Intel 18A is here, the node the entire comeback was built to reach, and the man who bet the company on it is gone. Pat Gelsinger, who staked his return and Intel’s future on five nodes in four years, was pushed out at the end of last year, months before 18A, the fifth node, the one that was supposed to vindicate him, actually shipped. Panther Lake is the first product on it. RibbonFET and PowerVia, the two firsts that died with 20A as a product node, live here instead, and on paper 18A is the most advanced thing Intel has built since it lost the lead.
18A pairs RibbonFET, Intel’s gate-all-around transistor, with PowerVia, its backside power delivery, and PowerVia is the real flag in the ground. Moving the power rails to the back of the wafer is something neither TSMC nor Samsung has in volume yet; their first gate-all-around nodes this year still run power down the front the way every chip ever has. So if PowerVia holds up at volume, Intel has shipped a genuine leadership feature first, for the first time since the FinFET days, which is the entire thing the comeback was supposed to produce. After a decade of catching up, Intel finally has a card the leader does not.
Intel Foundry needs marquee outside customers running real volume on 18A. What it has is a handful of announced names and a great deal of polite interest, while the leading-edge work that actually pays, the AI accelerators and the flagship mobile parts, still routes to TSMC. A node that wins the spec sheet means nothing if the companies placing the biggest orders do not trust the fab to build them, and a decade of Intel misses is a long shadow to argue against. Being ahead on the transistor and being a real foundry business are not the same thing, and 18A has to turn the first into the second by converting reputation into signed wafers. Reputation is the slowest thing in this industry to rebuild.
Gelsinger is gone, a new chief executive inherited the seat in the spring, and the foundry is bleeding money fast enough that the politics around Intel have left the commercial tier entirely. The United States government has reportedly taken an equity stake in the company, with outside investors putting capital in alongside it, the kind of intervention that only happens when a business is judged too strategically important to fail and too fragile to assume it won’t. The exact terms are still settling and worth watching closely. The signal under them is not subtle: Intel’s manufacturing is now national infrastructure, not just one company’s competitive headache.
A clean ramp on Panther Lake, with PowerVia holding up at real volume, lets Intel walk into customer meetings carrying proof instead of promises: a leading-edge Western fab with a feature TSMC still cannot match. That is the version everyone inside Intel is praying for. The other version needs far less to go wrong. Yields could come in soft, or PowerVia could prove a nightmare to manufacture at scale; either one is enough on its own. And even if the silicon is flawless, the biggest buyers can just keep sitting on their hands, waiting for someone braver to qualify the node first. However it lands, the most advanced thing Intel has built in a decade ends up another impressive piece of engineering that never became a business, right as the money and the patience run out together.
This is the most important node Intel has shipped in twenty years, and standing here in late 2025, I genuinely do not know if it saves the company. Perso, the engineering looks real: PowerVia is the most impressive thing Intel has done since Tri-Gate, and the part of me that grew up on Intel silicon wants this comeback to land. But wanting is not forecasting. Intel is ahead of TSMC on backside power, and the customers are still waiting to see whether anyone else qualifies the node first. The man who bet everything on this moment was pushed out before he got to watch it ship. 18A is Intel asking the whole industry to trust it again, and for once the technology actually backs the ask. That still might not matter. Nobody is obligated to care, and the buyers who count have spent a decade learning not to. Whether they sign is the only question left, and it is the one thing Intel cannot manufacture.