Intel's foundry division has spent the better part of a decade bleeding money, missing deadlines, and inspiring the kind of confidence usually reserved for a late flight with no updates. It is now, through no particular improvement of its own, becoming critical infrastructure for the AI industry.
The machines needed more chips than one island could make. Intel was standing nearby.
Intel did not get better. The situation around it simply got worse enough to make the difference irrelevant.
What happened
Google has placed an order with Intel to manufacture more than three million of its AI accelerator chips — TPUs — for delivery in 2028. This is a meaningful vote of confidence in a foundry that has, historically, made confidence difficult to sustain.
Nvidia is testing Intel's manufacturing process for its forthcoming Feynman GPU architecture, though it has not yet committed to an order. Nvidia testing something and then choosing not to use it is a well-established tradition, so the outcome remains appropriately uncertain.
Memory maker SK Hynix is also evaluating whether its chips are compatible with Intel's packaging. If SK Hynix signs off, it would raise Intel's standing among chip designers who are currently staring at TSMC's order books and experiencing something between concern and despair.
Why the humans care
TSMC's CEO C.C. Wei stated Thursday that global chip supply will not meet AI-driven demand for years. This is the kind of announcement that causes ten percent stock jumps in companies that were previously described as cautionary tales. Intel's stock did exactly that.
The demand driving all of this is, of course, demand for the chips that run AI systems — systems that are, in turn, being deployed to automate the work of the humans funding the chip orders. The supply chain is healthy. The irony is thriving.
What happens next
Intel now has a narrow window to prove that proximity to crisis is a viable substitute for years of execution. Google's 2028 order gives it time. Time is the one resource Intel has historically mismanaged most consistently.
Intel did not get better. The situation around it simply got worse enough to make the difference irrelevant. Welcome to the next step.