Full-Time
Posted on 6/28/2025
Photonic Fabric enables memory disaggregation
$175k - $215k/yr
Toronto, ON, Canada + 2 more
More locations: Santa Clara, CA, USA | Hillsboro, OR, USA
In Person
Celestial AI provides Photonic Fabric™, an optical memory interconnect for hyperscale data centers, to enable memory sharing and reduce DRAM needs. It moves memory data with light through optical interconnects, cutting DRAM requirements by up to 35% and enabling HBM disaggregation over optics instead of PCIe. Unlike traditional electronic interconnects, it targets an end-to-end optical memory solution that supports multi-tenant memory pooling and higher bandwidth in large clouds. Its goal is to deliver scalable, cost-efficient memory infrastructure to support Generative AI and other demanding workloads.
Company Size
51-200
Company Stage
Acquired
Total Funding
$3.8B
Headquarters
Sunnyvale, California
Founded
2020
Help us improve and share your feedback! Did you find this helpful?
Health Insurance
Vision Insurance
Dental Insurance
Life Insurance
Company Equity
Wilson Sonsini Goodrich & Rosati, JSA, and Latham & Watkins advised on US company Marvell Technology’s USD3.25 billion acquisition of...
Niobium expands leadership team with key hires in Product and Finance as it builds toward an encrypted cloud platform. March 10, 2026/in Niobium Company adds accomplished veterans from Groq and Celestial AI as it advances a vision for cloud computing where data never needs to be decrypted. DAYTON, Ohio - March 10, 2026 9:00 AM EST - Niobium, a leader in hardware acceleration for fully homomorphic encryption (FHE), today announced the addition of two senior executives to its leadership team: John Barrus as Vice President of Product, formerly of Groq, and Evan Scott as Vice President of Finance, formerly of Celestial AI. The appointments come as Niobium reaches a major milestone in its FHE Accelerator development program, a platform that could fundamentally change how sensitive data is processed in the cloud, enabling organizations to run computations on encrypted data without ever exposing it. The company recently announced a manufacturing partnership with SEMIFIVE and Samsung Foundry to bring the world's first commercially viable FHE accelerator ASIC to production, designed to enable encrypted computation at speeds practical for real-world applications. John Barrus joins Niobium from Groq. As Groq's VP of Product, Barrus helped build one of the most talked-about AI inference companies in the industry, which has been recently acquired by NVIDIA. Before Groq, Barrus held leadership roles at Amazon and Google, where he helped bring Google's Cloud TPU to market, a platform that has since grown into a multi-billion-dollar business. "When I first engaged with Groq in 2018, it was obvious what the team should be building and why it would matter," said Barrus. "I have the same conviction about Niobium. The hardware opportunity in FHE is real, the team is exceptional, and the timing is right." Evan Scott joins Niobium from Celestial AI, where he served as Senior Director of Finance and led financial planning and analysis, and managed investor reporting during a period of rapid growth in AI infrastructure, including the company's $250 million Series C-1 financing. Scott also helped guide the company through a major strategic transaction culminating in its acquisition by Marvell. His experience building financial infrastructure and managing investor relations at a fast-moving semiconductor startup gives him a strong foundation as Niobium advances toward commercial deployment and scales its operations. "What stood out to me immediately was the strength of the leadership team and the caliber of engineers building this technology," said Scott. "Niobium is tackling one of the hardest problems in computing - making encrypted computation practical - and the opportunity to help bring that to market was incredibly compelling." Building the CUDA of encrypted computation. Today, organizations that move sensitive workloads to the cloud must trust that their data will be protected during processing when the data is decrypted and exposed. This creates risk across industries from healthcare and financial services to government and AI, and it remains one of the biggest unresolved challenges in cloud computing. Niobium is addressing the two barriers that currently hinder broad FHE adoption: performance and usability. While the industry has long recognized the computational demands of FHE, Niobium's approach goes beyond raw acceleration. The company is developing higher-level software frameworks designed to make encrypted computing accessible to a far broader developer community, mirroring the role CUDA played in democratizing GPU-accelerated computing for NVIDIA. "If you want to know where next-generation computing is headed, follow the talent. John saw the AI inference wave early at Groq. Evan helped navigate Celestial AI through a landmark acquisition. The fact that they're both here tells a story about where encrypted computation is going," said Kevin Yoder, Niobium CEO. "We're building more than just an accelerator. We're building the platform that makes encrypted computation something that developers can actually use. John and Evan have done this before, at scale, and that's exactly what this stage demands." About Niobium. Niobium is building the first dedicated hardware platform designed to advance fully homomorphic encryption (FHE) into commercial applications. FHE keeps data encrypted even during computation, mathematically guaranteeing privacy. Niobium is headquartered in Dayton, Ohio, with offices in Portland, Oregon, and San Francisco, California. More information is available at niobiummicrosystems.com.
