On October 29, STMicroelectronics raised its 2026 forecast for AI data-centre revenue to about $1.3 billion, up from a July forecast of more than $1 billion, and reiterated its goal of well above $2 billion for 2027. The raise came on a third quarter in which revenue of $3.76 billion and a 37.3% gross margin both came in above the midpoint of the company’s own outlook, and operating income excluding one-off charges nearly doubled to $429 million, beating the $400.5 million expected by analysts polled by LSEG. CEO Jean-Marc Chery told analysts the demand was driven mainly by the company’s silicon photonics chips.

This desk has spent the year covering this lane through startups and their demonstrations. This is the same lane, but a different kind of event. A startup’s demo says a thing can be built. A foundry’s capacity conversion says it can be built at volume. An incumbent chipmaker raising a revenue forecast on the strength of its photonics line says it is being bought, at scale, by customers who are already paying. That is the industrial phase, and it arrived with a number attached. The number needs its caveats kept in the same sentence as the number: the $1.3 billion is a management forecast, not booked sales, and the well-above-$2 billion figure for 2027 is guidance a year out. The $1.3 billion is also AI data-centre revenue across the company’s whole offering for that market — silicon photonic ICs, electronic ICs, microcontrollers, and power solutions — not a silicon photonics revenue line. STMicro does not disclose a photonics-only figure, and none can be derived from the total. What the company does say, in Chery’s words on the call, is that the demand is driven mainly by its silicon photonic ICs. That is the claim worth examining, and the rest of this article is the examination.

What was actually announced

Precision first. STMicro reported third-quarter 2026 results on October 29, for the quarter ended September 26. Net revenues were $3.76 billion, up 17.9% year over year. Gross margin was 37.3%, up 410 basis points from 33.2% a year earlier, on lower unused capacity charges and a better product mix. Non-U.S. GAAP diluted earnings per share were $0.37. The company guided fourth-quarter revenue to $4.15 billion at the midpoint, plus or minus 3.5%, with gross margin of about 39% — above analysts’ estimate of $4.10 billion — implying full-year revenue of approximately $14.5 billion, up 22.8% year over year.

The segment detail is where the photonics story sits. Revenue from communication equipment and computer peripherals rose 64% year over year and 14% sequentially, driven by AI data-centre and low-Earth-orbit satellite revenues. The RF & Optical Communication product segment grew 46.2% year over year, while industrial rose 40%, automotive 5%, and personal electronics fell 4%. On the call, Chery said demand strengthened during the quarter, with the overall book-to-bill ratio exceeding 2 and backlog above $20 billion — roughly five quarters of revenue at the fourth-quarter guidance midpoint.

The forecast itself has a short history. In July, STMicro said AI data-centre revenue would exceed $1 billion in 2026. The October update moves it to about $1.3 billion — a $300 million upward revision on a forecast only three months old, made with two months of the year left to run. By late October, a $1.3 billion full-year figure is largely a readout of the order book, not a hope about demand that has yet to materialize.

Two remarks from the call deserve to be quoted rather than paraphrased. Asked whether heavy AI spending was sustainable, Chery told journalists that STMicro’s data-centre business was “much more related to the top five hyperscalers than the neo-cloud business” — a pointed distinction at a moment when smaller AI cloud providers’ debt levels have drawn scrutiny. And in the earnings statement he said, “In an environment where demand remains well above supply, we continue to expand capacity.” The company also said it is raising prices on selected products, according to CFO Lorenzo Grandi — the kind of pricing power a supplier claims only when it believes its customers have nowhere else to go.

What PIC100 is, and why it is the thing being bought

A photonic integrated circuit puts optical functions — modulation and photodetection — onto a silicon chip, so that data moves between servers as light through fibre rather than as electrical signals through copper. Light carries more data per lane, over longer distances, at lower energy per bit, and the distance problem is exactly what AI clusters have: as accelerator counts per cluster grow into the tens of thousands, the interconnect between them becomes the bottleneck, and copper runs that were adequate for a rack of servers are not adequate for a training cluster.

PIC100 is STMicro’s first silicon photonics technology, announced in February 2025 and built on the company’s 300 mm wafer platform in Crolles, France. Most silicon photonics production to date has run on 200 mm lines at merchant foundries; a 300 mm line means more dies per wafer and the economics of a mature high-volume CMOS fab. The platform supports 200 gigabits per second per lane using 100 Gbaud PAM-4 signalling, and it is designed into 800G and 1.6T optical modules — the module generations hyperscalers are deploying now and next. In March 2026, STMicro announced that PIC100 had entered high-volume production on 300 mm wafers for leading hyperscalers, and said it was planning capacity expansions to more than quadruple production by 2027, with further expansion in 2028.

