Theses›Datacentre & AI compute›Optical circuit switching
Optical circuit switching
Standalone optical circuit switching fails to reach venture scale through 2030 (a zero-success base rate over 20 years, outside Nvidia/Google-scale captive buyers)
- Bear
- Closing
- $25B by 2035
The callStandalone optical switches have never reached venture scale in 20 years; the limit is timing re-lock, so the value goes to the light source and sync layer, and the switch itself makes little.
- D1AI clusters head to hundreds of thousands of accelerators wired as one machine, and every electrical hop converts light to electrons and back, burning power.
- D2Optics is replacing copper inside GPU scale-up networks, NVIDIA's above all, on top of Google's TPU fabric, so the buyer set is widening.
- R1No independent switch company has scaled in 20 years: Polatis and Calient both sold for undisclosed sums, by all accounts under $100M.
- R2Google's switch is captive, Lumentum and Coherent absorb the merchant side, and incumbents likely take 60-75% of the re-sync pool, leaving $100-160M.
- R3A full switch system to hyperscaler qualification is capital-heavy: nEye at $152M raised, Oriole looking for about £150M.
- C1A named optical circuit switch shipping in volume inside an NVIDIA scale-up reference design, making this a multi-buyer market beyond Google.2027-28
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
- 1Whether a passive all-optical switch shows contention resolution in hardware; I put it at about 65% that it does not.Dec 2027 · open
- 2The re-synchronisation reframe is right: the architecture-level players that attack transceiver re-synchronisation out-progress the pure switch-medium players by end-2028, more capital raised AND more hyperscaler engagement.Dec 2028 · open
- 3Timing-architecture players out-raising and out-winning the switch-medium players; about 60%.Dec 2028 · open
- 4The zero-success base rate holds to 2030: no independent standalone-OCS company reaches a >$1B venture-scale outcome as a stay-independent business.Dec 2030 · open
- ·Lumentum's OCS number, the easiest running check on whether buyers beyond Google are taking volume.
Expressionslisted, private, technologies
- AMATApplied Materials
- cheap
- GOOGLGoogle
- cheap
- MSFTMicrosoft
- full
- NVDANVIDIA
- cheap
- TSEMTower Semiconductor
- expensive
- LITELumentum
- cheap
- COHRCoherent
- cheap
- HUBN.SWHUBER+SUHNER (Polatis)
- not screened
Private · 19 private companies, 13 of them in the US.
Texas Instruments DLP · Samsung Display · Pacific Light Technologies · Neurophos · Meadowlark Optics · Lumotive · Lightmatter · Jasper Display · Holoeye Photonics · Forth Dimension Displays · Envisics · Boulder Nonlinear Systems · nEye.ai (nEye Systems) · CALIENT Technologies · Polatis (HUBER+SUHNER Polatis) · Telescent · Google OCS (Apollo / Palomar) · nEye Systems (nEye.ai) · Corespan Systems (formerly Drut Technologies)
17 more private companies tracked, not shown
Who has already bet on itprivate rounds, from public sources
$152M raised by 2 companies, $393M of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.
Backed by a top-tier venture firm: nEye Systems (nEye.ai). Backed by a strategic investor: nEye Systems (nEye.ai).
- $152MnEye Systems (nEye.ai)
- Founded 2020
- ·Corespan Systems (formerly Drut Technologies)
- Founded 2018
What changedthe record, newest first
- 15 Oct 2026Two European optical-switching startups each raised $125 million in six weeks. In twenty years no company selling only an optical switch has become a large independent business, and investors are now betting heavily that this time is different. The money proves appetite, not outcomes: one of the two names no customer, and neither has said how it handles the timing re-lock that has held the category back.
- 11 Aug 2026A hyperscaler measured a 2x communication-time win from fast optical switching, told its own leadership, and did nothing, because the product that would let it act does not exist. The gap between a problem being known and being fixed is where the entrant lives, and it is usually mistaken for the problem not mattering.
- 28 Jul 2026The optical-switching team that measured how long receivers take to re-lock after a path change has now said plainly that the fix has no patent protection yet, and that making it work on production network cards requires the incumbent chip vendors to expose firmware hooks. The reframe still looks right; owning it looks harder than ever. And their next-generation switch engine turns out to be a merchant part from a large US chipmaker, which anyone can buy — another sign the value in this category pools in control software and endpoints, not the switching medium.
6 earlier
- 27 Jul 2026A third optical-switching team has independently concluded that the thing holding the category back is not how fast the switch moves but how long the receiver takes to re-lock afterwards, and has put a measured number on it: 1.8 milliseconds against roughly 5.7 seconds for standard equipment. When three teams converge on the same reframe, the interesting question stops being whether the reframe is right and becomes whether anyone can own it, or whether the big network-chip vendors simply absorb the fix.
- 21 Jul 2026An optical networking founder explains why the pitch is power rather than speed: the electricity a network burns grows with the square of how many chips you connect, and almost all of it goes to the switches. He also draws the line the category keeps blurring, between fixed high-speed links and networks that can re-route on the fly.
- 21 Jul 2026The founder of one passive optical switching company explains why almost everyone else builds circuit switches rather than packet switches: their switching methods take seconds to reconfigure, so the network has to be designed around fixed heavy-traffic routes. He also gives a concrete answer to the timing problem that has sunk this category before, which is worth checking properly.
- 23 Jun 2026The optical-circuit-switching market keeps getting bigger as Google's AI buildout pulls demand (one forecast revised up 40% in a quarter), and a listed supplier now guides to ~half that market by end-2026, but the standalone-switch startups still face a 20-year zero-success record; the durable value sits with the listed buyers and suppliers and, on the venture side, the laser and timing layer underneath.
- 18 Jun 2026The case for all-optical packet switching just got sharper: time-division circuit switches may cap out as line rates hit 1.6T, the first specific reason the harder packet approach could be the one that scales, though the piece that makes it work is still unproven in hardware.
- 14 Jun 2026In optical networking for AI, almost every startup is building essentially the same optical switch and competing on the switching medium; the more interesting early-stage bet may be the picks-and-shovels layer they all depend on (the tunable light source, and the timing/clock that lets fast switching work at all).