← All companies

Chips / Company profile

Celestial AI

What Celestial AI does

Celestial AI (celestial.ai) developed the Photonic Fabric, an optical interconnect technology platform aimed at improving data movement for AI systems—especially within racks, systems, and packages where electrical “beachfront” constraints can limit bandwidth, latency, and energy efficiency. The company’s core premise is that large AI workloads are increasingly constrained by memory capacity/bandwidth and data movement rather than compute, motivating an architecture that routes optical links “any point on the silicon” to the point of consumption, with claims of substantially higher bandwidth density and lower latency versus co-packaged optics approaches. Over time, Celestial AI positioned Photonic Fabric as a foundation for scale-up connectivity and disaggregated memory/compute designs, offering technology licensing as well as (at least in earlier stages) an Orion AI accelerator product path.

As of 2026-02-02, Celestial AI was acquired by Marvell Technology, Inc. and integrated into Marvell’s Data Center Group. Marvell’s acquisition communications describe Celestial AI’s Photonic Fabric chiplet approach as an all-optical scale-up interconnect enabling high bandwidth and reduced latency, including claims about a 16 Tb/s per chiplet figure and the ability to be co-packaged with custom XPUs and scale-up switches for large-scale deployments. Marvell also states it expects revenue contributions from Celestial AI starting in the second half of fiscal 2028, with ramp targets communicated in acquisition-related materials.

News

Company record

Feb 02, 2026 · Official · Marvell Technology, Inc. (Investor Relations)Marvell completes acquisition of Celestial AI

Marvell announced it completed its previously announced acquisition of Celestial AI and said Celestial AI’s Photonic Fabric technology would be part of Marvell’s Data Center Group; Marvell also reiterated financial impact expectations and revenue ramp targets.

Jan 21, 2026 · Official · Federal Trade CommissionFTC early termination granted for Marvell/Celestial AI transaction

The FTC’s early termination notice lists the transaction between Marvell Technology, Inc. (acquiring party) and Celestial AI, Inc. (acquired party) and shows granting status.

Dec 02, 2025 · Official · Marvell Technology, Inc. (Investor Relations)Marvell to acquire Celestial AI to accelerate scale-up connectivity

Marvell announced a definitive agreement to acquire Celestial AI, describing Photonic Fabric as a breakthrough optical scale-up interconnect and detailing transaction structure, valuation of upfront consideration, and anticipated timing of closing.

Aug 13, 2025 · Reporting · Finance Yahoo (Business Wire syndication)Celestial AI closes final Series C1 round ($255M)

Celestial AI announced the final close of its $255 million Series C1 funding round and added VentureTech Alliance and Samsung Catalyst Fund, while listing a large investor base.

Feb 04, 2022 · Official · Business WireCelestial AI raises $56M Series A funding led by Koch Disruptive Technologies

Celestial AI announced its $56 million Series A investment led by Koch Disruptive Technologies with participation from Temasek’s Xora Innovation fund, The Engine, Tyche Partners, M-Ventures, IMEC XPand, and Fitz Gate.

Source map · 2 recurring channels · 27 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuationNot disclosed
Tracked capital$311M2 sourced rounds

What it builds

Photonic Fabric™

Photonic Fabric is Celestial AI’s optical interconnect technology platform designed to enable optical compute-to-compute and compute-to-memory links for disaggregated AI compute and memory clusters, including claims that it can deliver optical connectivity beyond traditional silicon package “beachfront” constraints and to the point of consumption.

source ↗
Photonic Fabric Appliance

Photonic Fabric Appliance is positioned (in Celestial AI-related Optica/OFC materials) as a pathway to bypass bounded memory-to-compute ratios in GPUs, aiming for improvements in throughput, energy efficiency, and latency for disaggregated memory and compute designs.

source ↗

Milestones & partnerships