Closed · opens Mon 9:30 AM ET
The Marked Chart Wire · Tech and AI
AI and chips
Google's TPUv7 reportedly outperforms Nvidia's GB300 by 1.64 times in AI inference cost efficiency, according to preliminary benchmarks. Operating costs are estimated at $1.21 per chip-hour for Google versus $2.31 for Nvidia.
Block AI
The company revealed a model that processes audio directly without text conversion. It was trained on human dialogue using a single NVIDIA hardware unit.
AI Spending
Andy Jassy notes that running artificial intelligence models costs roughly 10 times more than needed for general software adoption. The retailer plans $220 billion in 2026 spending and has locked in $225 billion in future chip deals.
AI Spending
Nvidia CEO Jensen Huang said capital demand outstrips supply for AI infrastructure, though he expects the shortage to resolve. Goldman Sachs estimates AI spending will reach $7.6 trillion from 2026 to 2031.
Australian Data Centers
Less than a quarter of dozens of computing facilities in Sydney and Melbourne have secured confirmed electricity. NEXTDC shares plunged 41.7 percent after the firm lowered revenue forecasts due to delayed grid upgrades.
Nvidia
Nvidia introduced Minerva, an AI operations platform that monitors hardware and predicts failures in large data centers. IREN is already using the tool, while Tesla, IBM and Bitdeer plan to deploy it.
Quantum Computing
Quantum X Labs, where Gix Internet owns a 13 percent stake, reports lowering logical errors by 12.7 percent on one metric and 12.1 percent on another. IBM tests showed measurement bias falling below 6 percent.
Fast headlines, summarized. Not verified and not advice. Check the full story before you act.