Photonic light sources
Across the whole photonics stack, the layer that actually captures value is the LIGHT SOURCE (lasers-on-silicon, frequency combs, external laser modules), not the modulator or the passive PIC.
- Bear
- Open
- $72B by 2035
The callThe laser is the scarce part of the optical build-out, and NVIDIA has already bought the incumbents' capacity; newcomers wait for on-wafer sources after 2028.
- D1AI clusters move from copper to light, and every GPU-to-GPU and switch link needs a laser, so laser volume runs far ahead of the general laser market.
- D2Co-packaged optics keeps the laser off the hot switch die as a separately sourced part, which is what lets a merchant supplier exist.
- R1NVIDIA put $4B into Coherent and Lumentum with priority capacity, which probably locks the laser array design through Feynman in 2028.
- R2The private companies with the best shot are already late stage.
- R3If volume optical engines integrate the laser on the die by 2028, the discrete module stops being a market.
- C1A Rubin or Kyber-class co-packaged optics generation qualifies a merchant laser supplier beyond Coherent and Lumentum, opening about $500M of the market.2028
- 1By mid-2027 volume CPO / optical-engine deployments (NVIDIA / Broadcom-class) still source laser modules from merchant suppliers rather than vertically integrating.Jun 2027 · open
- 2Volume co-packaged optics still buys laser modules from merchant suppliers rather than making their own.Jun 2027 · open
- 3Lumentum pulls back toward Coherent's valuation, which would change the public-market read.Dec 2027 · open
- 4Through 2028 external/merchant light sources remain a SEPARATELY-SOURCED layer in volume AI-datacom optical engines, rather than being monolithically integrated onto the switch/compute die.Dec 2028 · open
- 5Teardowns of 1.6T-plus optical engines show the light source still separately sourced, or integrated on the die.Dec 2028 · open
- AAOIApplied Optoelectronics
- cheap
- CIENCiena
- fair
- COHRCoherent Corp
- cheap
- FNFabrinet
- cheap
- IPGPIPG Photonics
- full
- LASRnLIGHT
- full
- LBIRD.PALumibird
- full
- LITELumentum
- cheap
- MTSIMACOM
- full
Private · 7 private companies, 3 of them in Netherlands.
CScale · Photon IP · EDWATEC · Elytra · Xscape Photonics · Photon IP (likely now Photon Bridge) · Superlight Photonics
8 more private companies tracked, not shown
$86M raised by 4 companies, $159M of it in the last three years. Wave building: Money is arriving: recent rounds and tier-1 funds are in, but no runaway leader yet.
Backed by a strategic investor: Xscape Photonics, Superlight Photonics.
- $81MXscape Photonics
- Founded 2022
- $5MPhoton IP (likely now Photon Bridge)
- Founded 2020
- ·EDWATEC
- Founded 2023
- ·Superlight Photonics
- Founded 2020
- 16 Oct 2026Lumentum's CEO says he ships a third less than customers want and is sold out through 2027; a forecaster says the shortage clears this year. One of them is wrong by December.
- 28 Jul 2026Even the modulator, the part silicon supposedly commoditised, is drifting off silicon at the leading edge. The list of things photonics must borrow other materials for is growing, and that is where the profit pools.
- 21 Jul 2026The light source turns out to be a quarter to half the cost of an optical link, confirmed by two separate companies. That explains why Nvidia paid four billion dollars to lock up its laser supply. The open question is whether the incumbents it bought use an approach that stops working as channel counts climb.
4 earlier
- 19 Jul 2026The laser shortage is the most quantified bottleneck in AI infrastructure right now, and it has been for a year. Nvidia spent about $4bn locking up supply rather than building it. That tells you where the value sits and also why the listed names are the clean way to own it.
- 11 Jul 2026The standalone laser is dying as a category in two different markets at once — datacentre interconnects AND AR smart glasses. In both, value flows away from the discrete laser module to either the gain-medium maker (who even the giants keep buying from) or the integration platform that pulls the laser onto the chip. AR absorbs even harder than datacom because the OEM wants the whole device in-house, which is why an AR light-engine startup is structurally an acquisition, not a scale-up.
- 23 Jun 2026High-power lasers are a durable standalone category where the laser is the product sold into a fragmented end market (industrial, defense, medical, ultrafast); they are an absorption/acquisition base rate where the laser is a component inside a hyperscaler optical engine (datacom/CPO). The device-manufacturing layer stays merchant even for the largest integrators; the discrete module gets pulled on-wafer and value re-pools into the integration IP. A laser pitched at AI interconnects is the most absorbable possible framing.
- 17 Jun 2026A heterogeneous-laser-integration company confirmed every major XPU and switch player is now blocked on putting the laser (and now the amplifier) on-chip, especially for multi-wavelength. The merchant discrete-laser module is a bridge; the durable value is the wafer-level integration process foundries adopt.