AI has a compute problem.
We are reliant on compute architecture that was developed for gaming GPUs three decades ago, in a market that is capital-intensive, technically unforgiving, and strategically critical. For a long time, customers were willing to live with the trade-offs because there was no real alternative.
Recent AI progress has been driven by ever-larger pre-training runs, but we’ve now entered the inference era, where inference-time compute scaling and reinforcement learning training unlock capabilities that once would have required years of pre-training alone.
Inference at scale requires both high interactivity and high throughput.
Despite the different needs, training and inference now compete for the same scarce resource, and cost, scale, and energy are all under extreme pressure. Current GPU-centric hardware is approaching physical limits and cannot provide fast inference for every user simultaneously. Even as GPU packages grow larger and HBM generations advance, we are running into hard physical limits on how far this architecture can be pushed.
Paradigm shifts don’t come from incremental tweaks to the old regime, and solving the inference challenge is no different. This is where OLIX, founded by James Dacombe, comes in.
Rather than pushing a single chip to do everything, OLIX is building specialised silicon and an optical network to connect it, unrolling models across racks the way an assembly line works. It's a genuinely different approach to the industry's supply chain constraints, and one built on components that are actually available at scale.
We are proud to be among the first institutional investors in 2024. Since then, OLIX has raised 250 million dollars in financing, including a recent 220 million dollar round led by Hummingbird Ventures at a $1bn valuation. That level of conviction from the market reflects the sheer scale of the opportunity OLIX is going after.
If you’re excited about redefining the future of compute, OLIX is hiring in London, Austin, San Francisco, Toronto, and Bristol!