Intel's $21 Billion Foundry Loss Is Eating Its AI Profits Alive

The Valuation Gap

Intel’s stock has surged 278% over the past year. Its price-to-sales ratio now sits at roughly 8.2× — a level that ranks in the top decile of the company’s own history over the last decade. That kind of multiple does not come cheap. Investors are paying for a future in which Intel’s fabs stop burning cash and start generating returns that justify the valuation.

The problem is that the present tells a different story. In the second quarter of 2026, Intel’s foundry division — the arm tasked with building chips for other companies and competing directly with TSMC and Samsung — reported an operating loss of $2.1 billion. That figure consumed almost all of the $2.5 billion in operating profit generated by Intel’s data center and AI segment in the same quarter. The AI boom that is lifting Nvidia and sustaining Intel’s revenue growth is, paradoxically, subsidizing the very business that could define the company’s next decade — or destroy it.

Overall operating margin has collapsed to 7.6% over the trailing twelve months, down from roughly 34% at the company’s peak. Revenue grew 7.5% year over year to $570 billion. Growth is real. Profitability is not.

Who Pays, Who Profit

The foundry loss deserves scrutiny beyond the headline number. External customers — companies other than Intel itself buying chips from Intel’s fabs — contributed only $293 million to foundry revenue in Q2. That is roughly 5% of the segment’s total sales. The remaining 95% consists of internal wafer transfers, where Intel’s fabless divisions buy wafers from the foundry arm, and the loss is simply absorbed by the parent company.

This is not a business model that markets will reward at an 8.2× multiple. It is a cost center with a branding exercise.

To be fair, the trend within the foundry division is moving in the right direction. Losses improved by $348 million quarter over quarter. Wafer costs are declining as yield rates climb and cycle times shorten across the advanced node portfolio. Intel’s management reports that the cost of Panther Lake — its next-generation client CPU — has dropped approximately 50% since the start of the year, with another 20% reduction expected by end of 2026. CFO Kristen Holum has noted that these new products still trail the company-wide average margin in early lifecycle stages, but their share of total product mix is growing meaningfully.

The margin trajectory matters. But trajectories are not yet results.

The Capex Mountain

This is where the stakes become existential. Intel’s capital expenditure for 2026 has been raised to over $20 billion, with management signaling a significant increase for 2027. The company currently holds approximately $30 billion in cash and short-term investments, backed by a $10 billion revolving credit facility. That leaves a structural funding gap that management acknowledges could require raising capital in public or private markets — even as it points to roughly $10 billion in non-core assets that remain convertible to cash.

The question is not whether Intel can fund the buildout. It is whether the revenue generated by those fabs — especially from external foundry customers — will arrive fast enough to service the debt and satisfy shareholders who have already priced in success.

The Japan Lens

Forbes JAPAN’s coverage of this story carries a dimension that English-language wire desks have largely overlooked: the Japanese market watches Intel’s foundry bets with particular intensity because they sit at the intersection of three structural shifts. First, Japan’s Rapidus project — a state-backed effort to build advanced logic fabs with Intel’s technology licensing — depends on Intel proving that its own foundry model can work at scale. If Intel’s fabs remain unprofitable, the technical and financial rationale for Rapidus weakens considerably.

Second, Japanese investors in Intel stock face a currency hedge they do not get with TSMC or Samsung. The yen has strengthened meaningfully in 2025 and 2026, which compresses the dollar-denominated losses when converted back to yen — but also means any future profitability from the foundry business will be magnified for yen-based holders. This asymmetry is not reflected in current pricing.

Third, Japan’s semiconductor policy community views Intel’s turnaround as a proxy for the viability of allied foundry capacity outside Taiwan. Washington’s CHIPS Act incentives, which benefit Intel’s U.S. expansion plans, were partly justified on the grounds that a profitable Intel foundry would diversify global supply away from Taiwan-dependent production. If Intel cannot make its fabs profitable domestically, the policy architecture collapses with it.

What Changes the Story

The market is not asking Intel to eliminate foundry losses tomorrow. It is asking for a convergence: that losses shrink faster than the capex budget expands. That is the single metric that will re-rate this stock.

Consider the math. At the current pace, foundry losses are declining by roughly $350 million per quarter. Capex is running at over $5 billion per quarter annualized. The losses are not closing on the spending. They need to accelerate — significantly — or the valuation multiple becomes a trap for anyone buying at today’s levels.

The AI datacenter business provides a buffer. It generated 70% of Q2 revenue and counted for the bulk of the company’s operating profit. But dependency on a single high-margin segment to subsidize a money-losing division is a fragile architecture. Any slowdown in AI infrastructure spending — and cycles of that nature are predictable in semiconductor equipment markets — would expose the gap immediately.

Who Wins, Who Loses

If Intel’s foundry business reaches breakeven within 18 to 24 months, the current valuation becomes conservative. External foundry customers would begin to materialize beyond the current 5% of revenue, government subsidies from the CHIPS Act would reduce effective capex, and the company would re-enter the geopolitical narrative as a credible second source for advanced nodes — something both Washington and Tokyo have been desperate to see.

If the losses persist beyond 2027, the math forces a painful choice: further dilution through equity raises, asset sales that erode the company’s strategic options, or a strategic pivot that abandons the full foundry ambition in favor of a hybrid model. Each outcome carries reputational and market consequences.

The current stock price assumes the first path. The quarterly loss figures suggest it has not yet begun.

For Japanese observers and global investors alike, the coming three quarters will determine whether Intel’s foundry gamble is a turning point or a prolonged bleed — and whether the AI profits that sustain it today can survive the costs they are currently funding tomorrow.