Intel's 18A Yield Gains Put TSMC's 2nm Scale to the Test

By Tom Whitfield

Intel's Data Center and AI revenue rose 78% to $7.32 billion in the third quarter, even as Intel Foundry recorded a $1.85 billion operating loss. Those two numbers frame the contest with Taiwan Semiconductor Manufacturing: Intel has demand and a working 18A process, while TSMC has scale, pricing power and a 58.7% operating margin.

Intel and TSMC both reported third-quarter results in October. Intel's results showed that its 18A transistors work and that yields keep improving. TSMC reported what a mature leading-edge foundry looks like, with 2nm ramping across Hsinchu and Kaohsiung and gross margin at 66.4%.

Where the two foundries stand

Servers carried Intel's quarter. Demand for Xeon processors still outpaced supply, and that shortage capped how much the company could sell. The same AI buildout driving server demand also lifts orders for power, cooling and networking suppliers.

Intel Foundry's 18A production costs fell about 40% since the first quarter. Chips built on 18A went from less than 5% of client CPU revenue to nearly one-third of it. CFO David Zinsner said Panther Lake margins remain short of the corporate average, so the cost gains had not yet reached every product.

TSMC's gross margin slipped 1.3 percentage points to 66.4% because 2nm, now 8% of wafer revenue, carries lower margins early in a ramp. TSMC management treats that ramp cost as temporary. CEO C.C. Wei said demand remains "still far above our capability to supply today."

The operating comparison is stark:

• Growth engine: Xeon server CPUs at Intel; high-performance computing at 68% of revenue at TSMC.
• Leading node: 18A ramping at Intel, with its 14A 0.9 process design kit released; 2nm ramping at TSMC across Hsinchu and Kaohsiung.
• Foundry profitability: a $1.85 billion operating loss at Intel Foundry; a 58.7% operating margin at TSMC.

Intel sells everything it can build; TSMC earns more on what it sells.

Can Intel close the lithography gap?

Intel is moving early on high-NA extreme-ultraviolet lithography with ASML, trying to pull ahead in process technology. TSMC plans to adopt High NA starting in 2030, once tool costs make economic sense. Wei said TSMC remains the best company to offer the most advanced technology to its customers.

Scale lets TSMC wait. The foundry served 534 customers in 2025, so each node's cost spreads across many buyers. Intel's external foundry revenue was just $411 million, mostly from a nonrecurring program. That is a thin base for a business seeking outside customers.

A lithography lead without a customer list does not close the revenue gap.

What a 14A customer would settle

Intel's roadmap puts the 14A 1.0 process design kit in the first quarter of 2027. CEO Lip-Bu Tan also expects to announce advanced-packaging customers within the next two quarters, which would put external proof behind the foundry effort. TSMC's A14 targets volume production in 2028, roughly the same timing as Intel's 14A.

Funding is the wider gap. Zinsner said there is "probably not a dynamic where we're going to have positive free cash flow next year." A process roadmap consumes capital before outside customers generate revenue.

Investors have already marked down Intel's optimism. Intel shares fell 43.3% over the past month, against 14.25% for TSMC. The market wants signed customers before it pays for the roadmap.

Without a named 14A customer, Intel's 14A timetable remains a promise rather than a business.

Can physics bridge the economics?

Intel can match TSMC on 14A physics. Yield consistency, customer trust and economic scale are harder problems, and they decide which foundry wins volume orders.

Intel's roughly 40% cost decline at 18A since the first quarter shows that a new node can get cheaper quickly. TSMC's 534-customer base spreads those costs further, and its pricing power supports steadier compounding. Intel's Xeon momentum and its US fabs give the turnaround concrete drivers, while a named, high-volume 14A customer would change the comparison quickly.

The next checkpoints are Intel's 14A 1.0 design kit and its customer announcements; until then, TSMC's 2nm scale and 58.7% operating margin keep the commercial gap open.