2026-06-07 — views · STMicroelectronics

STMicroelectronics Doubles Its 2026 Data-Center Revenue Target to $1B on AI Photonics Demand

On June 2, 2026, STMicroelectronics roughly doubled its 2026 data-center revenue ambition to about $1 billion (from "nicely above $500 million") and said 2027 could double again, citing AI-infrastructure demand and a faster PIC100 silicon-photonics ramp. Shares rose more than 8%

What happened

On June 2, 2026, STMicroelectronics (NYSE: STM; Euronext Paris/Milan) raised its revenue ambition for data centers, telling investors the segment is now expected to reach “about $1 billion in 2026” versus a prior outlook of “nicely above $500 million.” The company added that, “assuming the current dynamic continues and with the current engagements we have, revenues could double in 2027,” up from a previous framing of “well above $1 billion.” In short, the 2026 target roughly doubled and the 2027 bar moved meaningfully higher in a single update.

The market reaction was sharp for a large-cap analog and embedded-chip maker not usually grouped with the AI-GPU names: shares closed more than 8% higher in Paris trading on June 2, and the stock had already risen more than 164% year-to-date through the prior close, per Yahoo Finance.

Why this is an AI-capex story, not a generic chip update

ST is best known for microcontrollers, power devices, and sensors that go into cars, phones (it counts Apple and Tesla among customers), and industrial gear. The new growth lever is different: optical and power content sold into AI data centers. Jefferies analysts estimate that optical products should account for roughly two-thirds of the expected 2027 data-center growth, with power semiconductors contributing the remaining third.

The optical piece centers on PIC100, ST’s silicon-photonics platform. The company entered high-volume production around March 9, 2026, on 300mm lines, with the platform supporting 800 Gb/s and 1.6 Tb/s optical modules used to move data between GPUs, CPUs, and storage in large AI clusters. ST said it plans to quadruple PIC100 capacity by 2027, backed by long-term supply commitments from hyperscale customers, and developed the platform in collaboration with Amazon Web Services. A follow-on PIC100 TSV variant targets near-packaged and co-packaged optics. Separately, ST disclosed a March 2026 collaboration with Nvidia on Physical AI platforms.

The numbers

Metric	Prior outlook	New outlook (June 2, 2026)
2026 data-center revenue	”nicely above $500 million”	about $1 billion
2027 data-center revenue	”well above $1 billion”	could double vs 2026
PIC100 module speeds	n/a	800 Gb/s and 1.6 Tb/s
PIC100 capacity plan	n/a	quadruple by 2027
STM share move, June 2	n/a	up more than 8% (Paris)

Practitioner note

Treat the dollar figures as ambitions, not formal segment guidance: ST framed 2027 with conditional language (“could double,” “assuming the current dynamic continues”), and the company has not historically broken out data center as a standalone reported segment, so the path from ~$1B to a “double” depends on hyperscaler co-packaged-optics adoption timing. The optical-versus-power split is a Jefferies estimate, not a company disclosure. Note also that one retail-data aggregator reported a much larger intraday pop (around +15%); the better-attested figure from mainstream financial coverage is “more than 8%” in Paris, so size any read on the move conservatively.

Under-considered angle

The interesting tell here is who is doing the buying. ST’s photonics ramp is anchored by a co-development with AWS and long-term hyperscaler commitments, which makes this less a bet on Nvidia’s GPU cycle and more a bet on the interconnect and optics layer of AI build-outs — the plumbing between accelerators. As 800G transitions toward 1.6T and co-packaged optics moves the laser closer to the switch ASIC, the value migrates from merchant-GPU vendors toward silicon-photonics and optical-module suppliers. ST’s jump suggests the market is starting to price a broader “AI optics” supply chain (transceivers, PICs, lasers, power delivery) rather than treating AI exposure as synonymous with GPU compute alone — a rotation worth watching across analog, optical, and power-semi names that rarely show up in AI headlines.