Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI
Training Data
20 HOURS AGO
Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI
Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI
Training Data
20 HOURS AGO
Google AI infrastructure chief Amin Vahdat explains why frontier-scale computing is an exercise in workload forecasting, system reliability, power planning, and hardware–model co-design—not simply buying more FLOPs.
Amin Vahdat describes how Google builds infrastructure for rapidly evolving AI workloads. He contrasts theoretical chip speed with goodput under frequent failures, explains why model and software advances often drive more efficiency than hardware, and traces the TPU’s evolution from a risky inference bet to a flexible accelerator platform. The conversation then explores co-design with DeepMind, agent-driven demand for CPUs and storage, optical networking, gigawatt-scale power constraints, dedicated serving clusters, open standards, and the possibility of multi-megawatt racks in orbit by 2036.
00:02
00:02
Specialization is a bet on the future
04:13
04:13
AI racks reshape the building itself
09:01
09:01
Goodput, not theoretical FLOPS, is what matters
16:35
16:35
The TPU began as a contrarian bet
23:37
23:37
Co-design unlocks huge optimization opportunities
29:05
29:05
Changing the chip architecture mid-flight
35:06
35:06
Agents remove the human rate limit
39:17
39:17
Optical switching leaves the bits untouched
42:59
42:59
Power is the fundamental constraint
49:51
49:51
Inference cannot rely on retired training clusters alone
1:02:44
1:02:44
The 2036 rack may consume multiple megawatts
