Paul van Gerven
4 April

Intel’s recently separated manufacturing unit expects to rekindle profitability by about 2027.

Intel expects to be ahead of the competition at the 14A node in terms of power, performance, density and cost, CEO Pat Gelsinger revealed during a company webinar on Tuesday. Introduced into manufacturing in 2027, Intel 14A will go head-to-head with TSMC’s A14 chips, production of which will start in 2028, and Samsung’s SF1.4 node, which is also scheduled for production in 2027.

The webinar’s purpose was to sketch the path to profitability of Intel Foundry, Intel’s manufacturing arm that was recently separated from the product division. The new legally distinct unit reported pro forma sales of 18.9 billion dollars in 2023, down from 27.5 billion dollars in 2022. The 2023 operating loss amounted to 7 billion dollars.

Intel Foundry expects losses to peak this year before starting to turn things around and reach break-even operating margins midway between now and 2030. This timing roughly coincides with Intel 18A, Intel 14A’s immediate predecessor, reaching maximum production volumes. The performance and cost of Intel 18A are roughly on par with competitors’ offerings, Intel believes.

Intel leadership

As previously announced, Intel 18A will be manufacturing-ready later this year. It was commonly assumed that the node would subsequently start ramping to meaningful volumes by late 2025, but it turns out that won’t happen until (late) 2026. “The ramp of the 18A product line really only hits the large volumes in 2027 and 2028,” Gelsinger said.

Bits&Chips event sponsor registration

For comparison, TSMC started N5 (5nm) risk production in March 2019, ramping started a little over a year later and the initial maximum capacity was reached in mid-2021 (capacity was later expanded). N5 is still the foundry’s biggest earner, bringing in 35 percent of revenue in Q4 2023. The N3 node was taken into production in late 2022 and contributed 15 percent of revenue in Q4 2023.

Back to leadership

Gelsinger also gave an expanded analysis of Intel’s stumble last decade. He’d already admitted on several occasions that his company had been too slow to embrace EUV, but Tuesday he added that the capacity intensity of EUV wasn’t compatible with the previous business model.

“We were optimized for speed of node transitions and always being able to shift just one more wafer. The substantial majority of the free cash flow that the tools generated in our fabs were in the first few years of their service,” Gelsinger explained. This doesn’t work for EUV scanners, which are much more expensive than their DUV predecessors. “There was an incredible growth in capital intensity and this necessitated extending the life of both nodes and the tools that delivered them.”

In other words, even if Intel had somehow managed to maintain technological leadership, the advent of EUV would still have forced it to adopt a foundry-like model, which maximizes value extraction from manufacturing equipment as well as monetization of IP and process technology. In the EUV era, “we see that we’re very competitive now in price performance and back to leadership.”

Intel Foundry aims to be the world’s second-largest contract manufacturing facility by 2030, at which point it will derive 30 percent of revenue from external customers.