Celestial AI, whose photonic-based architecture scales across multichip systems for AI workloads, raises a $56M Series A at a $150M post-money valuation
Kyle Wiggers / VentureBeat :
Context & Ripple Effects
This is the opening move of what became one of the steepest funding arcs in photonic computing: Celestial AI's $250M round led by Fidelity at a $2.5B valuation in March 2025 traces directly back to this $56M Series A at just $150M post-money — roughly a 16x valuation climb across three follow-on rounds (the $100M Series B and a $175M Series C in between).
The thesis investors were underwriting was that copper interconnects cap how far multichip AI systems can scale, and light can lift that ceiling. A month after this raise, Luminous Computing pulled in a larger $105M Series A for its own light-based AI accelerator, confirming photonics had become a funded category rather than a one-company bet.
First-order effects
- Celestial AI gains the capital to push its photonic architecture beyond single-chip demos toward multichip systems, competing for AI workloads where data movement between processors and memory — not raw compute — is the bottleneck.
- At a $150M post-money, early backers buy into an unproven optical-interconnect approach while it is still cheap relative to the silicon-accelerator crowd like Deep Vision, whose edge-inference chips raised at comparable stages with a different architectural bet.
Second-order effects
- Luminous Computing's $105M Series A weeks later shows rivals' fundraises feeding off each other: each photonic round de-risks the category for the next investor, forcing both companies to differentiate on whether light accelerates computation itself or just moves data between chips.
- If optical interconnects prove out, incumbent accelerator vendors face pressure to open their chip-to-chip interfaces to third-party photonic links rather than keep scaling monolithic dies.
Third-order effects
- AMD Ventures' participation in Celestial AI's later round points toward a structural endgame where photonics stops being a startup niche and becomes a supply-chain layer absorbed by the major AI-compute platforms.
- Sustained mega-rounds for optical interconnect suggest AI infrastructure capital is bifurcating: compute itself stays concentrated in a few giants, while the interconnect and memory-movement layers around them attract platform-scale bets on startups.
The trend: Photonics is graduating from speculative chip startup territory to a core, separately capitalized layer of AI infrastructure, with interconnect specialists raising ever-larger rounds as multichip scaling hits electrical limits.