Google need to catch up ! The only thing that is they have their own best chips and Gatekeeper of…
Google need to catch up ! The only thing that is they have their own best chips and Gatekeeper of lot of internal models ! Fast inference and less run cost ! do anyone support the deep mind division of Google ! Is it the internal conflicts or mismanagement or a safe play ?! Distribution channels and etc . They are losing the moat with small startups growing @evanotero @_vamsibatchu_ .
Hope they will come back to the front lines .
Context
Google Cloud's docs describe TPU7x as the first release in the Ironwood family, Google Cloud's seventh-generation TPU, designed for large-scale AI training and inference. Documented per-chip figures are 192 GiB of HBM, about 7,380 GB/s of HBM bandwidth and 4,614 TFLOPs of peak FP8, with 9,216 chips per pod. A Google Cloud blog post of 6 November 2025 describes Ironwood as hardware behind models like Gemini, with 4 chips per host and 64-chip cubes.
That Google has its own chips is grounded by those first-party pages. The figures are vendor specs, not measured performance, and no cost per token or latency claim was found in the Google pages read. Third-party cost comparisons, such as a SemiAnalysis post and relays reporting up to 50% lower inference cost than B200 or B300, were seen at search-result level only, so those figures are unverified. Fast and cheap inference, losing the moat to startups and the two named handles are the author's opinion with no inspected evidence, and current docs cannot show a trend.
Related work
- Inside the Ironwood TPU (Google Cloud blog, 6 Nov 2025) ↗First-party.
- TPU InferenceX (SemiAnalysis, 7 Sep 2026) ↗Snippet only; not read.
Watch next
- TPU7x per-chip-hour pricing, and a full read of the SemiAnalysis post for its workload and latency assumptions.
Sources
- TPU7x (Google Cloud docs)docs.cloud.google.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 14 September 2026 at 03:03 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →