DCD Magazine #62 - The coming wave
Will compute at sea sink or swim?
Will compute at sea sink or swim?
Details disclosed in a report from research firm SemiAnalysis
As company seeks to increasingly sell TPUs, not just lease them
The hardware, dubbed Frozen v2, is slated for launch in 2028
Initial $35 billion to serve Anthropic, via Fluidstack
We love Nvidia chips, say Nvidia-backed firms
Company sees strong Q1 for FY2026
“Purpose-build architectures” designed to support model training and inference workloads
Follows announcement that Broadcom will develop Google TPUs until 2031
Financial terms of agreement have not been shared
Financial terms of agreements have not been disclosed
Researchers invited to use not-yet-public chip that’s five times more reliable than its predecessor
Google says focusing on second modality with increase its ability to deliver mission sooner
Plus the other clouds set to adopt the latest generation of Nvidia AI infrastructure
And a new unnamed XPU customer is teased
All-hands meeting coinciding with GA of Ironwood release reveals extent of AI demand
The future of processors lies in specialized architecture, Google’s VP and GM of AI and infrastructure, Amin Vahdat, said
Fluidstack said to be among those deploying the chips
Joining former Google TPU lead Richard Ho on rumored Broadcom project
Not only is AI company using competitor's cloud, also using Google chips, sources say
Add new regions, bring availability to company partners, and plan new service
Comes in both 256 and 9,216 chip configurations
Freespace optics company reveals new optical chip
Currently available in preview
Maruthy Vedam goes from TPUs to satellite semiconductors
A4 VMs with B200s also available in preview
Acquired double the Nvidia chips as Meta, according to Omdia
Trillium can now be accessed by Google Cloud customers
Qolab secures seed funding from Development Bank of Japan