The 800G and 1.6T figures are worth unpacking for readers who do not live in transceiver datasheets. An 800G module moves 800 gigabits per second in aggregate; a 1.6T module moves 1.6 terabits, assembled from lanes — eight lanes of 200 gigabits each for 1.6T, four for 800G — and the PIC100’s 200G-per-lane capability is what makes the arithmetic work. The next step on the roadmap, which STMicro has already named, is PIC100 TSV: a version integrating through-silicon vias to increase optical connectivity density, aimed at the optical engines for near- and co-packaged optics. PIC100 is the pluggable-module product today and the intended engine for the packaged-optics products tomorrow, and the capacity expansion is sized for both.

Why a forecast raise from an incumbent is different

The desk’s startup CPO arc has been a story about possibility: can co-packaged optics be built, at yield, at a price a switch vendor will pay? The STMicro story is about a different question. A diversified chipmaker does not raise a revenue forecast by $300 million on a three-month-old guide because a technology is promising. It does so because orders are in hand, the book-to-bill exceeds 2, backlog is five quarters deep, and the capacity it has is not enough — which is why it is expanding that capacity more than fourfold by 2027. The forecast is a lagging indicator of demand that has already arrived, and the capacity expansion is the forward-looking bet that it will keep arriving.

There is a second difference, and it cuts in the opposite direction. A startup’s entire valuation can ride on one photonics product; STMicro’s cannot. AI data-centre revenue of $1.3 billion is about 9% of the company’s implied $14.5 billion full-year revenue, and its gross margin of 37.3% — while up sharply — is still well below the 44% to 46% intermediate target management reaffirmed on the call. The photonics business is pulling the company forward; it has not yet transformed it. That is the honest scale of the claim, and paradoxically what makes it credible: a company with a $14.5 billion revenue base and a backlog of $20 billion does not need to inflate a photonics story.

The third difference is about who bears the manufacturing risk. The foundry stories this desk covered in November — UMC reportedly winning PIC orders at seven times CMOS pricing, Tower moving Innolight’s near-package optics device into early production — are stories about merchant capacity being built and booked. STMicro is not a merchant foundry customer in this lane; it is the manufacturer, running its own 300 mm line in Crolles. When it says it will more than quadruple production by 2027, it is describing its own factory, its own yield ramp, and its own capital spending. That vertical integration is a strength and a concentration of risk at the same time: there is no second source to fall back on if the Crolles ramp stumbles.

The industrial-phase read

Set the STMicro announcement against the rest of the month’s optical-interconnect news. On November 2, Marvell and Lumentum announced a definitive agreement for Marvell to acquire Lumentum in an all-stock transaction with an implied enterprise value of $115 billion — a consolidation of the laser supply chain, expected to close in the first half of calendar 2028, subject to shareholder and regulatory approvals. On November 3, Tower and Innolight put a near-package optics device into early production on a merchant foundry line. In the same week, UMC was reported to have won photonic IC orders from Innolight and SILITH, with a foundry at 92% utilization converting capacity for silicon photonics.

Each of those stories answers a different question. Marvell–Lumentum answers who owns the supply chain. Tower–Innolight answers whether the next architecture can be built. UMC answers whether merchant foundry capacity exists for it. STMicro answers whether an incumbent can sell it at scale and raise a forecast on the strength of it. Four answers to four different questions, all arriving within a month, is what a lane entering its industrial phase looks like. The startup arc was about whether the technology works. The industrial phase is about who makes it, who owns it, and who buys it — and this month supplied an answer to each.

There is a counter-reading, and it should be stated rather than suppressed. Demand running “well above supply,” a book-to-bill above 2, and prices rising are also the classic markers of a market in which buyers double-order to secure allocation, and in which today’s order book overstates tomorrow’s consumption. Chery’s remark that the business is tied to the top five hyperscalers is a defence against exactly that concern — the largest cloud operators are the buyers least likely to be ordering speculatively — but it is a defence, not a proof. The industrial phase is real; whether it is durable at this rate of growth is the question the next two quarters will answer.

What is undisclosed, and why it matters

An honest accounting of what is not on the record is part of the story. STMicro does not disclose a silicon photonics revenue figure, and the $1.3 billion cannot be decomposed into photonics, electronic ICs, microcontrollers, and power solutions. No customer or module vendor is named for PIC100. No wafer-start or yield figure is given for the Crolles line, and the “more than quadruple” capacity expansion has no disclosed baseline. No average selling price, margin figure, or order value is attached to any photonics product, and the 2027 figure is a goal, not a contracted backlog.

