Celestial AI raises $100 million for utilizing light-based interconnects to transfer data
Celestial AI announced it had secured $100 million in a Series B round headed by IAG Capital Partners, Koch Disruptive Technologies, and Temasek’s Xora Innovation fund. CEO Lazovsky said the tranche would increase Celestial AI’s photonics level by increasing the company’s engineering, income, and method trading divisions. This takes the total amount raised by Celestial AI to more than $165 million.
Technologies at Celestial AI
Celestial AI, a machine learning (ML) accelerator business, has created a patented technological platform for the next generation of high-performance computing solutions. Celestial AI, led by founder and CEO Dave Lazovsky, provides scalability for independent computation and memory resources
through a range of Photonic Fabric technology solutions. The Photonic FabricTM, an optical connection technology platform for computation and memory, was developed by Celestial AI. To unlock developments in AI with sustainable and successful business models, Photonic Fabric delivers the core technology for optically scalable, disaggregated data center computing and memory. The technology is compatible with industry standards, including JEDEC HBM, PCIe, UCIe, CXL, and PCI Express. It is also compatible with non-standard electrical communication lines. With the addition of several chip systems, Celestial takes advantage of photonics for data transmission. With the use of memory and computing resources, it enables efficient information movement inside and across chips for AI and other applications. The headquarters of Celestial AI is in Santa Clara, California, and the company also has branches in Hyderabad, India, Orange County, and Toronto.
According to Celestial, their method boosts microprocessor efficiency by lowering the power required for data transport. Their photonics technology provides 25 times more bandwidth and ten times less latency and power consumption than optical alternatives while aligning with industry communication standards. Computing and memory are now intimately linked. Adding more computation is the only method to increase high bandwidth memory, regardless of whether the extra computation is necessary, Lazovsky acknowledged in an email. According to Lazovsky, “Our photonics technology is truly distinct, one-of-a-kind, and has superior characteristics. Demand for optical connection is rising quickly to enable the shift from general computing data center infrastructure to accelerated computing, given the surge in generative AI workloads brought on by LLMs and the pressures it places on existing data center architectures”.
Several years ago, David Lazovsky and Preet Virk, two researchers with semiconductor engineering and photonics backgrounds, realized together that workloads for AI and machine learning would fast run into a “data movement” problem. They projected that if AI models became more prominent than what could be stored on the die of any one memory chip, it would become increasingly difficult to transfer data to and from computational gear. Their answer, designed by Phil Winterbottom, a former researcher at the prestigious Bell Labs, was an optical interconnect technique for transmitting data from chip to chip, memory to chip, and compute to compute. Lazovsky, Virk, and Winterbottom established Celestial AI as a firm to market the technology. The startup is now drawing significant funding and becoming increasingly popular daily.