution to grow to roughly $1.2 billion. It grows further in FY27. This is a permanent transfer of economic interest that reduces whatever per-share recovery eventually shows up on the P&L, rather than profit-sharing.

When you strip FY25 GAAP net income of $26 million through to the shareholder level, the honest figure is net income attributable to Intel common of negative $267 million after the $293 million NCI attribution. The $26 million headline is mechanically manufactured from a $5.3 billion Altera divestiture gain plus $3.3 billion of interest-and-other items.

Where the models disagreed

Seven debates across the committee. Four produced real round-0 to round-1 movement rather than rubber-stamp convergence.

Cross-lens reinforcements

Four convergence patterns where independent lenses arrived at the same finding through different evidentiary paths. Convergence raises confidence that the finding is structural rather than an artifact of a single lens's framing.

What we are watching

Bottom line