Direct answer
Two of the researchers behind AlphaChip — the AI that once designed a chip in a shape that baffled its own engineers — have started a new company, Ricursive Intelligence, to let AI design chips end to end. Unveiled at TechCrunch Disrupt this week, the startup’s pitch is simple: designing a chip should take a week, not years, and the resulting chips could help solve the looming compute crunch.
What happened
Founders Anna Goldie and Azalia Mirhoseini took the stage at TechCrunch Disrupt to detail what Ricursive Intelligence is building: AI that designs chips, learns from each iteration, and improves subsequent designs — automating and accelerating the whole chip-design process.
Their argument starts with a stark number. Goldie told the audience that compute demand is expected to grow “at least 10,000x” in the next five years — a figure she described as potentially conservative. Building 10,000 times more data centres isn’t the answer, she said. Instead, chips need to be more efficient: using data-centre bandwidth better, using wafers and fabs better.
The technical approach is what makes this interesting. Traditional chip design partitions a chip into dozens of smaller blocks, each optimised separately. Ricursive’s AI optimises the entire chip at once, discovering unconventional layouts humans wouldn’t attempt. The precedent is AlphaChip, the founders’ earlier system, which produced a circular “donut-shaped” arrangement of chip components that initially surprised engineers but ultimately worked.
Goldie sketched the payoff: if designing a chip takes six months to two years, you have to be extremely selective about which chips get designed. But if it takes a week, custom chips for niche applications — space, AR/VR, even biological computing — become financially viable. Humans, she said, bring the ideas and creative thinking; execution gets handled by AI.
Source: TechCrunch
What it means for UK small businesses
You won’t be designing chips. But you absolutely pay for compute — in the price of AI tools, cloud bills, and the performance of the software you depend on. If AI-designed chips deliver even a fraction of what’s promised, the downstream effect is cheaper, faster AI services for everyone, including small firms.
The nearer-term lesson is about where the AI race is moving: the bottleneck is shifting from models to hardware. We’ve covered the money flowing into chip and data-centre infrastructure (AI data-centre funding, Sunmmio’s $277M chip raise). Businesses that treat AI as a line item in their costs should watch this space — efficiency gains upstream are the most reliable route to lower AI prices downstream.
There’s also a strategy point for founders: Ricursive’s whole premise is that AI handles execution while humans keep the creative work. That division of labour — humans on ideas, AI on grind — is exactly how small teams should be thinking about their own AI adoption right now.
Honest limits / what to watch
- It’s a stage pitch, not a product. Ricursive detailed its approach at Disrupt but announced no chips, no customers, no timelines. Treat this as a direction, not a delivery.
- Fabs are still the bottleneck. Designing a chip in a week is one thing; manufacturing it still means booking time at a fab, with all the cost and queueing that involves. AI design doesn’t remove the physical constraint.
- The 10,000x demand figure is a forecast, not a fact. It’s the founders’ estimate, made to justify their approach. Directionally the demand growth is real — but treat round numbers from founders selling a vision with appropriate scepticism.
FAQs
What is Ricursive Intelligence?
A new startup founded by Anna Goldie and Azalia Mirhoseini that is building AI systems to design computer chips — automating and accelerating a process that currently takes months or years.
Who are the founders behind Ricursive Intelligence?
Anna Goldie and Azalia Mirhoseini, the researchers who previously developed AlphaChip, the AI system known for producing a circular “donut-shaped” chip layout that surprised engineers but worked.
How does AI-designed chip design work?
Instead of engineers partitioning a chip into dozens of blocks and optimizing each one, the AI optimizes the entire chip at once, learns from each iteration, and produces unconventional layouts — including wave-like patterns — humans wouldn’t typically try.
Why does AI chip design matter for the compute crisis?
The founders estimate compute demand will grow at least 10,000x in the next five years. Rather than building 10,000x more data centres, more efficient chips would squeeze far more performance from existing hardware, bandwidth and fabs.
When will AI-designed chips be available?
It’s early days — the startup detailed its approach on stage at TechCrunch Disrupt but hasn’t announced products or customers. Manufacturing (fabs) remains a bottleneck AI alone can’t remove.
