Can Intel's foundry strategy succeed, or is the US already too late to reclaim chip manufacturing
Can America Get Back Into the Chip Business? It's Complicated.
Based on analysis of a 105-node, 330-edge knowledge graph mapping the causal structure of Intel’s foundry strategy and US semiconductor manufacturing.
What We’re Actually Talking About
Computer chips — the tiny pieces of silicon inside your phone, laptop, and car — are almost entirely made in Taiwan and South Korea. The United States used to make them. Now it mostly designs them and has other countries build them.
Intel is one of the last American companies that still makes its own chips. It wants to go further and become a “foundry” — a factory that makes chips for other companies the way TSMC (a Taiwanese company) does. The US government wants this to happen too, badly enough that it passed a law called the CHIPS Act and put billions of dollars toward making it real.
The question this analysis tries to answer: is that actually going to work?
A knowledge graph — think of it like a map where the dots are ideas and the lines between them show how one thing causes or blocks another — was built to trace all the forces pushing this outcome in different directions. Here is what that map shows, in plain terms.
The Central Problem: The Chicken-and-Egg Trap
At the heart of everything is what the graph calls the “Yield-Volume Paradox.” This sounds technical, but the idea is simple.
To make chips reliably, you need to make a lot of them. Every time you run a factory, you learn something. You fix tiny problems. You get better. That accumulated experience is called “yield” — the percentage of chips that come out working instead of broken. TSMC has been doing this for decades and their factories work extremely well.
Intel’s new factory process (called 18A) is unproven. To get customers, Intel needs to show it works. But to prove it works, Intel needs customers putting real chips through the factory so it can learn and improve. Customers won’t commit until it’s proven. Intel can’t prove it without customers committing.
This is a classic chicken-and-egg problem. The graph identifies it as the single most important bottleneck in the entire system. Almost every other problem either feeds into it or flows out of it.
The Big Bet and Why It’s Still Up in the Air
The graph has a node for something called “Intel Foundry National Champion Bet.” Think of this as the core question: does Intel become America’s official chip factory, backed by the government, trusted by big tech companies, and competitive with TSMC?
Here is what makes this interesting: that node has more connections than almost anything else in the graph — 50 of them — but its weight (a measure of how settled or certain the outcome is) sits at only 5.6 out of 10. Most nodes that connected to everything else would be weighted much higher, because high connectivity usually means something is central and established.
The reason it’s low is that the forces pushing for this outcome and the forces pushing against it are roughly balanced right now. On the “for” side: the US government has put in equity, Congress passed the CHIPS Act, Samsung (a major competitor) is having its own factory problems, and Nvidia is considering an alliance. On the “against” side: Intel keeps losing money on its foundry business, TSMC is expanding into Arizona anyway, Japan is building its own chip industry, and a key structural problem — explained below — keeps getting in the way.
The graph is not saying Intel will fail. It’s saying the outcome is genuinely undecided, and the map shows why.
The Trap Inside the Fix
Here is one of the most non-obvious things the graph reveals.
The US government, worried about what happens if Intel fails, took an equity stake in the company. That makes sense: you don’t want your national chip factory to collapse. But this created a side effect.
One of the best ways Intel could fix its chicken-and-egg problem is to spin off its foundry business into a separate company — or sell a big chunk of it to a partner, possibly TSMC. A separate company wouldn’t be competing with its own customers. It could say: “We only make chips, we don’t design chips that compete with yours.” That would make fabless companies (companies that design chips but don’t build them) more willing to trust Intel’s factory.
But the government equity stake means the government effectively has veto power over major structural changes to Intel. A spinoff or sale requires government approval. And the government’s interest is in keeping Intel intact as a national asset — not in restructuring it in ways that might look like selling off a strategic resource. So the policy designed to save Intel is also the policy that blocks the fix that would most directly solve Intel’s customer acquisition problem.
The graph connects these explicitly, and it’s not a subtle relationship: government equity stake enables the “national champion” outcome and simultaneously perpetuates the trust problem that is preventing that outcome from becoming real.
The Engineer Drain Problem
Intel has been laying off engineers as part of a restructuring effort to cut costs. This makes financial sense in the short term: fewer people, lower expenses.
The problem is that chip manufacturing knowledge lives in people’s heads. The specific tricks, the workarounds, the tribal knowledge about why a particular process step needs to happen in a particular way — this accumulates over careers. When experienced engineers leave, they take that knowledge with them, and it’s very hard to get back.
The graph models this as a reinforcing spiral. Restructuring causes engineer departures. Engineer departures slow yield improvement. Slow yield improvement means the chicken-and-egg paradox persists longer. The paradox means the factory keeps losing money. Losing money motivates more restructuring. Which causes more departures.
