Artificial intelligence (AI) chipmaker Luminous Computing nabbed $105 million in a Series A funding round led by a handful of big names, including Gigafund, Bill Gates, and 8090 Partners.

Founded in 2018, the startup aims to develop a supercomputer that can realize AI applications the likes of which have only ever been seen on the silver screen.

“Artificial intelligence is able to do things that used to be exclusively in the domain of human experts. AI can now write poetry, write functional code, write speeches that sound like JFK,” Luminous Computing CEO Marcus Gomez said. “But, there’s this gap in capability.”

Gomez explained that while AI may be able to play chess or go, Siri doesn’t actually run on your phone, and your car still has a steering wheel. “We’re not at Star Trek yet.”

“Artificial intelligence was supposed to be the foundational tool for moving the needle on the survival and progress of the species, and it’s not yet that,” he added.

What frustrates Gomez is the barrier to realizing this futuristic vision of AI isn’t the compute or even the software.

“We spent three years really doing a deep dive on how these AI systems are built to really understand where the bottlenecks are,” he said. “It turns out the bottleneck isn’t on computation. It’s on communication.”

The problem with existing compute architectures, according to Gomez, is most of the die space available for compute is occupied by large caches to store the data until it’s ready to be processed and the physical interfaces that keep those caches full.

“Why do we end up in a situation where most of the chip whose purpose in life is doing computations is not compute? It's because you don't have the bandwidth going into that chip to feed it,” he said.

And that’s the problem Luminous aims to solve.

However, Gomez emphasizes that Luminous isn’t just trying to solve the connectivity problem.

“Just because we are building a better interconnect, it doesn't mean we're a pure interconnect company,” he said. “We do build the whole system, including all the compute chips, and the switches, and all the software, and we package the whole thing together.”

In fact, the chipmaker’s end goal is to have a rack-mount system that can outperform Nvidia’s DGX servers by a wide margin available for large hyperscale customers within 24 months.

The funding round includes contributions from more than half a dozen investors. These include the aforementioned plus Neo, Third Kind Venture Capital, Alumni Ventures Group, Strawberry Creek Ventures, Horsely Bridge, and Modern Venture Partners.

“If we can build the product AI customers have always dreamed of, but wasn’t possible, Luminous will dominate the future of computing,” Luke Nosek, managing partner at Gigafund, said in a statement.

The funding will enable Luminous to double its engineering team, accelerate the development of custom silicon and software, and ultimately bring its first products to market. High on the chipmaker’s list of priorities is to grow its silicon photonics expertise, something Gomez says will be critical to eliminating the data movement bottlenecks facing large-scale AI today.