TSMC Eyes Musk's Terafab — And Intel's 14A Anchor Just Slipped

On April 7, 2026, Intel publicly joined the venture with Intel's next-generation 14A node named as the project's \"sole manufacturing process technology,\" per Yahoo Finance. Musk said at the time that Tesla, SpaceX and xAI would use the 14A node at the full-scale Terafab. It was the only committed external node win in Intel Foundry's pipeline, and the anchor on which the segment's external-revenue ramp narrative was built — external foundry revenue ran just $293M of the $5.77B Q2 FY2026 segment top line, per Intel's Q2 FY2026 earnings release dated July 23, 2026.

If Taiwan Semiconductor comes in under either Plan A (TSMC owns and operates, Terafab is anchor customer) or Plan B (SpaceX majority owner, TSMC provides process technology), Intel's 14A \"could drop out of the game\" entirely at Terafab — even as Apple's, NVIDIA's, AMD's, Broadcom's, Google's and Microsoft's evaluations of 14A test chips remain in their early stages.

Taiwan Semiconductor doesn't need Terafab to survive, with Q2 2026 foundry market share at 72.5%, per TrendForce, and a Q3 FY2026 revenue guide of $44.6B to $45.8B disclosed in YTD Q3 monthly sales — August 2026 alone hit NT$403.2B, a record. But the project lets TSMC do three things at once: lock in North American capacity for an anchor customer set already consuming 75%+ of TSMC's wafer revenue; park an exclusive fab footprint in Grimes County, Texas, alongside its existing Arizona expansion; and frame a credible competitive answer to Intel on rocket-grade and automotive AI silicon without taking balance-sheet risk on a greenfield.

Under the Plan A structure Culpan floated, TSMC would own and operate the plant, with SpaceX and Tesla contributing either equity, a wafer-volume commitment, or both. Plan B leaves majority ownership with SpaceX and assigns TSMC to operations and process technology — closer to the model TSMC uses with Sony at its Japan JASM fab. Either way, TSMC gains a captive customer without funding the full capex; the customer pays for the build, and TSMC takes a management + node-licensing annuity.

Through Q2 FY2026, Intel Foundry delivered $5.8B in revenue (+6% sequentially) on 18A output running 25% above internal target — and still lost $2.09B at the operating line, per Intel's Q2 FY2026 prepared remarks dated July 23, 2026. Of that $5.77B segment revenue, external customers contributed just $293M, per the Intel Q2 FY2026 10-Q. The Terafab win was supposed to be the inflection: a contracted external node anchor from an AAA-rated customer base over a10–15-year horizon.

TSMC, meanwhile, is in a different place. Q2 FY2026 gross margin sat at 64.2% (TTM, per the Aug 2026 monthly disclosure), and North American customers now generate over 75% of wafer revenue. A captive Terafab customer in Texas doesn't dilute TSMC's Taiwan pricing — it monetizes US capex at full corporate economics, similar to how it monetizes the Arizona expansion. For an Intel Foundry that loses $0.36 on every revenue dollar at the segment line, that is a structurally different math problem.

The customer shockwave runs two directions. Downstream, the AI silicon buyer cohort — NVIDIA, Broadcom, AMD, Apple — has been evaluating 14A test chips for the back half of 2026 per Piper Sandler, with Intel confirming \"two prospective\"14A customers still in early rumor-testing as of January 2026. If TSMC takes Terafab, those buyers now have an alternative captive path to leading-edge US capacity — and that weakens Intel Foundry's bargaining position on price, volume and PDK timing.

Upstream, the EUV and deposition equipment base (ASML, Tokyo Electron) doesn't change day-one — the same tools ship into a TSMC-operated fab as into an Intel one — but the order-book allocation shifts toward TSMC's tool pipeline. Memory, the third leg of Terafab (logic, memory, and packaging under one roof), keeps SK Hynix in scope as a partner on HBM and advanced packaging — though TSMC and Intel both retain their own packaging capacity.

Short-term (days to quarters): the next hard catalyst is TSMC's Q3 FY2026 earnings call on Thursday, October 15, 2026, where Texas capacity and any Terafab linkage is on the agenda. Intel reports Q3 FY2026 in late October. NVIDIA, Broadcom and AMD report through November. A definitive Plan A or Plan B announcement — or a hard \"no\" from TSMC — is the binary for Intel shares between now and year-end.

Long-term (1–3 years): the structural shift matters more than the daily tape. TSMC's Q2 2026 72.5% foundry share is the highest in industry history, per TrendForce. Every additional captive US customer widens TSMC's moat against Intel and Samsung Foundry, both of which compete on US-made-leading-edge capacity. The 14A node — even if Intel signs a different anchor — becomes harder to underwrite without the Terafab customer base, and Intel Foundry's external-revenue line may stay sub-$1B through 2027 absent a major new win.

A week ago, the Intel bull case was that18A external revenue would lift the Foundry segment out of losses, anchored by Tesla/SpaceX/xAI at 14A. As of the morning of Oct 5, 2026, that anchor is in renegotiation, TSMC is the higher-probability partner, and Intel shares are pricing that. The Intel Foundry story is not over, but the external-revenue case has narrowed to the 18A evaluations underway at NVIDIA, AMD, Broadcom, Apple, Google and Microsoft — and the longer Terafab stays in discussion, the more leverage TSMC holds on price and process commitment when a deal lands.

For investors, the trade is binary: either TSMC signs (and Intel's Foundry external narrative thins further) or TSMC doesn't (and Intel keeps its anchor). The asymmetry is on the Intel side — a no-decision still keeps the overhang alive, while a TSMC yes compresses the bull case immediately. The base-rate read: TSMC's market cap reaction (+3% Taiwan, +1% US pre-market) implies the market is leaning toward a yes, and is right to.