Light, the new compute substrate.
Why photonic computing matters now, who's building it, and where the capital is concentrating.
The thesis
For seven decades, transistor density doubled on schedule. Moore's law was a budget item. Today, copper wires connecting GPUs draw more power than the GPUs themselves; data center networks are the bottleneck, not the compute. The economy of every additional NVIDIA cluster is shaped by the speed and energy cost of moving data between chips, not the speed of computing on them.
Photonic computing changes the substrate. Photons don't dissipate energy in motion the way electrons do, don't generate heat at scale, and naturally support certain operations (notably matrix multiply via interference) that take billions of transistors to emulate. A handful of startups and a few quiet incumbents are now shipping silicon that does compute and interconnect with light, not electrons.
The four buckets
Pure-play photonic compute. Q.ANT, Lightmatter, Celestial AI, Lightelligence. Building processors whose multiply-accumulate units are made of light. Smaller in revenue, largest in upside, highest in technical risk.
Optical interconnect. Ayar Labs, nEye Systems, and the I/O chiplet vendors. Photonic plumbing between GPUs. Earlier to revenue than pure-play compute and strategic to NVIDIA, AMD, and Intel — all three of which are shareholders in Ayar Labs.
Optical infrastructure. Coherent (COHR), Lumentum (LITE), Marvell (MRVL), IPG, nLight. The picks-and-shovels: lasers, transceivers, DSPs, EML modules. Real revenue today, locked supply contracts with NVIDIA, sensitive to hyperscaler capex cycles.
Photonic foundries. GlobalFoundries (GF Fotonix), TSMC (COUPE), AMF Singapore (now part of GFS). The wafer fab capacity that everything else depends on. Geopolitically strategic — silicon photonics manufactured outside China is suddenly a national security category.
The catalyst calendar
Lightmatter is the bellwether IPO. $4.4B private valuation, $850M raised, Google still re-investing as of Series D in late 2025. A filing on EDGAR — which the IPO Watch on this terminal is configured to detect — would set the tone for every other photonic name's multiple.
Ayar Labs is the next likely public listing. Strategic premium because of the AMD/Intel/NVIDIA cap table; the kind of company that a competitor could buy outright before it ever IPOs.
Q.ANT — the smallest in raised capital but the most validated in technology — sits at $80M Series A with Druckenmiller's family office now on the cap table. A 2027 Series B is likely; an IPO is years out. But every Q.ANT customer announcement reads through to QUBT, which uses the same TFLN material system on its own Tempe fab.
The risk
Photonics has been five years away from breaking out for two decades. Most pure-play names trade at 50–200x sales because they have essentially no sales. Power is concentrated: NVIDIA has locked EML laser capacity at Lumentum through 2027, foreclosing new entrants on the supply side. The market for photonic compute is real but the architecture winner is not yet clear — silicon photonics has ecosystem advantages, TFLN has physics advantages, all-optical has the highest ceiling and the longest timeline.
For a portfolio: optical infrastructure is the lower-risk way to participate, the pure-play names are sized for asymmetric upside, and the IPO calendar is where the multiples will reset.
What this terminal does
Tracks the public universe with daily quotes and rule-based catalyst scoring tuned to photonic-specific events (PFIC, S-1, TFLN, DARPA HARQ, tape-out, design win). Watches SEC EDGAR for IPO filings on the seven private companies most likely to file in 2026–2027. Routes high-conviction catalysts through SwarmCast — a six-agent panel where the Outsider archetype's dissent is the most important signal when the panel doesn't converge.