Article: OLIX just closed a $312 million Series B round, pushing its valuation to $3.3 billion. The money will fund the first shipments of its inference chip, which eliminates High Bandwidth Memory (HBM) and could flip AI-hardware economics.

Why the chip matters

Inference chips run the models behind chatbots, recommendation engines and video analytics. Most of today’s chips lean on HBM—a fast but pricey, power-hungry memory stack placed next to the processor. OLIX’s design throws HBM away, routing data through on-chip resources and cheaper memory. If the architecture lives up to its claims, AI service providers could slash both capital outlay and power bills while keeping throughput steady.

The funding backdrop

  • Lead investor: Fundomo (New York)
  • Other backers: Arm, Hudson River Trading, the UK government’s Sovereign AI venture fund
  • Notable participant: Reed Hastings, co-founder of Netflix
  • Board addition: Stanford professor Nick McKeown

The round ranks among Europe’s biggest semiconductor raises and follows OLIX’s 2024 launch in London. The company aims to ship its first customer units in the second half of 2027.

Stakes for the AI hardware market

  • Potential winners: Cloud providers and edge operators that replace HBM-laden accelerators with a cheaper, lower-power alternative.
  • Risks: HBM still dominates the highest-throughput workloads. Skipping it may cap performance, forcing OLIX to focus on a narrower set of use cases.
  • Cost implications: Dropping HBM could dramatically cut bill-of-materials and cooling needs, lowering total-cost-of-ownership for AI workloads.

What’s still unclear

No benchmark data have been released, so real-world speed-per-watt and latency numbers remain speculative. Competitors might roll out their own memory-light architectures, and supply-chain constraints could delay volume production.

What to watch

  • Performance validation: Independent testing of the chip’s inference speed and power draw.
  • Customer uptake: Which AI service providers sign up before the 2027 delivery window.
  • Ecosystem support: Compatibility of tooling and software stacks with existing AI frameworks.

If OLIX proves an HBM-free chip can meet modern model performance demands, it could force a rethink of AI accelerator design and reshape cost structures across the industry. The next few months of data will reveal whether the promise lives up to the hefty valuation backing it.