All-hands meeting coinciding with GA of Ironwood release reveals extent of AI demand
Google will have to double its "serving capacity" for AI every six months to meet the demand for AI, Google Cloud VP Amin Vahdat has reportedly said.
CNBC reported Vahdat's comments, which were made in a presentation during an all-hands meeting on November 6, the same day the Ironwood TPU became generally available. In the presentation, seen by CNBC, a slide covering AI compute demand said "now we must double every 6 months.... the next 1000x in 4-5 years."
Ironwood SuperPod– Google Cloud
Vahdat added: “The competition in AI infrastructure is the most critical and also the most expensive part of the AI race." But he noted that while they will invest, they aren't interested in trying to "outspend the competition."
Google needs to “be able to deliver 1,000 times more capability, compute, storage networking for essentially the same cost and increasingly, the same power, the same energy level,” Vahdat said. “It won’t be easy, but through collaboration and co-design, we’re going to get there.”
During Google's latest earnings call, the company raised its capex guidance for the year for the second time. At the start of the year, estimates were for around $75bn in spend, rising to $85bn in Q2, and now peaking at between $91bn and $93bn. Of that, the "vast majority" went on technical infrastructure - 60 percent on servers, and 40 percent on data centers and networking equipment.
Since the November 6 meeting, Google has announced plans to invest $40bn in cloud and AI infrastructure in Texas, $2bn in a data center in Turkey with Turkcell, and $6.38bn in expanding data center and office space in Germany.
In addition to increasing its infrastructure, Vahdat said that the company will continue to improve its models and utilize its custom silicon to improve its capabilities.
The latest iteration of Google's Tensor Processing Units (TPUs) – dubbed Ironwood – officially became generally available to customers on November 6. The chips can be scaled up to 9,216 chips in a single superpod. Ironwood's specs were first revealed in April 2025, offering 192GB of 7.4Tbps of high-bandwidth memory (HBM) per chip, a 6x increase on Google’s sixth-generation Trillium TPU, in addition to 1.2Tbps bidirectional Inter-Chip Interconnect (ICI) bandwidth, a 1.5x improvement over Trillium. Anthropic has already signed on to use up to a million TPUs in a deal worth tens of billions of dollars.
Google's CEO, Sundar Pichai, also addressed concerns about an "AI bubble" during the meeting, reported CNBC.
An employee asked how the company is protecting itself against a potential "burst," to which Pichai said: “I think it’s always difficult during these moments because the risk of underinvesting is pretty high. I actually think for how extraordinary the cloud numbers were, those numbers would have been much better if we had more compute.”
He added that "there will no doubt be ups and downs."
Google will have to double its "serving capacity" for AI every six months to meet the demand for AI, Google Cloud VP Amin Vahdat has reportedly said.
CNBC reported Vahdat's comments, which were made in a presentation during an all-hands meeting on November 6, the same day the Ironwood TPU became generally available. In the presentation, seen by CNBC, a slide covering AI compute demand said "now we must double every 6 months.... the next 1000x in 4-5 years."
Ironwood SuperPod– Google Cloud
Vahdat added: “The competition in AI infrastructure is the most critical and also the most expensive part of the AI race." But he noted that while they will invest, they aren't interested in trying to "outspend the competition."
Google needs to “be able to deliver 1,000 times more capability, compute, storage networking for essentially the same cost and increasingly, the same power, the same energy level,” Vahdat said. “It won’t be easy, but through collaboration and co-design, we’re going to get there.”
During Google's latest earnings call, the company raised its capex guidance for the year for the second time. At the start of the year, estimates were for around $75bn in spend, rising to $85bn in Q2, and now peaking at between $91bn and $93bn. Of that, the "vast majority" went on technical infrastructure - 60 percent on servers, and 40 percent on data centers and networking equipment.
Since the November 6 meeting, Google has announced plans to invest $40bn in cloud and AI infrastructure in Texas, $2bn in a data center in Turkey with Turkcell, and $6.38bn in expanding data center and office space in Germany.
In addition to increasing its infrastructure, Vahdat said that the company will continue to improve its models and utilize its custom silicon to improve its capabilities.
The latest iteration of Google's Tensor Processing Units (TPUs) – dubbed Ironwood – officially became generally available to customers on November 6. The chips can be scaled up to 9,216 chips in a single superpod. Ironwood's specs were first revealed in April 2025, offering 192GB of 7.4Tbps of high-bandwidth memory (HBM) per chip, a 6x increase on Google’s sixth-generation Trillium TPU, in addition to 1.2Tbps bidirectional Inter-Chip Interconnect (ICI) bandwidth, a 1.5x improvement over Trillium. Anthropic has already signed on to use up to a million TPUs in a deal worth tens of billions of dollars.
Google's CEO, Sundar Pichai, also addressed concerns about an "AI bubble" during the meeting, reported CNBC.
An employee asked how the company is protecting itself against a potential "burst," to which Pichai said: “I think it’s always difficult during these moments because the risk of underinvesting is pretty high. I actually think for how extraordinary the cloud numbers were, those numbers would have been much better if we had more compute.”
He added that "there will no doubt be ups and downs."