IREN secured $2.8B in new GPU financing covering 90% of capex, plus $3.6B of Microsoft contract financing at 6%.
Cheap capital pledged against contracted demand is how these compute landlords fund the buildout with minimal dilution. $IREN
SK Hynix now expects memory to stay in shortage until the end of 2030.
Pricing power is migrating down the stack as component supply, not just raw silicon, sets the ceiling on the AI buildout. $MU
Perplexity's Portable Computer runs local models on a single NVIDIA DGX Spark, and pairing two or four scales up to frontier-class open models like Nemotron Ultra.
We've said every agent needs its own computer, and this is that thesis shipping as product. $NVDA
OpenAI's Jalapeño inference ASIC, fabbed on TSMC N3P, beat Nvidia and AMD accelerators on several efficiency metrics in SemiAnalysis testing.
Every custom-silicon program still routes through one foundry, so more inference ASICs means more leading-edge demand for TSMC. $TSM
Amazon's Anthropic stake is now marked at $190B, up from $60B in Q4 2025 and a 214% gain in three quarters.
The strategic value compounds as Anthropic trains and runs inference on AWS silicon, keeping the return inside Amazon's own stack. $AMZN
"Nebius felt like the grown-ups of the neo cloud space."
Shay Boloor on episode 52, on the $NBIS Blackwell auction that cleared 15% above the company's own best price, and why management says all of 2027 could book at those levels.
OpenAI signed a 20-year lease for up to 8 IT-GW at the old Portsmouth diffusion plant site, with Nvidia as exclusive compute host and a direct investor.
Nvidia's reach now extends past selling chips into owning stakes in the power and real estate underneath them. $NVDA
Cerebras missed revenue at $180.1M but grew cloud 281% YoY and carries $25.4B in remaining performance obligations.
The backlog is the story: inference demand is booking faster than the vendor can recognize it, with revenue set to triple in 2027. $CBRS
