← AI Daily Briefing

AM Intelligence Orders 20,000 Nvidia Rubin GPUs, No Renter Named Yet (October 07, 2026)

October 07, 2026 · 7m 45s · Listen

Somebody just ordered a hall's worth of Nvidia's next-gen chips, and nobody's said who's moving in. So who's stuck with the bill if nobody does? AI Daily Briefing, I'm Bill. I'm Cassidy. Also today: AMD's memory squeeze, California laws that are actually signed, and a quick post-raise buy. First, the order with no tenant. Here's Sean at SOTKN:

AM Intelligence, the AI cloud set up by the founders of Indian power group Greenko, has placed two more binding orders for about 20,000 Nvidia Rubin GPUs. They go into about 70 megawatts of halls in India and Malaysia, with delivery due in the second quarter of 2027. On the neoclouds layer, that is a purchase, not a rental. The company has not named who will rent the new capacity.

Twenty thousand Rubin GPUs, about 70 megawatts across India and Malaysia, delivery Q2 2027. And AM Intelligence bought them outright. Nvidia gets paid as systems ship, so if no renter turns up, those are AMI's empty halls. And split the money carefully. What's confirmed in the release is six billion committed for the first 100 megawatts. The four billion for these 20,000 chips? That's two people talking to Mint. So nobody should be backing out a per-GPU price from that. Including me, and I really want to. What I'll say is there's probably demand out there. It just doesn't have a name yet. But they're planning another 300 megawatts at over twenty billion, so somebody has to sign a lease eventually. It's Anthropic's reported $518 billion in mostly non-cancelable commitments, flipped around. There, the buyer of compute is locked in. Here, it's the builder. Binding orders, tenant to be announced. The money story's showing up at every layer now. Here's Wilson Sonsini:

As the 2026 legislative session came to a close, California passed a wide-ranging assortment of privacy and artificial intelligence (AI)-related bills. In total, Governor Gavin Newsom signed over two dozen privacy and AI bills into law.

Okay, after a morning of GPUs with no named renter, here's something that actually binds people who ship code. California employers can't rely solely on an automated system to discipline or fire someone anymore. So if you sold an HR tool that auto-flags terminations, there's now a human in that loop by law. Signed, not proposed. And look at the spread. Firing decisions, bias in clinical decision support, kids on companion chatbots, emotion recognition at work. Different problems, different rules. Which is exactly the distinction Washington keeps blurring into one word. The chatbot one's the sleeper, though. Independent child safety audits, penalties often assessed per child. Per child. Multiply that by your user base and legal just cancelled their weekend. And the state's registering the AI auditors itself. So who lands on that list, and how fast, decides whether these audits have teeth or turn into a checkbox. That's the next fight, and honestly it's a supply problem too. Diego Almada Lopez, writing in Crypto Briefing:

Mass production of Helios began in July 2026. Initial shipments were scheduled for the end of Q3 2026, with significant volume increases expected in Q4 2026. Volume shipments are expected to serve major customers including OpenAI, Anthropic, and Microsoft. AMD has also found that demand outran its own projections. Interest in Helios has far exceeded the company’s initial forecasts for 2027, prompting AMD to plan substantial production increases over the next three to five years.

Okay, now I can actually picture the box. Seventy-two MI455X per rack, 432 gigs of HBM4 on each, so about thirty-one thousand gigabytes of memory in one rack. After that HPE billion-dollar order, that's the spec I wanted. Still nobody's handing me tokens per second on ROCm. And forget the July production date. Q4's the one to watch, when volume ships to OpenAI, Anthropic and Microsoft. Meanwhile Lisa Su spent early October in Seoul asking Samsung for more HBM4. CEOs don't make that trip when they've got plenty of memory. Yeah. You can respin a chip. You can't respin Samsung's yields. And a fast accelerator without enough memory bandwidth just sits there waiting. Also, 'demand blew past our 2027 forecast' is AMD grading its own homework. Fair, but hold it up against the AM Intelligence order. There, the hardware's coming and the renter has no name. Here, the buyers are named and memory's the scarce part. Same money-and-supply story, from the chip all the way down to the neocloud, and it's bigger than anything anyone launched this week. From Dealroom:

Bellevue-based Temporal has acquired Oso, an AI cybersecurity startup. The deal, announced in October 2026, is an acquisition for an undisclosed sum. Why now? The purchase comes just weeks after Temporal raised a $550 million Series E round. The fresh capital gives the app development company firepower to expand through acquisition.

Last one. Temporal, the Bellevue workflow-engine company, just bought Oso, the authorization and security shop. Price undisclosed. And it's just weeks after they closed a $550 million Series E, so that money didn't sit in the account long. Four for four today, Bill. Even the software story ends up being about where the capital goes. Yeah, except I actually like this one. You run a long agent workflow, and what bites you is who's allowed to do what at each hop. Putting authorization inside the orchestration layer instead of bolting it on later? That's the right place for it. And it pairs with the California package from earlier. Those audit and hiring rules are on the books now, so a platform that logs who authorized each step has something real to show an auditor. That's the generous read. The eyebrow read is half a billion raised, and the first headline move is an acquisition. I'll call it strategy once the integration actually ships. Got feedback, a story idea, or a correction? Email us at aidailybriefing at lantern podcasts dot com. We read every note, and your input helps make the AI Daily Briefing better.

We're watching for AMD's expected Q4 2026 Helios volume increases, with shipments to OpenAI, Anthropic, and Microsoft while HBM4 supply stays tight. We're also watching California's SB 947 limits on automated decision systems in employee discipline and termination, which take effect July 1, 2027.

Links to every story are in the show notes, so dig into whatever caught your ear. That's AI Daily Briefing for today. We'll be back tomorrow. This is a Lantern Podcast.