LIVE
+++ Amsterdam's Wonderful lands $550M at $5B valuation +++ Pentagon Adds ChatGPT and Grok to GenAI.mil +++ US government backs OpenAI's fair-use case against the NYT +++ Runway's Solaris generates app interfaces in real time +++ OpenAI Delays Astra Model After Hugging Face Hack +++ OpenAI hit with 30 new suits over Tumbler Ridge attack ++++++ Amsterdam's Wonderful lands $550M at $5B valuation +++ Pentagon Adds ChatGPT and Grok to GenAI.mil +++ US government backs OpenAI's fair-use case against the NYT +++ Runway's Solaris generates app interfaces in real time +++ OpenAI Delays Astra Model After Hugging Face Hack +++ OpenAI hit with 30 new suits over Tumbler Ridge attack +++
All news ›
AI IN LIFE AI IN LIFENEWS
DAILY
DE EN
CHIPS

Google to unveil two custom AI chips a year, not one

The AI infrastructure chief says the two-year generation cycle is over; Google also wants 60 percent more space in Taiwan.

Google to unveil two custom AI chips a year, not one

Symbolic image: a robotic arm places a large accelerator chip onto a circuit board on an assembly line, with a worker seen from behind at the control station.

Google plans to introduce two of its own AI chips per year instead of one new generation every two years, and to move even faster after that.

At a glance

  • Amin Vahdat, Google senior vice president for AI infrastructure, spoke in Taipei on September 2, 2026, at Semicon Taiwan.
  • The cadence shifts from one generation every two years to two chips per year, with a faster pace signaled afterward.
  • Google plans 60 percent more development space in Taiwan and higher research spending there.
  • Not established: chip names, manufacturing partners, investment figures and a start date for the new schedule.

Google is moving its in-house AI chips to a twice-a-year release rhythm, replacing the roughly two-year gap between generations, and wants to go faster still. Amin Vahdat, the company's senior vice president for AI infrastructure, said so in Taipei on September 2, 2026, according to Nikkei Asia.

A cadence change, not a product launch

What Vahdat described is a schedule, not a new chip. Instead of one generation every two years, Google aims to put out two chips a year, with an even quicker pace after that. He framed the change as a response to the race for leadership in artificial intelligence.

Taiwan gets more room

The second half of the announcement concerns Taiwan, where Google intends to expand the space it uses for development by 60 percent and to spend more on research there. The island hosts the contract manufacturers and advanced packaging plants the chip industry depends on. Neither a headcount nor a completion date was given.

What is not established

Only the opening of the Nikkei Asia report is readable without a subscription, so several details cannot be confirmed here and are deliberately left out:

  • which chip families or generations are affected
  • manufacturing and packaging partners
  • the size of the Taiwan expansion in money terms
  • when the new cadence begins

It is also unclear whether the faster schedule covers Google's full accelerator lineup or only part of it.

◈ AI-GENERATED REPORT · SOURCES LINKED

FAQ

How often will Google release new AI chips?

Two chips per year, replacing the previous pattern of one generation every two years, with an even faster pace signaled for later.

Who announced the faster chip schedule?

Amin Vahdat, Google senior vice president for AI infrastructure, in a keynote at Semicon Taiwan in Taipei on September 2, 2026.

What is Google planning in Taiwan?

A 60 percent increase in the space it uses for development plus higher research spending. No headcount or investment figure was given.