GlobalFoundries closed at $38.62 on 6 November, 58% below the $92.55 it printed on 26 May, and 15% below where it stood two days earlier, before a third quarter that beat on revenue, margin and earnings451. The market sold the outlook, not the quarter: net capital expenditure of about $4 billion in 2027 against $1.3 billion this year, and a share count heading to 600 million as the US government takes its stake26. We read the same call and heard who pays. Japan is funding up to $21 billion of new fabs at Malta that GF will operate rather than own, Washington has reallocated $955 million of its CHIPS award to photonics and packaging, customers are prepaying for capacity through 2029 and will cover a third of next year's capex, and a 35% investment tax credit covers part of the rest892. The stock is being priced on the size of the cheque. It should be priced on who signs it.
This piece belongs to our equities section, opened in November 2026 around one frame: the next cycle is being built in silicon and in qubits, and the companies that own the fabs, the tools and the machines are where the record will be written. The analysis is as of 9 November 2026.
Where we are now
The year in one line: $34.92 at the end of 2025, $92.55 on 26 May after the quantum unit launched and Mubadala sold a $2 billion block into the strength, $38.62 today, still up 11% for the year and still below the $47 IPO price of 20214513. The third quarter: revenue $1.90 billion, up 13%; a record non-IFRS gross margin of 31.4%, up 540 basis points in a year; non-IFRS earnings of $0.53 against roughly $0.50 expected; a $441 million non-cash charge for the 9.9 million shares issued to the US government that turned the IFRS line into a loss1. Blended wafer pricing rose about 5% year on year, the first increase since 2023, and utilization ran in the high 80s2. The fourth-quarter guide is $2.0 billion at 31.5% with earnings of $0.591.
The mix has changed under the surface. Smart mobile, the handset business that was more than 60% of revenue at the IPO, is 32% and shrinking 19% a year. Communications infrastructure and datacenter grew 81% to $316 million, its eighth straight quarter of double-digit growth; home and industrial IoT grew 49%; automotive 20%2. Silicon photonics, the optical chips inside the transceivers that connect AI servers, was above $200 million of revenue last year, is about $500 million this year, and management pulled the $1 billion annualized target forward a year to the end of 2027, with capacity booked through 2027 and increasingly through 20282.
At $38.62 the stock trades at 16 times forward earnings, 9.9 times EBITDA and 1.8 times book, with about $2.1 billion of net cash4. TSMC trades at 17.8 times and 14 times, Tower, the closest photonics peer, at 21.7 and 20.815. Nine brokers cut targets in the week after the quarter; the consensus is still $64.7774.
The bull case
1. AI optics is revenue, not a story
Every AI data centre is a network problem before it is a compute problem, and the network runs on optical transceivers whose photonic chips are made in a handful of fabs. GF's is one of them, with more than 50 customers including seven of the top ten transceiver makers, and it is sold out2. The communications segment grew 81%; the SiGe process behind 200-gigabit lanes is sold out; photonics capacity in Singapore is being tripled over the next year2. This is the only part of the semiconductor cycle where a trailing-edge foundry is a bottleneck, and GF is charging for it: the first price increase in three years, a record margin, and a stated path to 35% gross margin at the end of 2027 and 40% at the end of 20283.
2. Someone else pays for the fab
The two new 300mm fabs at Malta are funded under the US–Japan investment framework, up to $21 billion, with the first online in 2029 at 900,000 wafers a year; GF is the technology provider and operator8. The federal government moved $955 million of its award to the photonics and packaging lines GF needs most, and takes about 7% of the equity in return, which is the dilution the market sold92. TSMC signed a $2 billion, five-year contract for US-made silicon interposers at Malta, which makes GF a supplier to the AI leader's own packaging10. Customers are prepaying for capacity through 2029, covering a third of 2027 capex, and the 35% tax credit covers more; management says the payback is about two years2. A foundry whose expansion is financed by a sovereign, a government and its customers is not spending $4 billion. It is spending a fraction of it and keeping the capacity.
