The Optical Circuit Switching Boom: One Market Is Real, One Isn't There Yet

Read a market-size headline in optics this season and you will see the same shape over and over: a niche networking technology compounding at an eye-watering rate. SNS Insider puts optical circuit switching at $0.40 billion in 2025, rising to $4.93 billion by 2035 at a 28.51% CAGR. The number is real, and it is not cherry-picked — MarketsandMarkets independently lands on the same ~28.5% annual rate, off a slightly larger base, to $2.52 billion by 2032.

The problem is not the arithmetic. It is what the headline implies: that a large, fast-growing market exists today for investors to buy. It does not. What exists is a small, one-customer market that a research model compounds into a large market a decade out, and whether that forecast ever becomes revenue depends on a second market that has not yet shipped a single unit. The two are not the same forecast, and conflating them is where the money gets lost.

The 28.5% is a forecast, not a going-concern number

The CAGR is a compounding extrapolation, and it compounds off a floor that is close to nothing. Before the current AI build-out, the optical circuit switch — a device whose tiny mirrors redirect light between fibers without converting it to electrical signals — was largely a lab-automation and telecom niche measured in the low tens of millions of dollars annually. Industry research firm Cignal AI puts today's total addressable market at "multiple billions" only because of the AI ramp now underway. Remove that ramp and the base that the 28.51% multiplies is essentially the revenue of a single product line inside a couple of companies.

Table 1 shows how wide the map of "this market" already is. The three OCS-specific forecasts disagree with each other by 2030 even though two of them share a growth rate, and a separate, broader "optical switches" market — which several headlines merge with OCS — is a different, slower, much larger sector.

Source	Market	Base	Endpoint	CAGR
SNS Insider	Optical circuit switching	$0.40B (2025)	$4.93B (2035)	28.51%
MarketsandMarkets	Optical circuit switches	$0.56B (2026)	$2.52B (2032)	28.5%
Cignal AI	Optical circuit switching	—	$8B+ (2030)	—
SNS Insider	Optical switches (broader)	$7.05B (2025)	$20.60B (2035)	11.36%

Cignal AI, the firm doing the most detailed bottom-up work, is also the most aggressive: it raised its OCS forecast to more than $8 billion by 2030 from $2.5 billion by 2029 as of last December. Its reason for the jump is worth reading closely — the increase is "driven almost entirely by new visibility into the Scale Up (GPU) opportunity." That is a polite way of saying the biggest part of the forecast is not asking investors to believe in a product that exists. It is asking them to believe in a product that has not shipped.

Two markets, and only one has revenue

Semiconductors repeatedly split into two markets that share a label but not an economics. Optical circuit switching has split the same way, and treating them as one number is the core analytical error.

Market one — scale-out, TPU — is the only volume application and is real. Google built it. The company has deployed tens of thousands of OCS ports into its spine layers and AI cluster reconfiguration, spent more than $1 billion doing it, and reports roughly 40% power savings, a 30% cost cut, and markedly better uptime versus electrical switching. Google's incentive to promote its own build is also structural: OCS keeps light in the optical domain, sidestepping the power and heat of electrical packet switches that each convert light to electricity and back.

But market one has exactly one customer. That has two consequences for the suppliers who want the business. The first is that Google is migrating from internally built OCS to commercial suppliers, which is the multi-hundred-million-dollar opening Lumentum and Coherent are chasing. The second is that the entire near-term revenue of the industry sits in one buyer's hands, and that buyer built the switch itself for years. This is a land-conversion event, not a greenfield market — the value is in winning the volume the incumbent already makes internally.

Market two — scale-up, GPU — is the part of the forecast that does the heavy lifting and does not yet exist. Cignal AI expects it to contribute significantly only in 2028/2029, eventually rivaling and exceeding the scale-out market, on the back of Nvidia's planned integration of OCS into its upcoming NVL576 architecture. The forward condition here is the sharpest one in the piece: Nvidia is already slipping its electrical scale-up roadmap — its Kyber rack system was pushed to 2028 on manufacturing difficulties — and the skeptics argue that electrical switching (NVLink, NVSwitch) may keep optical circuit switching a niche "patch panel" suited to static configurations, not the dynamic fabrics the largest clusters need. If Nvidia's OCS never reaches volume, nearly all of Cignal's upgrade sits on a sub-market that did not ship.

The suppliers are priced like the boom already happened

For a retail investor the decision is not whether OCS is real technology. It is whether the forecast is already in the price — and for the two merchant suppliers that stand to capture the bottom-up volume, it substantially is.

Lumentum is the purest OCS exposure, guided by management to roughly $100 million a quarter by the end of 2026 and positioned to take about half the market. Its fiscal 2026 fourth quarter delivered $1.01 billion of revenue, up 109% year over year, and it calls OCS one of the AI-driven growth drivers "only beginning to contribute." The market has noticed: the stock trades at roughly 29 times trailing sales and about 109 times trailing EV/EBITDA, up sharply from a $144 low within the past year. Coherent, the other merchant OCS supplier working on the interim 2.4T "Lite" configuration with Google, carries a $60 billion valuation at roughly 67 times trailing EV/EBITDA. Neither is priced for a technology that fails to scale; both are priced for the scale-up market of 2028 arriving on schedule.

The discipline for a stock like this is the same as for any market-size headline that compounds off a small base: the forecast is not revenue, and the multiples already assume the optimistic scenario. The numbers that would distinguish the real boom from the model are specific and observable. Does Google shift its in-house OCS volume to third-party suppliers, and does Lumentum's ~$100 million-a-quarter run-rate materialize by the end of this year? Does Nvidia's NVL576 actually use OCS at volume in 2028, or does the delay that already hit Kyber extend to the optical switch? The first two are land-conversion facts that would show up in reported revenue. The third is Cignal AI's own admitted "matter of opinion, not fact." The 28.51% CAGR is back-loaded by design — the money is not in believing the headline, it is in waiting for one of those checkpoints to prove the second market is real.