Comment Intel's fledgling foundry business is in trouble. The division is bleeding billions each quarter and now the chipmaker has revealed that it won't even manufacture parts on its own long-awaited 20A node.
TSMC will be eating Intel Foundry's lunch through much of 2025, while Chipzilla focuses on ramping production of its mainstream 18A process tech. In fact, it appears that the Taiwanese foundry giant will be responsible for manufacturing the majority of Intel's 2024 product lineup – including newly launched Lunar Lake mobile chips and upcoming Arrow Lake desktop parts.
Can CEO Pat Gelsinger really claim "five nodes in four years" when one of them – 20A – will never see the light of day?
Intel is, of course, handling packaging, and its Xeon 6 roadmap will continue to be manufactured in-house – albeit on its Intel 3 process node. That's a small consolation prize for Gelsinger, who is desperately trying to convince customers to embrace Intel Foundry.
It's a lot to ask anyone to invest in your process tech when you yourself aren't using it. And, given Intel's shoddy track record of meeting deadlines, we'd forgive customers for being a little nervous right about now.
Mixed signals, worse optics
The decision comes amid reports that 18A has failed to impress. Per Reuters, Broadcom – one of the leading designers of network ASICs and semiconductor IP – was reportedly in talks to manufacture chips with Intel. But after receiving test wafers built on 18A last month, it concluded they weren't viable for volume production.
If that sounds familiar, an earlier report by the Financial Times alleged that SoftBank had also approached Intel about manufacturing AI chips – but walked away from the talks after determining the foundry couldn't ramp volume fast enough.
Those reports paint a very different picture of Intel's foundry business than Gelsinger. He has insisted that, despite financial headwinds and the extreme cost associated with the plan, Intel's fabs are garnering considerable interest in the weeks since it released the process design kit (PDK) for 18A.
"We now have about a dozen customers that are actively engaged with around that PDK. We have eight tape-ins that we expect to finish by the middle of next year." he said during Deutsche Bank's Technology Conference last week. "I'm happy to update the audience that we're now below 0.4d0 as defect density."
Ben Sell, VP of technology development Intel, cited 18A's strong yields as the deciding factor in the chipmaker's decision to move production of Arrow Lake off of 20A and instead to TSMC.
The trust is gone
Unfortunately for Gelsinger and his compatriots, positive vibes won't change the reality that Intel Foundry is in deep trouble – and investors are losing faith in what, to them, probably looks like a money pit.
Since the start of the year, Intel's share price has plunged nearly 60 percent. The precipitous decline has spurred multiple class action suits from investors who feel Gelsinger and CFO David Zinsner misled them about the health of the Foundry division. To make matters worse, the x86 giant now faces the very real possibility of losing its spot on the Dow Industrial Average.
Regardless of whether those cases go anywhere, investors' frustrations are understandable. Under the new reporting structure put in place earlier this year, the division posted a $7 billion operating loss in 2023.
The current fiscal year is already shaping up to be far worse. In the first half of 2024, Intel Foundry had already racked up $5.3 billion in operating losses, weighing down Intel's Product division like a boat anchor. And, by Gelsinger's own estimate, its fabs are unlikely to break even until 2027.
That's a far cry from the 60 percent gross margin and 40 percent operating margin executives have been promising for years now.
The situation is dire enough that, faced with $1.6 billion in losses for second quarter of 2024, Gelsinger announced the business would shed 15 percent of its workforce — about 16,000 employees — and slash capex spending by more than 20 percent to $25–$27 billion in 2024 with more cuts planned for the next year.
No choice but to embrace TSMC
Intel's financial collapse no doubt played into the chip shop's decision to abandon its 20A production run. The process node, envisioned as a sort of testbed for 18A, was never going to see mainstream deployment outside of Intel's internal products – at least as we understand it.
If 18A is really yielding as well as Gelsinger and Sell claim, then the decision to ditch 20A makes sense even if the optics are terrible. Intel's Product division is carrying Foundry on its back right now, and it would almost certainly have cost more to ramp production of 20A than to move Arrow Lake to TSMC.
If things go Gelsinger's way, this concession will be short-lived. Speaking with analysts during the Q2 earnings call, Gelsinger acknowledged Intel's growing reliance on TSMC – but insisted that the majority of Intel's products would return home starting in 2026.
To this end, its next-gen Clearwater Forest Xeons and Panther Lake client CPUs, due out next year, are slated to use its 18A process tech. The implication here is that Intel not only wants to bring its CPUs back in house, but its GPUs, NICs, and AI accelerators as well.
However, that's assuming Intel's shareholders have the patience to see Gelsinger's grand plan through to its conclusion.
- Intel Arrow Lake to be made elsewhere as 20A process node canned
- Intel's 120 TOPS Lunar Lake AI PC chips have landed
- Dow-ward spiral: Intel share price drop could see it delisted from blue-chip index
- Gelsinger opens up about Intel troubles amid talk of possible split
Too big to fail?
Gelsinger will reportedly present potential remedies to Intel's board later this month, and is said to be weighing even more drastic measures – including the possibility of spinning off Foundry as an entirely separate business.
While shareholders might find it appealing to get Foundry's billions in operating losses off Intel's books, it's hard to see how it would survive on its own.
What's more, Gelsinger has all but bet Intel's future on his contract manufacturing aspirations panning out. He has committed to building more than $100 billion worth of fab capacity around the globe, with many of these projects backed by private equity groups like Apollo and Brookfield Asset Management.
Uncle Sam has a vested interest in Intel Foundry's success too. Not only is Intel slated to receive $8.5 billion in subsidies and another $11 billion in loans under the US CHIPS and Science Act, Chipzilla is America's foundry.
One of the chief goals of the bill was to alleviate US reliance on foreign fabs. Intel is the only US fab operator still producing leading edge chips, after GlobalFoundries abandoned its 7nm process tech back in 2018.
A spin off of Intel's Foundry business would not only jeopardize Gelsinger's vision, but has the potential to destabilize US national security policy.
Short of the US nationalizing Intel Foundry – a decision we have no doubt would be even less popular than whatever Gelsinger is cooking up – walking away from the division is going to be difficult. If it fails, there goes Intel Products' roadmap – you know, the thing actually making money right now.
Regardless of what the future holds for Foundry, whether or not Gelsinger himself will be there to see it through is another matter entirely. Lest we forget former CEO Bob Swan was ousted for far less. ®