Breaking

Ricursive Founders Say AI-Designed Chips Could Broaden Recursive Self-Improvement Race

October 2, 2026 at 10:08 EDT

Foundation Models
Infra & Chips
AI Agents

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini are arguing for a broader view of recursive self-improvement: not a single AI lab abruptly winning, but a feedback loop in which models help design better chips and those chips make stronger models possible. In an AI Deep Dive episode published on September 28, 2026, the pair framed chip design as one of the hardest bottlenecks in AI progress and said automating more of it could accelerate the hardware cycle behind frontier systems.

September 2026 · Ricursive Intelligence

AI chips enter the self-improvement loop

Ricursive is arguing that recursive self-improvement may depend on a hardware-software cycle: AI helps design better chips, and those chips make stronger AI systems cheaper and faster to build.

$335M

Reported backing

$4B

Reported valuation

~10x

Performance per cost target

weeks

Desired design cycle

The economic claim, drawn to scale

Baseline architecture

~10x

Model-specific chip architecture

The wager is that workload-specific silicon can sharply improve performance per total cost of ownership.

How the loop is supposed to compound

1. Model

AI proposes chip-design choices.

→

2. Silicon

Better hardware supports stronger AI.

→

3. Learning

Each design informs the next attempt.

Why engineers care

Advanced chips move on slower timelines than frontier models. Automating design, verification and learning across chips could reduce a major bottleneck.

Why caution remains

Earlier successes centered on floorplanning, while full chip development still depends on physics, process rules, verification and governance.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.