Google is courting the very cloud providers built around Nvidia's chips, offering its in-house Tensor Processing Units as a cheaper and more reliable alternative—while Nvidia responds with financial guarantees designed to lock those customers in.
The Chip War · Google vs Nvidia
Google Raids Nvidia's Cloud Territory With Its Own Chips — and Its Own Playbook
Google is offering its in-house TPUs to the "neocloud" providers built around Nvidia hardware — underwriting demand with billions on its balance sheet. Nvidia is answering with financial guarantees of its own, deepening a circular-financing arms race.
$3.2B
Google backstop for the Fluidstack New York facility — reportedly scalable to $19.4B
~14%
Stake Google takes in Fluidstack as part of the deal
$6.3B
Nvidia capacity guarantee to CoreWeave through 2032, plus a $2B equity stake
Ironwood vs Trillium
Google's 7th-gen TPU: ~4× the per-chip performance
Liquid-cooled Ironwood pods pack 9,216 chips delivering 42.5 ExaFlops — a real technical case for competing on demanding workloads.
The circular-financing loop
Chipmaker
funds the demand
→
Neocloud
builds data centers
→
Silicon deployed
demand secured
Both sides now use their own balance sheets to underwrite the buyers of their chips. Google chases Nvidia's installed base — CoreWeave, Crusoe, Fluidstack — while TPUs at Fluidstack's New York site are slated to serve Anthropic at multi-gigawatt scale from 2027.
Google's momentum
~4× per-chip gains with Ironwood
Anthropic expanding to multi-gigawatt scale
Financing muscle to fund the data centers
The catch
Nvidia's software ecosystem stays dominant
TPUs unproven outside Google's own walls
Adoption limited beyond marquee names
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…