Positron Raises $230M Series B: Memory Chip Startup Challenges Nvidia Dominance

by

$230M Investment Led by Qatar Sovereign Fund

  • Positron raised $230 million in Series B funding
  • Qatar Investment Authority (QIA) led the round
  • Claims equivalent performance to Nvidia H100 with 66% less power

What Happened?

AI chip startup Positron raised $230 million in its Series B round.[TechCrunch] Qatar Investment Authority led this round. Founded in 2023, this Nevada-based startup previously raised $51.6 million in Series A last year, bringing total funding to over $300 million.[VentureBeat]

Positron’s core weapon is high-speed memory chips. They targeted the memory bandwidth bottleneck in AI inference workloads. According to the company, their Atlas system currently on sale achieves 93% memory bandwidth utilization. This contrasts sharply with typical GPUs that hover around 10-30%.[VentureBeat]

Why Does It Matter?

Frankly, there have been many startups claiming to challenge Nvidia. Groq, Cerebras, SambaNova, and more. But what makes Positron different is their approach.

While most competitors emphasize compute power, Positron focused on memory. They targeted the fact that the compute-to-memory ratio in transformer model inference is nearly 1:1. Theoretically, this makes sense.

What I find more noteworthy is Qatar’s involvement. Qatar launched state-owned AI company QAI last December and announced a $20 billion AI infrastructure initiative with Brookfield.[Semafor] This aligns with Middle Eastern countries’ efforts to reduce Nvidia dependence.

There are real customers too. Cloudflare and Parasail are conducting long-term tests with Atlas.[Gulf Times]

What Comes Next?

Positron will use this funding to accelerate development of their next-gen chip, Asimov. The Titan system featuring this chip is scheduled for release in 2026. It will pack 2TB of memory per accelerator, capable of running models up to 16 trillion parameters on a single system.[Gulf Times]

However, there are realistic challenges. The current Atlas is FPGA-based, making it more expensive than general-purpose ASICs. Real competition becomes possible only when Asimov ships on time. And the question is whether they can close the performance gap with Nvidia Blackwell already on the market.

Frequently Asked Questions (FAQ)

Q: Is Positron’s chip really better than Nvidia?

A: For inference workloads specifically, they claim 3.5x performance per dollar and 66% lower power consumption compared to Nvidia H100. However, this is company benchmark data. Nvidia still dominates training. Since inference and training have different requirements, the choice depends on use case.

Q: Why is Qatar investing in AI chips?

A: Middle Eastern countries are pursuing AI sovereignty. Qatar announced $20 billion in AI infrastructure investment, and securing Nvidia alternatives is strategically important. US chip export regulations are accelerating this movement.

Q: Can I buy Positron chips now?

A: The Atlas system is currently supplied to select cloud companies. General enterprise sales remain limited. The next-gen Titan system is scheduled for 2026, so if considering large-scale adoption, it may be worth waiting.

If you found this article useful, please subscribe to AI Digester.

References

Leave a Comment