The restructuring that is financially necessary is, through this chain, undermining the technical progress the restructuring is supposed to protect.
The Machine Bottleneck Nobody Has an Answer For
There is one company in the world — ASML, based in the Netherlands — that makes the most advanced chip-printing machines. These machines use a technology called High-NA EUV (extreme ultraviolet lithography) and they are required for the most advanced chips. Each machine takes years to build and costs several hundred million dollars. ASML makes roughly six to eight of them per year.
The graph shows that Intel’s most advanced planned process (14A), Japan’s national chip program (Rapidus), China’s ambitions, and multiple other programs all depend on getting some of these machines. The total demand far exceeds the supply.
The graph models this as an “allocation race” — everyone is competing for a fixed number of machines per year. What it does not model is any path where the supply constraint gets resolved. There is no node for “ASML builds a second factory” or “alternative EUV supplier emerges.” The physical bottleneck is treated as fixed, and the race for machines is treated as a zero-sum competition.
The Workforce Problem With No Solution
The analysis found six separate nodes representing workforce problems: not enough engineers, not enough skilled technicians, a crisis in the pipeline of people being trained, gaps that will get worse by 2030.
Every one of these nodes has outgoing edges — they make other problems worse. None of them receive edges from anything that resolves them. There is no “expand university programs” node, no “immigration policy change” node, no “workforce development initiative” node that feeds back in.
This doesn’t mean those solutions don’t exist in the real world. It means the graph, as built, treats workforce as a fixed external constraint — something that shapes outcomes but cannot itself be shaped. If that framing is correct, and if workforce turns out to be the binding limit, then all the equipment investment and government funding in the world hits a wall of “not enough people who know how to run these factories.”
The Single Highest-Stakes Near-Term Test
The graph assigns a weight of 9.8 — out of 10 — to a single edge: the connection between a company called Terafab (associated with Elon Musk) and the resolution of the core yield-volume paradox. That is the highest-weight “potentially resolves” edge in the entire graph for the central bottleneck.
By comparison, the Apple deal (which would be a major win) is weighted 8. Government subsidies are weighted 9 but address financial losses rather than the core paradox directly. The TSMC joint venture is weighted 9 for resolving the paradox but 7 for undermining Intel’s independence and 7.5 for facing the government veto.
The Terafab edge is the single highest-leverage near-term variable the graph identifies. Whether Terafab actually places chip orders on Intel’s 18A process by 2026 is the most direct observable test of whether the resolution pathways in the graph are activating or remaining hypothetical.
What the Graph Says About China
China is under export controls that prevent it from getting the most advanced chip-making equipment. The theory behind this policy is that without the best machines, China stays stuck at older, less capable chips.
The graph shows an unexpected chain: China’s factories figured out how to use older machines (DUV technology) to approximate more advanced chips through a technique called multi-patterning. That workaround generated revenue. That revenue is being used to fund development of a domestic EUV machine — the very technology the export controls were meant to deny.
The graph models a December 2025 prototype. If that prototype progresses toward production capability, the entire “control the machines, control the capability” framework becomes less durable. The graph doesn’t model a Western policy response to that scenario, leaving it as an open question.
The Bottom Line
The graph does not say Intel will succeed or fail. It shows the structural shape of why the outcome is genuinely uncertain.
The core mechanical problem is the yield-volume paradox: you need volume to prove yields, you need proven yields to get volume, and Intel is trying to break that cycle from a standing start while competing against a company (TSMC) that has been compounding the opposite advantage for thirty years.
The most important finding is not any single node but a relationship: the policy interventions designed to secure Intel’s future — government equity, veto power over structural changes — are structurally connected to perpetuating the trust problem that is preventing Intel from winning foundry customers. The fix and the blocker are the same mechanism viewed from different angles.
The most uncertain single outcome is whether Intel becomes America’s chip factory by design or whether TSMC’s Arizona expansion, combined with Intel as a secondary player, becomes the de facto US semiconductor base — not through anyone’s plan, but as the equilibrium the forces in the graph are actually pointing toward.
The most consequential unmodeled variable is Samsung. The entire graph is implicitly conditioned on Samsung continuing to underperform. If Samsung fixes its factory problems, the customers Intel is counting on have somewhere else to go.
And the most honest thing the graph says is this: America has the money, the policy intent, and one company willing to try. What it does not yet have is the proven yield, the experienced workforce, the customer trust, or a clean path through the structural paradoxes that the map shows in detail. Whether those gaps close in the next two years is what the graph identifies as the actual decision point — not the policy, not the funding, but the factories either working or not.