Celestial AI, which aims to use light to speed up data transfer inside servers, raised $250M led by Fidelity at a $2.5B valuation, taking its funding to $515M
- BlackRock and AMD Ventures participated in $250 million round — Startups are trying to develop ways to speed up AI computing
Context & Ripple Effects
Celestial AI had already progressed from a $56M Series A for its photonic architecture to a $100M Series B for optical interconnect technology and then a $175M Series C. The new round materially extends that financing arc and lifts cumulative funding to $515M.
The company is funded into a competitive optical-computing field: Lightmatter previously raised a $154M Series C backed by Fidelity and GV and later reached a $4.4B valuation. Fidelity's lead role therefore connects Celestial AI to a financier already active in the category.
First-order effects
- Celestial AI gains $250M to continue developing and commercializing optical links between compute and memory inside servers, with a $2.5B valuation setting the reference point for this financing.
- Fidelity becomes the round leader, while BlackRock and AMD Ventures add institutional and strategic backing to Celestial AI's investor base.
Second-order effects
- Lightmatter and other optical-computing rivals face a better-capitalized Celestial AI, increasing pressure to demonstrate that light-based interconnects can move from technical promise into server deployments.
- AMD Ventures' participation aligns a chip-industry investor with Celestial AI's interconnect approach, making the startup more relevant to hardware partners evaluating ways to reduce data-movement constraints.
Third-order effects
- If comparable funding continues, optical interconnects could become a distinct, heavily financed layer of AI-server infrastructure rather than a peripheral component bet.
- The widening valuations and repeat participation by large financial investors suggest that AI infrastructure finance is reaching beyond processors to the data-transfer bottlenecks surrounding them; commercial adoption remains the key test.
The trend: AI infrastructure investors are increasingly funding specialized hardware that targets the movement of data between compute and memory, not just the compute chips themselves.