Marvell Industry Analyst Day: Celestial AI lands a Hyperscaler Win. Celestial AI CEO Highlights Major Hyperscaler Win and Photonic Fabric Adoption at Marvell Industry Analyst Day Celestial AI CEO Dave Lazovsky used Marvell's Industry Analyst Day to spotlight the company's most significant commercial milestone to date, announcing that a top-tier hyperscaler has selected Celestial AI's photonic connectivity for its next-generation AI processors, with co-designed technology now moving into volume manufacturing. Lazovsky said this engagement represents one of the industry's earliest large-scale transitions from copper to optical XPU interconnect inside AI systems. Lazovsky told analysts that AI workloads are evolving faster than the data-center infrastructure designed to support them. With compute no longer the bottleneck, he said hyperscalers now face acute constraints around bandwidth, latency, power, and cost. Next-generation fabrics require sub-200-nanosecond link latency, at least a 4 - 5x reduction in power consumption, and a cost structure of roughly five to eight cents per gig per side. Celestial AI's Photonic Fabric, he said, uniquely matches these requirements while enabling switch-class bandwidth at the XPU level - 16 Tbps per direction per chiplet, or 64 Tbps bidirectional in a single package. He emphasized that the hyperscaler deployment provides a first-mover advantage as system designers shift from electrical traces to optical connectivity earlier in the architecture. Customer volumes for chiplet attach can reach hundreds of thousands of units per quarter, creating a substantial near-term revenue engine, while Celestial AI's optical HBM-pooling design allows memory to be extended tens of meters away from compute while maintaining the latency budgets required for large-scale training workloads. - Celestial AI's Photonic Fabric spans chiplet-level XPU connectivity and optical memory interconnect for HBM pooling. "The timing of this combination couldn't be better," Lazovsky said. "Latency translates directly to revenue for our customers, and Photonic Fabric delivers the bandwidth density, energy efficiency, and cost structure required for AI at scale." Earlier this month: Marvell agreed to acquire Celestial AI in a definitive deal valued at approximately $3.25 billion upfront, with contingent equity that could lift the total above $5 billion based on revenue milestones. The acquisition targets Celestial AI's Photonic Fabric optical interconnect platform, which delivers 16 Tbps chiplets designed for next-generation scale-up fabrics connecting XPUs across packages, systems, and racks. Marvell expects the business to begin contributing revenue in the second half of fiscal 2028, reaching a $500 million annualized run rate by the fourth quarter of fiscal 2028 and $1 billion by the fourth quarter of fiscal 2029. The deal positions Marvell to extend its connectivity portfolio - including Ethernet switching, DCI optics, custom silicon, and its UALink-based scale-up roadmap - into the short-reach optical domain that AI systems increasingly require as copper reaches limits in bandwidth, latency, power, and thermal reach. Celestial AI's Photonic Fabric chiplets integrate electrical and optical components in a compact form factor and can be co-packaged directly with XPUs and scale-up switches, enabling optical links to originate from the center of the die and freeing die-edge area for additional HBM capacity. The company reports active engagements with multiple hyperscalers preparing multi-rack AI clusters that will transition high-speed XPU links from copper to optics. Under the agreement, Marvell will pay $1 billion in cash and issue approximately 27.2 million shares valued at $2.25 billion based on the 10-day volume-weighted average price. Earn-outs totaling another ~27.2 million shares will be paid if Celestial AI achieves at least $500 million in cumulative revenue by the end of Marvell's fiscal 2029 (one-third of the earn-out), and up to $2 billion in cumulative revenue by the same date (full earn-out). Closing is expected in the first quarter of calendar 2026, subject to regulatory review.
