TULL

Investor and former Legendary Entertainment CEO Thomas Tull

Featured image by China Photos/Getty Images

AI

Thomas Tull’s USIT leads $175M round into cost-saving startup Celestial AI

Thomas Tull’s U.S Innovative Technology Fund has led a $175 million Series C into Celestial AI, a photonics chip startup aiming to bring AI computing costs down.

In a move highlighting investor appetite for solutions to the extreme computing and energy demands of AI, photonic interconnect startup Celestial AI has raised a $175 million Series C led by US Innovative Technology Fund, a VC firm led by investor and Legendary Entertainment founder Thomas Tull.

The Santa Clara-based startup raised a $100 million Series B less than a year ago from investors including Temasek, Koch Disruptive Technologies and Samsung Catalyst.

The investment underscores Tull’s budding leadership in AI deals. The firm led a $320 million Series C into AI GPU cloud computing startup Lambda in February. The startup is attempting to aggressively undercut costs and compete with big tech cloud providers like AWS and Google Cloud.

USIT primarily invests in defense tech, leading a $200 million investment last year in Shield AI, which is using AI to automate military aircraft.

Celestial AI uses light and lasers to transmit data within computer systems. The startup claims that light photonics reduces energy costs and speeds up data transmission relative to traditional silicon-based wiring.

Startups like Celestial AI are betting on the increased need for faster computing solutions. Many startups and large language model providers have struggled to find the necessary computing power to fill their needs.

Other photonics startups have likewise raised large sums. Boston-based Lightmatter raised a $155 million Series C-2 in December led by GV. Ayar Labs added $25 million to its Series C last May, bringing its total raise to $155 million from investors including Intel Capital and Nvidia.

Eliyan, which enables chips to be combined in ways that squeeze out more performance, just this week raised a $60 million Series B from Tiger Global and Samsung Catalyst.

Energy demands for AI have skyrocketed as models have gained traction. Independent estimates and studies have suggested that training a large language model like OpenAI’s GPT-3 could consume 1,300 megawatts of electricity, roughly the amount 130 US homes use in a year. That figure is only one estimate: By 2027 the entire AI sector could use 85 to 134 terawatts of power, about what the Netherlands utilizes annually, according to an estimate by data scientist Alex de Vries in the journal Cell.


Investor and former Legendary Entertainment CEO Thomas Tull

Featured image by China Photos/Getty Images

Learn about our editorial standards.