2026 · 06 · 08 · Mon

Updates for 6/8

Anthropic hired OpenAI's #2 custom chip engineer, a bet on owning inference silicon. Perplexity now lets AI write its search code, cutting token costs 85%.

A · Theme of the day

Before the IPOs, Anthropic poaches OpenAI's chip engineer

The race for inference silicon turned into a talent war ahead of both IPOs.

OpenAI loses its #2 chip engineer to Anthropic

GPT (OpenAI)

What changed

Lost Clive Chan — self-described second engineer on its custom chip program — to Anthropic (The Decoder), a key departure as both labs head toward IPOs.

Compared to before

Until last month, OpenAI led in custom silicon via its Broadcom partnership, with Clive Chan — a Tesla Autopilot ASIC veteran — near the center of the program.

Why it matters

Chip design is a long-term margin story for IPO investors, so losing a key architect dents OpenAI's hardware narrative. Nothing changes for GPT API users today.

Source: openai.com　·　→ View on chaos map

Anthropic hires the engineer who built OpenAI's chip program

Claude (Anthropic)

What changed

Hired Clive Chan, self-described second engineer on OpenAI's chip program (The Decoder). Reports say Anthropic is weighing building its own AI chips.

Compared to before

Anthropic has run entirely on cloud GPUs from AWS and Google, while OpenAI's Broadcom deal gave it a clear path to owning inference hardware.

Why it matters

If Anthropic builds chips, Claude API pricing could fall over two to four years — relevant for large workloads. For most developers it's background context for now.

Source: anthropic.com　·　→ View on chaos map

B · Theme of the day

Perplexity rewrites its search engine in real time — with AI

Perplexity's AI now writes its own search routines, cutting token costs 85%.

AI now writes Perplexity's search code on the fly — 85% cheaper

Perplexity

What changed

Replaced fixed search APIs with "Search as Code" (The Decoder): models write Python search routines in a sandbox, beating OpenAI and Anthropic on benchmarks.

Compared to before

Perplexity's search ran on a fixed set of API calls — the same pipeline for every query — making it hard to balance accuracy against cost per query.

Why it matters

Complex queries should feel noticeably sharper for daily users, and API customers should watch for pricing changes. It still doesn't cover Maps, Shopping, or ad data.

Source: perplexity.ai　·　→ View on chaos map

Past updates

A daily archive of changes actually applied to the site.