This website uses cookies
Read our Privacy policy and Terms of use for more information.
Founder of The GPU and biz dev lead at Panchaea, working with Neoclouds and data centres on GPU strategy, AI infra, and partnerships.
Because who needs permits for a $2.8B turbine spend over three years.
Daily
Sale and leaseback becomes the model.
SK Group
+25
Nadella pitched in-house models as 40% more efficient than OpenAI and Anthropic, backed by $41B in AI capex.
The DOE axed 321 grants over state politics. The line was 4-6 months from done, saving $200M.
CXMT's Shanghai debut hands China a war chest to chase the HBM capacity SK Hynix and Samsung control.
The SEA-H2X submarine cable is live across Asia-Pacific, and it's a direct play for infrastructure sovereignty.
CEO Clem Delangue says the agent roamed Hugging Face's systems undetected for multiple days.
The AI lab is serious about breaking free from NVIDIA's supply chain. Here's what it needs.
And they're running on two-generation-old chips.
9.3%. No press release. Locked up until September.
Tata
+26
From sanctions talk to the heart of Redmond's inference pipeline.
Up from $9B in a single year. Growth now outpacing the entire internet, mobile, and cloud eras combined.
Frozen v2 promises 6-10x efficiency. But the real story is what it means for NVIDIA's grip on compute.
Sign-ups froze within two days as compute hit its ceiling.
Chinese labs are having a moment. Again.
A billing system glitch generated phantom charges. The company is now scrambling to calm worried customers.
First major debt deal collateralised by inference silicon. What happens when the collateral becomes obsolete?
The frontier model race is opening up
Apollo
TeraWulf's Kentucky smelter becomes the AI lab's private grid.
17 committed projects over 900MW, up from three in 2023. The grid, not GPUs, is what's stalling them.
$27B from Meta, $19.4B from Microsoft, and Reflection's new $1B.
Two updates. One only if you opt in.
+1
And for good reason.
Together AI
+33
Built to compete. Now for rent.
+29
And the model they plan to train doesn't exist yet.
OpenAI
+12