Marvell acquires Celestial AI for $3.25B as photonics becomes the next big frontier in AI chip design. Marvell Technology is making a major move into next-gen chip architecture with a $3.25 billion acquisition of Silicon Valley startup Celestial AI, the company announced Tuesday. The deal comes at a time when every major chipmaker is racing to develop faster, more energy-efficient ways to connect AI processors in large-scale data centers, and Marvell is placing its bet squarely on photonics. "Under terms of the deal, Celestial AI will receive $1 billion in cash and 27.2 million shares of Marvell common stock worth $2.25 billion," Reuters reported. Celestial AI, founded in 2020 by Preet Virk, David Lazovsky (CEO), and Michelle Tomasko, has spent the past five years developing optical interconnects that use light instead of electrical signals to move data between AI chips and memory. It raised $250 million earlier this year in a Series C1 round to speed up development, positioning itself as one of the more ambitious startups building technology specifically for the AI hardware supply chain. Celestial AI, backed by AMD Ventures, is developing a bridge technology using a unique form of photonics. This approach stands apart from competitors and aims to enhance energy efficiency and reduce latency. Its flagship product, the Photonic Fabric, is designed for AI computing systems and acts as the backbone for data center computing, networking, and memory. The technology offers significant improvements in system performance and energy efficiency, aiming to support sustainable and profitable AI advancements. Marvell's interest is straightforward. The surge in generative AI has pushed cloud providers to overhaul their data center plans, and demand for faster chip-to-chip communication continues to rise. Companies such as Amazon and Microsoft already rely on Marvell to build custom silicon for their in-house AI chips. Adding Celestial's photonics work gives Marvell a new path to improve bandwidth and efficiency at the infrastructure layer, where performance bottlenecks have become increasingly visible. CEO Matt Murphy said the company expects total revenue of around $10 billion next fiscal year, with data center sales rising 25 percent. He told analysts the Celestial acquisition opens a new addressable market worth roughly $10 billion on its own. "We're going to have a silicon photonics powerhouse at Marvell when this is all done," he said. Marvell's $3.25B bet: Celestial AI's photonics tech to power next-gen data centers. The deal gives Marvell access to Celestial's optical fabric, which competes with Broadcom's and Nvidia's efforts. Big cloud companies are expected to begin installing photonics-based systems around 2027 or 2028 for high-volume workloads, and Murphy believes adoption will scale quickly after that. Marvell plans to fold Celestial's technology into its next-generation photonics-related infrastructure products, setting the stage for what it hopes will become a long runway of revenue expansion. Under the terms of the acquisition, Celestial AI will receive $1 billion in cash and 27.2 million Marvell shares valued at $2.25 billion. Marvell expects the deal to close in early 2026, with meaningful revenue contributions beginning in the second half of fiscal 2028. The company projects an annualized run rate of $500 million by the end of that year, doubling to $1 billion by late fiscal 2029. As part of the broader partnership, Marvell issued Amazon a warrant that allows Amazon to purchase up to $90 million of Marvell stock based on its purchases of photonic fabric products through 2030. The warrant covers roughly one million shares at an exercise price near $87. Marvell also delivered a bullish near-term forecast. The company expects fourth-quarter revenue of around $2.2 billion, slightly above analyst expectations compiled by LSEG. Third-quarter revenue rose 36.8 percent to $2.07 billion. Even so, shares have slipped more than 15 percent this year as competition intensifies across custom AI silicon and networking. For Marvell, the Celestial AI acquisition signals a clear direction: the next phase of AI hardware will be shaped by photonics, and it wants to be one of the companies defining that shift.
Marvell Technology is in advanced talks to acquire Celestial AI for over $5 billion, involving cash and shares. The acquisition aims to integrate Celestial's photonic technology, which enhances data flow speed in AI systems, into Marvell's portfolio. This move could make Marvell more attractive to cloud providers by offering a combined electrical and optical solution. Celestial AI has raised $250 million in venture capital, totaling $515 million in funding.