Each omission has a plausible innocent explanation. Customer names are confidential in this market; capacity baselines are commercially sensitive; a forecast is not a contract. But the pattern of omissions also means the most interesting question — how much of the $1.3 billion is actually photonics, and how much is the microcontrollers and power products riding alongside it — is unanswerable from the public record. Chery’s statement that demand is driven mainly by silicon photonic ICs is the only bridge across that gap — a qualitative statement from the chief executive, not a number. Do not convert it into a photonics revenue estimate.

What to watch

The signals that will move this story from forecast to evidence are specific. First, the fourth quarter: whether STMicro reports revenue at or above the $4.15 billion midpoint, and whether communication equipment and computer peripherals grow again at anything like the 64% year-over-year rate of the third quarter. Second, a photonics split: any disclosure that separates silicon photonics revenue from the broader AI data-centre total. Third, the capacity marker: disclosure of Crolles wafer starts, ramp timing, or a dated milestone on the quadrupling plan. Fourth, a customer disclosure: a hyperscaler or module vendor naming PIC100 in a product or a system. Fifth, the PIC100 TSV step: whether the through-silicon-via platform reaches production on a cadence that suggests the pluggable business is funding the packaged-optics roadmap, rather than the roadmap remaining a slide.

The fair summary is a narrow one. On October 29, STMicroelectronics raised its 2026 AI data-centre revenue forecast to about $1.3 billion, up from more than $1 billion in July, and reiterated a goal of well above $2 billion for 2027, after third-quarter revenue of $3.76 billion and a 37.3% gross margin. CEO Jean-Marc Chery said demand was driven mainly by the company’s silicon photonics chips, and its PIC100 platform — in high-volume production on 300 mm wafers in Crolles for 800G and 1.6T modules since March — is set for a more-than-fourfold capacity expansion by 2027. The figures are forecasts, not booked sales, and the photonics share of the total is undisclosed. What is confirmed is adjacent and substantial: a book-to-bill above 2 and a backlog above $20 billion. The lane’s industrial phase now has an incumbent’s number on it. Whether the number holds is the next story.

Sources and further reading

Reuters (Nathan Vifflin and Fabrizio La Rocca), “STMicro lifts data centre revenue forecast after Q3 profit beat,” October 29, 2026 — 2026 AI data-centre revenue forecast raised to about $1.3 billion from a July forecast of more than $1 billion; 2027 goal of well above $2 billion reiterated; Q3 revenue $3.76 billion; gross margin 37.3%, up from 33.2% a year earlier, expected to reach about 39% in Q4; operating income excluding one-off charges nearly doubled to $429 million, beating the $400.5 million expected by analysts polled by LSEG; Q4 revenue guidance of $4.15 billion, plus or minus 3.5%, above analysts’ estimate of $4.10 billion; Chery: demand driven mainly by silicon photonics chips; data-centre business “much more related to the top five hyperscalers than the neo-cloud business”; supply planning three to four years ahead; prices rising on selected products.

STMicroelectronics, third-quarter 2026 results and earnings call, October 29, 2026 — net revenues $3.76 billion, up 17.9% year over year; gross margin 37.3%, up 410 basis points year over year; non-U.S. GAAP diluted EPS $0.37; communication equipment and computer peripherals revenue up 64% year over year and 14% sequentially; RF & Optical Communication segment up 46.2% year over year; industrial up 40%, automotive up 5%, personal electronics down 4%; book-to-bill above 2; backlog above $20 billion; AI data-centre demand driven mainly by silicon photonic ICs, electronic ICs, microcontrollers, and power solutions; 2026 AI data-centre revenue expectation of about $1.3 billion; full-year revenue implied at approximately $14.5 billion; intermediate gross margin target of 44% to 46% reaffirmed.

STMicroelectronics, “STMicroelectronics Enters High-Volume Production Of Its Industry-Leading Silicon Photonics Platform To Support AI Infrastructure Demand,” March 2026 — PIC100 technology in 300 mm high-volume production for leading hyperscalers; plans to more than quadruple production capacity by 2027 and further expand in 2028; 800G and 1.6T PIC100 transceivers; PIC100 TSV platform roadmap; manufacturing in the Crolles, France 300 mm fab.

STMicroelectronics, PIC100 announcement, February 20, 2025 — PIC100 as the company’s first silicon photonics technology, enabling 200 Gbps per lane and 800 Gbps / 1.6 Tbps optical modules; silicon photonics and BiCMOS technologies manufactured in the Crolles 300 mm fab; AWS quoted as collaborating on the technology. This publication’s account of the Marvell–Lumentum agreement (November 2, 2026), the Tower–Innolight production milestone (November 3, 2026), and the reported UMC photonic IC orders (November 2, 2026) is drawn from its own prior coverage. The $1.3 billion and well-above-$2 billion figures are management forecasts and guidance, not booked sales, and no silicon photonics revenue figure is asserted here, because none is disclosed. This article is analysis, not investment advice.