The AI boom is running on borrowed money
The Wall Street Journal reports Broadcom is working to arrange more than $50 billion in financing for the custom AI chip it is building with OpenAI, with Apollo Global Management and Blackstone among the lenders in talks.
· The Aigentic · Morning Brief
The chips keep getting bigger, and so do the loans it takes to buy them. Overnight, Wall Street’s biggest private lenders lined up to bankroll the next wave of AI hardware, a former Treasury secretary asked what happens if one promise in the chain breaks, and Samsung signaled just how much cash the boom is throwing off right now. Here is what moved overnight.
The Wall Street Journal reports Broadcom is working to arrange more than $50 billion in financing for the custom AI chip it is building with OpenAI, with Apollo Global Management and Blackstone among the lenders in talks. Oracle is separately talking with Apollo and Goldman Sachs about money to buy chips, and SpaceX has discussed about $40 billion of its own. The talks are early and the amounts could change. Broadcom already set up a financing platform in June, anchored by Apollo and Blackstone, to help fund more than 20 gigawatts of computing capacity for AI developers including OpenAI and Anthropic through 2028. The new pattern is hard to miss: the AI build-out is increasingly paid for with private credit, not cash on hand.
Bloomberg reports former Treasury Secretary Robert Rubin is worried about “circularity risk,” the web of commitments between AI companies, their suppliers and the lenders who have borrowed against those promises. “What happens if they can’t fulfill those commitments or all those borrowed against them?” he asked at the Greenwich Economic Forum, adding that the risk of cascading failures is not “near zero.” He also warned AI could hit knowledge workers “very hard.” As for whether the spending pays off: “I don’t know, but neither does anybody else.” Bloomberg notes the 10-year Treasury yield just hit its highest level since 2002.
the aigentic
Know what matters in AI.
The stories shaping AI, the tools worth trying, and what they mean for your work.
Morning Brief + Closing Time. Two emails every weekday.
Free. Unsubscribe anytime.
Euronews reports Samsung estimated third-quarter operating profit of 107.4 trillion won, about $80.2 billion, up 782.5% from a year earlier. If confirmed, it would be the largest quarterly operating profit ever reported by a tech company, as AI data centers snap up high-bandwidth memory. TSMC separately booked a record quarter of about $46.8 billion in revenue. Samsung shares still fell 2.42% in Seoul as investors questioned how long the memory boom can last.
Axios reports a bipartisan Senate permitting bill would make data centers pay for more of the grid they use. New sites of at least 20 megawatts would have to cover the added costs they put on the power system, and regulators could charge large data centers even more and use the excess to cut other customers’ bills. Harvard’s Ari Peskoe said, “I don’t think we’ve seen it before.” With the Senate out until after the midterms, both sides have time to push for changes before a possible lame-duck debate.
CNBC reports AI agent startup Manus raised more than $500 million in its first round since Beijing forced Meta to unwind its $2 billion acquisition. Boyu Capital and IDG Capital led, with Tencent, HSG and ZhenFund following on. Manus didn’t disclose a valuation; Bloomberg had reported the round would double it to $4 billion.
Anthropic released Claude Haiku 5.5, its cheapest and fastest small model, priced at $0.10 per million input tokens for prompts up to 100,000 tokens. The company says it runs about 75% cheaper on average than Haiku 4.5, and it is the first Haiku with an adjustable effort setting. Anthropic also halved the price of Sonnet 5.5 cache reads and is adding monthly API credits for Max and Team subscribers.
TechCrunch reports Mecka AI, which pays people to record themselves making coffee or fixing cars while wearing body sensors, raised a $60 million Series B led by Sequoia, with Nvidia and Microsoft’s M12 joining. The pitch is to do for robots what Scale AI and Mercor did for chatbots: supply the human data they learn from.
Deals Desk
New compute and power deals, newest first.
Oct 7 · CoreWeave × AdaniConneX — Sole-tenant data center lease at the Taloja campus in Navi Mumbai, India: 240 MW across three 80 MW buildings, with an option to expand by up to 240 MW more, running Nvidia Vera Rubin; first phase mid-2028. They are CoreWeave’s first data centers in India. Bloomberg says the investment totals multiple billions over the project’s life; no exact figure was disclosed. Source: CoreWeave release
Oct 7 · AirTrunk × SMBC, MUFG Bank, UOB and other lenders — An additional $1 billion, funded by a green loan, for liquid cooling to enable AI workloads at the 300-plus MW TOK1 hyperscale campus in Inzai, Chiba, Japan. AirTrunk is backed by Blackstone; the tenant wasn’t named. Source: Reuters via KELO
Oct 6 · Google × Black Hills Corp. — Power supply agreements for Google’s planned data center in Cheyenne, Wyoming: up to 590 MW of grid-connected service, including 564 MW of new gas generation, plus about 2.1 GW of third-party resources via a private microgrid. Service starts late 2027 and ramps to peak load in 2030, with terms through 2048. Google’s contract value wasn’t disclosed. Source: Black Hills release
Full tracker → theaigentic.com/deals
Yesterday’s movers
Biggest AI and tech winners and losers from Wednesday’s (Oct. 7) U.S. cash session among the names we follow:
Historic market snapshot — October 8, 2026 (day % change as of that edition; not live)
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| MU | Micron | 1,088.00 | +4.06% |
| VST | Vistra | 166.72 | +3.88% |
| SMCI | Super Micro | 44.94 | +3.41% |
| INTU | Intuit | 297.24 | +2.56% |
| SNDK | Sandisk | 1,692.42 | +1.92% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| NBIS | Nebius | 237.15 | -5.09% |
| CRWD | CrowdStrike | 265.44 | -4.81% |
| IONQ | IonQ | 41.34 | -4.50% |
| ON | onsemi | 82.51 | -4.40% |
| TER | Teradyne | 411.78 | -4.31% |
Prices are Wednesday’s U.S. closing prices versus Tuesday’s close. Market data is a snapshot, not live.
Watch
Three picks this morning: the open-model race in six minutes, a lab that wants AI to run real experiments, and a cheap trick for making your own agents smarter about what deserves attention.
Fireship — A $6.3 billion open-weight model just got embarrassed by the French...
Fireship runs through a busy week in open-weight models, with Mistral, Reflection AI and Moonshot all making big moves. It is fast and dense, the way Fireship always is. Watch this if you want the state of the open-model race in about six minutes.
Latent Space — AI Scientists Are Here: Autonomous Labs & Synthesis Superintelligence — Periodic Labs
Latent Space hosts swyx and Brandon sit down with Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk, who argue the next frontier won’t come from more internet data but from letting models run experiments in the physical world. They explain why failed experiments may be some of the most valuable training data, and how autonomous labs could compress decades of materials-science trial and error into months. Watch this if you want a serious look at AI doing science, not just math.
Cole Medin — Jev Can SERIOUSLY Supercharge Your AI Second Brain (Here's How)
Cole Medin shows how he uses Jev, a fast, low-cost decision model, to make thousands of quick calls inside his AI second brain: which emails need a reply, which news is worth reading, which ideas he has already seen, and which inputs look like prompt injection. He walks through the prompts, the confidence scores and the places Jev still falls short. Watch this if you’re building an agent that runs on its own and want it to make cheap, fast judgment calls.