FRI 23 OCT 2026 · 2 MIN READ
A thin Friday: Nvidia and Google used GTC Berlin to pitch Rubin, power and fleet tooling.
Google says A5X Rubin instances reach general availability next month. Nvidia says 200GW of U.S. projects wait in interconnection queues, and Firebird holds an export license for 45,000 Blackwell GPUs. Nothing else met the sourcing bar today.
General AI Infrastructure
Google Cloud sets Rubin A5X for next month
TakeawayGoogle says A5X instances on Nvidia Vera Rubin NVL72 reach general availability "next month."
- Virgo Network clusters can scale to 80,000 Rubin GPUs in one site, per Google.
- Multi-site deployments can reach up to 960,000 Rubin GPUs, per Google.
- Google is adding local regions for EMEA data-residency needs.
AngleThese are Google's own scale figures for a network design, not capacity it has installed. Watch the November launch for regions, pricing and how many Rubin GPUs customers can reserve.
Buyer Signals & Deals
Firebird gets license for 45,000 Blackwell GPUs
TakeawayNvidia says Armenia-based Firebird holds a U.S. export license for 45,000 Grace Blackwell GPUs in Kazakhstan.
- First phase targets commercial launch in Q3 2027.
- A planned second phase would bring potential capacity to about 500MW.
- Kazakhtelecom has deployed 400 H200 and 480 B300 GPUs, and plans over 1,000 Vera Rubin GPUs in Ekibastuz.
AngleExport licensing, not demand, is the gating item for capacity outside core markets. No dollar figure or financing was disclosed, and the timeline is the vendor's own.
Neocloud & GPU Cloud Moves
Nvidia launches Minerva fleet tool; IREN is a user
TakeawayNvidia launched Minerva, an AIOps platform that tracks "AI factory yield": GPU hours, goodput, uptime and token output.
- IREN is named as a current user; Lambda, Aolani and Bitdeer AI plan to deploy it.
- Dell, Lenovo and Supermicro will offer it in their AI factory portfolios.
- Nvidia's claims of longer time between interruptions are unverified.
AngleOperators are being handed a vendor-built metric for utilisation. Whether Minerva's goodput definition matches what customers are billed on is the open question.
Power & Datacenter Buildout
Nvidia says power is now the top bottleneck
TakeawayNvidia cites over $300 billion in backlogged power equipment and over 200GW of U.S. interconnection-queue projects.
- Nvidia's DSX Flex syncs AI factory demand with grid conditions and onsite generation.
- Phaidra reports about 10% more compute in the same power envelope.
- GE Vernova, Schneider Electric and Siemens Energy are named partners.
AngleBoth headline figures are Nvidia's own and uncited in the post. The 10% gain is a vendor partner's claim, but flexible-load tooling is where interconnection delays may get negotiated down.
How this is made. Compiled each weekday morning by the Rack Rate desk's research agents from primary sources, with every item linked to the document it came from, and reviewed before it publishes. The Angle on each story is the desk's own read; it does not recommend suppliers, and Rack Rate's own data is labelled as such. Corrections: daily@rackrate.ai.