Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking about AI servers easing off during peak grid demand, and more!
AI in Action: AI is calculating how long to hold its breath
At Nscale’s AI factory in Iceland, a set of AI server racks received the kind of signal a utility sends during peak demand and responded by backing off. In a pilot unveiled at Nvidia’s GTC Berlin, Emerald AI’s Conductor software, working with Nvidia’s DSX power-management technology, cut the racks’ power use by about 25 percent, held that cut for 30 minutes and then returned to normal. It’s Emerald’s seventh pilot, following tests in Santa Clara, Phoenix and Chicago and a UK trial with National Grid, and the first with Conductor fully integrated into Nvidia’s DSX stack.
Grids are sized for a small number of peak hours each year, which leaves spare capacity sitting idle most of the time, and Emerald CEO Varun Sivaram calls power access the main constraint on Europe’s AI build-out. A data center that can ease off during peak hours can connect using capacity that already exists instead of waiting years for new plants and power lines. Conductor sits between the grid and the data center and reshuffles workloads in real time; with Nvidia’s DSX Flex, operators can defer flexible jobs while prioritizing critical inference. An earlier trial with EPRI, Nebius and National Grid pushed further, cutting a Blackwell Ultra cluster’s draw by up to 40 percent during simulated grid stress while critical work kept running.
The same Iceland site has been testing the other half of the pitch. Using Nvidia’s DSX MaxLPS across four GB300 NVL72 racks, Nscale ran 49 percent more inference throughput and managed 37 percent more GPU capacity inside an unchanged 264-kilowatt power budget. More work per watt in normal hours, and less draw on request when the grid is strained: that combination is what Emerald and Nvidia are selling to utilities. A temporary dip in a cluster’s demand is a different measurement from reducing the energy a workload needs to finish, Converge Digest notes, since deferred jobs still have to run eventually. Emerald is already moving past pilots, with its first DSX Flex deployment alongside Silicon Valley Power.
🔥 Rapid Inferno 🔥
Commentary: Anthropic’s S-1
Analysis: The Hater’s Guide to Blackstone
OpenAI trims valuation target again to $1.25 trillion
“Multiple people with knowledge of the Morgan Stanley-led process said demand from clients was weaker than in the past. ‘There has been zero clamouring,’ said one of the people. ‘All the prior times [OpenAI] have been on there’s been a frenzy. This time is a bit of a thud.’ Some investors were also suffering from fatigue, or ‘indigestion’, having invested heavily in prior rounds, said some of the people.”
Google’s own AI is making the company less valuable
“Alphabet shares tumbled 13 per cent on Thursday, wiping a record $500bn from its market value after the owner of Google burnt through cash last quarter and revealed ‘AI mode’ searches were threatening profits at its core advertising business.”
“The Silicon Valley group’s cash flow fell deeper into negative territory in the three months to the end of September, reaching minus $9.1bn as it lifted its AI spending to more than $50bn in the quarter.”
The debt faucet is being turned off for neoclouds per Elon Musk
“Elon Musk said SpaceX has closed a $40 billion debt deal to pay for the chips needed to fuel its artificial intelligence ambitions. People familiar with the discussions, though, told Bloomberg News that talks over pricing were still ongoing.”
“He went on to suggest that debt markets had all but shut for neoclouds depending solely on revenue streams from AI data centers. ‘That means the 200+ neoclouds are going to see guppy-level death and consolidation,’ Musk wrote.”
Apollo Global Management: No Signs of AI Payoff Outside the Tech Sector
“Despite massive investments and widespread hype around artificial intelligence, the technology has delivered zero financial payoff to corporate America outside of tech. […] And there are still no signs of AI boosting productivity growth, see the third chart below. With no impact on the bottom line, mainstream corporate America is pulling back and starting to view token spending not as a source of growth but as a cost center that weighs on margins.”
Close associate of Jensen Huang found to be Chinese intelligence asset
“A top executive at a Singapore-based data center company was under scrutiny for possibly steering advanced Nvidia A.I. chips to China. And now American officials had concluded that she was a Chinese intelligence asset. […] Surveillance of the executive, Alice Huang or Huang Le, a Chinese citizen, revealed that she had developed a close relationship with Jensen Huang, Nvidia’s chief executive.”
Per Bloomberg in 2025: “Singapore-based Megaspeed has imported tens of thousands more chips than Nvidia saw when checking its data centers.”
Jensen Huang denied the possibility of AI chip diversion to China
IMF warns that AI bubble threatens global financial markets
SK Hynix pulls Solidigm’s IPO amid AI bubble concerns
Banks shelve bond sale for Anthropic-tied Texas data center
China rushes to shore up markets as AI trade sours
Don’t be like Andy Yen. Proton CEO succumbs to AI psychosis.
📖 What We’re Reading
Data quality is the gold standard and a key requirement for building superior AI models. Engineers are well aware that AI models rarely fail because of the wrong algorithm, but rather because of wrong timestamps.
