Malta's $21 Billion: Japan Buys the Wafers, the Tool Chain Gets Paid First, and GlobalFoundries Gets 4% Diluted

On Oct. 14, 2026 GlobalFoundries announced it had been selected as technology provider and operator for two new 300mm fabs at its Malta, New York site under a US–Japan initiative worth up to $21 billion, with Japan providing the investment. Reuters reported that the two plants will be jointly owned by the US and Japanese governments, and the Wall Street Journal noted the funding was enabled by the US trade and investment agreement with Japan. Leaves GlobalFoundries running fabs it does not own, starting in 2029.

The asymmetry is the whole trade. GlobalFoundries avoids writing a $21 billion cheque, collects operating revenue from 2029, and gives away roughly 4% of the company to get it. Japan converts a macro pledge into physical capacity it owns and controls, without buying a single share of the foundry. Japan ends up owning the asset and outsourcing the operating risk.

That last omission matters more than the headline. The capacity is real and the money is committed, but the revenue case for 2029 rests on arrangements nobody has published. Investors are underwriting a 2029 option whose exercise price is not yet visible.

GlobalFoundries describes the demand behind the build as automotive, industrial and connected infrastructure markets. Its own automotive portfolio is a list of mature-node parts: 32-bit microcontrollers, non-volatile memory, analog and power management, battery management systems, DC-DC converters, motor drivers and 48-volt battery systems, plus radar, LiDAR and camera sensors. The programme leaves out every leading-edge logic node.

The payoff for GlobalFoundries is a captive book rather than a spot market: capacity a sovereign has already paid for, aimed at customers it already supplies. On its first-quarter call the company said automotive revenue was on track for a sixth consecutive year of double-digit growth in 2026, driven by sensor and power ramps. A sovereign-funded captive order book converts a 2029 capacity event into revenue that is contracted years earlier.

The same documents that put automotive and power chips at Malta also put silicon photonics, silicon-germanium and advanced packaging inside the same money. The Commerce amendment redirects up to $955 million of GlobalFoundries' existing CHIPS award to those technologies, requires the company to expand capital investment at Malta and Burlington, Vermont to $5 billion, and names the first US-based source of silicon interposers. The build funds the sensors and the optical interconnect side by side.

None of this makes Malta an AI fab. There is no GPU logic in the announcement. But reading the programme as automotive-only misprices two things: the photonics platform that now sits inside the same walls and the same capital plan, and the escalation of Washington's equity take from about 1% in July to roughly 4% in October. The chip content is mature-node, but the packaging and photonics content is the fastest-growing part of the company's revenue mix.

Two 300mm fabs at roughly $10 billion apiece is a $21 billion capital programme — on its own larger than the $18.1 billion the Americas will spend on all 300mm fab equipment in 2026, a figure SEMI expects to reach $40.6 billion by 2029. Across the global build-out, 300mm equipment spending runs at roughly three times fab construction spending, $149 billion against $47.5 billion in 2026. Equipment, not construction, absorbs most of a fab's capital.

The spend flows to the broad-line 300mm vendors, not to the AI tool chain. Applied Materials and Lam Research sell deposition, etch and inspection steps into every mature-node fab, KLA sells the metrology, and Tokyo Electron sells the coating, developing, etch and cleaning lines that standardise a 300mm flow. ASML is the clearest exclusion: with no leading-edge logic in the announcement, the lithography franchise that dominates the AI build-out does not collect this money. The vendors paid by Malta's cheque are the ones the AI cycle already de-rated.

That table is the cleanest cross-market conclusion in the trade. The Japanese vendor that captures a Japanese sovereign programme trades at 4.4x trailing EBITDA and about seven times earnings, with a 4.7% dividend yield and net cash, while the US vendors that capture the AI programme trade at 26x and 28x. Both chains are roughly half off their 52-week highs. The cheapest way to own Malta's cheque is Tokyo, not Silicon Valley.

The October selloff also explains the timing risk. The tool chain is not short of demand; it is short of supply. Lam Research's chief executive said he expects ten new tier-one fabs online over the coming year and visibility through 2028, while China was 35% of the company's September-quarter revenue with a significantly lower share expected next. Malta's tools will be booked into a supply-constrained queue, which protects pricing and also sets the 2029 start date.

GlobalFoundries' existing Fab 8 campus at Malta runs on 15 megawatts of contracted demand against 90 megawatts planned. The New York Power Authority has already allocated an additional 15 megawatts of low-cost ReCharge NY power and approved procurement of 90 megawatts of high-load-factor power on the energy market. Two new 300mm fabs need more firm power than the campus has ever contracted for.

Washington's roughly 4% stake is worth about $975 million against a $24.4 billion market capitalisation, and it trims trailing earnings per share from about $1.28 to about $1.23 — before Malta contributes a single wafer. At the Oct. 23 close the shares changed hands at 11.9x trailing EBITDA and 34.7x trailing earnings, on trailing revenue of $6.94 billion and a 27.4% gross margin. Investors are paying 34.7x trailing earnings for capacity that does not ship until 2029.

The reaction was brief. Shares closed at $46.70 on Oct. 13 and $48.94 on Oct. 14 on 13.8 million shares against 4.95 million the day before, then gave the move back: $46.11 on Oct. 19 and $44.39 at the Oct. 23 close, roughly 52% below the 52-week high of $92.19 and below both the 50-day and 200-day moving averages. The market has already decided the sovereign cheque is not worth a re-rating.

Japan's $21 billion buys something narrower and more durable than an AI boom: an owned, US-located supply of mature-node automotive and power silicon that enters the US market duty-free, underwritten by a state rather than by a fab's own balance sheet. The equipment vendors collect the cheque in 2027 and 2028, GlobalFoundries collects revenue in 2029, and Washington's 4% is collected now. Malta is an equipment event with a foundry option attached.

That option is not worthless. Trailing revenue of $6.94 billion at a 27.4% gross margin, $796 million of trailing free cash flow and a net debt position of $591 million leave room to fund a 2029 ramp without a capital raise, and the photonics work inside the same walls gives the mix a second engine. The uncertainty is concentrated in three places — contracted volumes, firm power and the December tool queue — and all three resolve well before 2029. An investor is being asked to underwrite a 2029 option with no 2029 volumes, and the next disclosure arrives on Nov. 4.