Washington’s $36B Intel Feedback Loop

Washington’s $8.9 billion stake in Intel was worth roughly $36 billion eight months later. The paper gain looked like vindication, yet it arrived before Intel had proved a broad external foundry customer base. In semiconductor industrial policy, public equity creates a peculiar loop: the stock can rise because policy improves Intel’s odds, and the policy can look successful because the stock rose.

Equity makes the state both sponsor and scorekeeper

A subsidy can be judged by the factory, equipment, workforce, or production it was meant to create. An equity position adds another score—the value of the government’s stake. The two measures can reinforce each other even when the underlying foundry problem remains unresolved.

The mark-to-market gain undercut early claims that public ownership was simply a bad trade. But it also changes the incentive structure. Washington benefits on paper when investors assign Intel better odds of securing customers and capacity utilization, while Washington can influence those odds through policy.

Intel shares rose 10.64% after President Trump said Apple had agreed to work with Intel to design and build chips in America. Intel was then more than 520% higher over the prior year. Investors were not merely pricing a possible Apple relationship. They were also pricing Intel’s place in the domestic chip strategy.

Separate reports said the administration had pushed Intel to expand local capacity and pressed prospective partners including Apple to use its fabs. The reports remain unconfirmed. They do not establish that any eventual contract was coerced, commercially unsound, or even completed. But they explain why investors can price foundry progress and continued public support into the same security. They need no conspiracy theory—only a belief that government involvement improves Intel’s odds of winning customers.

On July 17, Intel fell 5.84% during a broad chip-stock sell-off. Even a 520%-plus annual rise did not make the stock a stable policy verdict. Market price reflects expectations; it does not establish operational earning power.

A foundry customer is proof only when it can say no

Since Intel relaunched its foundry strategy in 2021, it has needed to fix manufacturing, finance new capacity, and build a genuine customer-service operation. Washington can support the first two. It cannot directly manufacture the third.

A durable external foundry must persuade customers to choose its process technology, cost structure, reliability, packaging, and support over credible alternatives. That choice requires more than a meeting, an evaluation kit, or a politically useful announcement. It requires a production commitment that survives contact with the customer’s economics.

Fortinet becoming the first named external customer for Intel 4 is real counter-evidence to the idea that Intel cannot earn outside demand. The agreement is confirmed, and no government mandate for it has been publicly documented. It is a concrete customer win. It is also one named customer on one process node—proof of possibility, not a broad customer base.

At 14A, Intel reportedly promised Apple and SpaceX a testing toolkit before any final production commitment. Supplying the toolkit marks progress because qualification has to begin somewhere. It is not a production award. The gap between those states is where technical evaluation becomes commercial trust.

Two former leaders of the CHIPS Program Office made the same diagnosis more bluntly: the government deal did not solve Intel Foundry’s core shortage of external customers. Their point was not that public support has no value. Supply-side support cannot be counted twice—first as financed capacity, then as proof that independent demand exists.

The more Washington helps create Intel’s demand, the less that demand can certify Intel’s competitiveness.

Washington can still help form useful initial demand. An early workload can qualify equipment, establish operating history, and give a buyer a second source in a concentrated market. A politically initiated conversation can also end in a rational commercial contract. The test is not whether government opened the door, but whether customers return on the merits.

A fab is only one link in the sovereignty chain

Governments often treat domestic chip sovereignty as a factory-building problem because factories are visible and ribbon-compatible. But a leading-edge fab works only when process execution, lithography equipment, advanced packaging, and qualified customers work together. Lose any one, and a large capital project becomes a very expensive partial solution.

Intel’s use of ASML’s High-NA EUV systems for some Panther Lake chips is tangible operational evidence. Each machine costs about $400 million, and using the next-generation equipment demonstrates process ambition. But Panther Lake is an Intel product. The milestone shows that Intel is investing in manufacturing capability; it does not establish that outside designers will entrust production to it.

Intel committed $20 billion to two Arizona fabs in March 2021. Its 2026 European plan is a €5 billion expansion in Leixlip after the company canceled a planned €30 billion Magdeburg factory in 2025. Leixlip is a real investment. Magdeburg shows how strategic capacity plans can shrink even after entering the industrial-policy narrative.

Even a working fab does not finish the chip. The United States has only 3% of global advanced-packaging capacity, while AI packaging has become more dependent on TSMC and its partners in Taiwan. Packaging was once treated as what happened after the important work. In heterogeneous AI systems, it increasingly determines whether that work can be assembled at all.

At TSMC, chips at 7 nanometers or smaller represented 77% of second-quarter wafer revenue. That is the incumbent scale Intel must complement or displace. Washington’s stake can improve Intel’s financing conditions and strategic relevance. It cannot bypass missing packaging capacity, process execution, or customer qualification.

Scarcity makes demand steering stronger and less legible

AI buyers already operate inside an allocation regime, which makes public demand steering unusually powerful. Even with a $200 billion U.S. expansion, Micron said it could meet only about 50% to 66% of requested supply for some key customers. TSMC’s N3 logic-wafer capacity has been described as one of the AI industry’s largest constraints.

Under those conditions, buyers have commercial reasons to diversify foundries even without political pressure. They want more supply, greater geographic resilience, and less dependence on one constrained producer. That is the strongest case for state-aligned AI industrial policy: public leverage can accelerate qualification of an alternative that private buyers already have reason to want.

Scarcity also obscures the counterfactual. If Apple or another customer chooses Intel, was the decision driven by process merit, capacity availability, supply-chain diversification, political pressure, or some combination? The contract can be commercially sound and still remain inseparable from policy.

When government directs demand toward a producer, it transfers real economic value by improving utilization and credibility. It can also weaken price discovery or buyer choice if the preferred capacity is not competitive on its own. SEMI, whose members include Micron and Samsung, warned that intervention in memory pricing or production capacity could worsen shortages. The warning concerned memory rather than Intel’s logic foundry, but the mechanism travels: steering supply in a constrained system can move the shortage rather than solve it.

Washington should use leverage to qualify an alternative, not to grant permanent preference. The intervention proves most when it creates another viable option—and least when Washington must keep supplying that option’s customers.

The scorecard must separate leverage from proof

Intel’s share price cannot carry the full weight of this policy because Intel is changing inside a market that is changing around it. Nvidia, Qualcomm, MediaTek, Arm, Google, and Amazon are contesting the data-center CPU market. Arm says its architecture already accounts for more than half of hyperscale cloud computing. Intel is trying to restore a foundry while its historic x86 demand base loses exclusivity.

Intel’s operating structure reflects the task. The company appointed a dedicated foundry leader covering advanced packaging and back-end manufacturing while retaining separate leadership for front-end technology and manufacturing. The division exposes where delivery must work across the stack, not just in the capital budget or stock chart.

Policymakers should count incremental qualified domestic capacity, process and packaging delivery, movement from evaluation to production, and customers that commit voluntarily and return. Those measures do not dismiss the federal gain. They define what it still needs to become.

If Washington’s roughly $36 billion position helps Intel build qualified capacity, and customers keep choosing that capacity after preference disappears, the paper gain will have marked a bridge to commercial power. If customers arrive only while the state keeps forming demand, the gain merely prices the policy that produced it. The $36 billion means most when Intel’s next customer remains free to say no.