3. The diversification already happened
The bear case on GF for five years was handset exposure. It is now a third of revenue and falling, while automotive, IoT and communications are two-thirds and growing double digits2. Renesas signed a multi-billion-dollar long-term agreement in February; Monolithic Power starts volume in Singapore early next year; Marvell is expanding SiGe at Burlington141. And the quantum unit, honestly: it exists, PsiQuantum's chips are built by the thousands at Malta, the $375 million Commerce award is a letter of intent, and there is no disclosed quantum revenue1112. It is an option on the foundry, the same way IBM's is an option on the software. We do not pay for it here and we do not need it.
The bear case
Now the argument against, which is what the stock did last week.
The capex cliff and the dilution are real numbers. Net capex of $4 billion in 2027 after three years at a tenth of revenue, against free cash flow this year of roughly $700 million; a third prepaid still leaves more than the company generates26. The share count rises from 558 million to about 600 million as the government takes its stake, a $441 million charge is already booked, and the $500 million buyback authorization is exhausted with no replacement12. Jefferies put it in one sentence: the capex and the dilution limit the per-share upside6.
The core is still shrinking and the growth engine is crowded. Smart mobile fell 19% in the quarter and is guided down 22% in the fourth; 2025 revenue grew 0.6%, partly on price cuts to dual-sourced customers2. Utilization is only in the high 80s before photonics capacity triples, and TSMC, Tower, STMicro and UMC are all entering photonics, a point raised on the call3. If AI capex digests in 2027, GF will have tripled capacity into it.
The overhang. Mubadala owns about 73%, roughly 400 million shares, and sold $2.8 billion of stock this year at $42 and at $9013. The government will own 7% and has registered shares for resale. The stock fell 58% in five months and 15% on a quarter that beat; nine brokers cut targets in a week7. A 157-million-share float with a seller that size above it is a stock that cannot rally far without being sold into.
GlobalFoundries is the cheapest way we can find to own the physical bottleneck of the AI build-out, the optical chips every data centre needs and few fabs can make, and the market has just priced its expansion as if GF were paying for it alone. It is not. The fabs are Japan's, the photonics lines are Washington's, the capacity is prepaid, and what GF keeps is the margin: a record now, 35% next year by management's own target.
Our view: GlobalFoundries trades at $55, in a $52–58 band, by the end of Q4 2027. That is 42% above today, below the consensus target and a long way below May. The path runs through the fourth-quarter print and the first 2027 guide in February, the groundbreaking at Malta, and the photonics run-rate updates through the year. Conviction is moderate-to-low — about 55% — and we say why: the Mubadala overhang is permanent, the float is thin, and the 2027 capex number can get bigger before it gets smaller.
Our bet runs one way: up, on photonics and on who pays for the fabs. We are not betting on quantum revenue inside the horizon. If GF closes a week below $32 first, we were wrong — see below.
What would prove us wrong
One direction, one closing level, and the fundamentals that would hollow it out.
$32 is the floor of the 52-week range and below the 2025 year-end price. A week closing beneath it says the market has decided the capex cycle destroys value regardless of who funds it, and the thesis is closed against us, whatever happens afterwards.
Any of these weakens the thesis while the price holds, and we will log it as it happens: the February 2027 guide putting net capex above $4.5 billion or gross margin below 31% for the year; photonics revenue growth slowing below 50%; the Japan-funded Malta fabs failing to break ground by the end of the first quarter of 2027; or a further Mubadala block below $40.
GlobalFoundries trading in the $52–58 band on the NASDAQ by 31 December 2027. If neither the target nor the invalidation resolves by then, the thesis is booked as it stands — unresolved and closed. It does not get extended.
When a condition triggers, we book the result and publish the piece. As always.
This is analysis, not investment advice. Do your own research. The thesis is not changed after the fact.