What a time to be live. The main thing happening today is OpenAI publishing over 700 maths papers generated by an unreleased AI model. Here's your Order of Play.
13:00 - Daniel Hulme, Global Chief AI Officer at WPP
13:30 - Roman Hölzl, Co-Founder & CEO at RobCo
14:00 - Jonny Cliffrod, Partner at Entrepreneurs First
14:10 - Vihan Singh, Founder & CEO at Omnibook
14:15 - Deepanshu Rohilla, Co-Founder & CEO at Argon Robotics
14:20 - Tjalling van der Schaar, CEO at Collective Intuition
14:25 - Jaayanth SK, Co-Founder & CEO at Enginuity Labs
14:30 - Joel Edholm, Founder at Goldfish
14:35 - Arthur Garzon, Founder & CEO at Nario Technologies
14:40 - Vayu & Grace, Co-Founders at Lattice Energy
14:45 - Daniel Yamamoto, Co-Founder at Artificial Humans
14:50 - Raghav Kumar, Co-Founder & CEO at Caldera
14:55 - Vaibhav Garg, Co-Founder at Saturn Labs
What’s Happening Today
OpenAI just published 722 maths papers generated by an AI model.
The papers are grouped into 372 families and cover everything from number theory and combinatorics to mathematical physics. They were produced by an unreleased internal model after OpenAI gave it around 4,000 problems. The company says the average result used roughly three hours of ChatGPT Pro style thinking, and has also released 10 summaries of the model's reasoning, including work on the Riemann zeta function.
Many of the papers come with Lean proofs, allowing a computer to check the formal logic, but others remain unformalised and could contain errors. OpenAI decided which results were significant enough to publish, meaning the collection hasn't been independently assessed in full.
The release comes days after an independent advisory group at the Institute for Advanced Study published guidelines for AI labs sharing maths. It asked labs to stop testing hard problems exclusively on proprietary models and to avoid turning results into marketing. OpenAI says it drew on that advice, but hasn't named the model or published the prompts. Mathematicians can read the papers, but they can't reproduce how most of them were made.
We're getting into all of it on today's show at 12 BST.
Daily Op-Ed — SpaceX's $40B bet on Nvidia
The FT reported yesterday that SpaceX is in talks to raise $40B of debt to finance the purchase of more Nvidia chips & hardware. I have had a few people ask me about SpaceX’s insane capex considering they just raised $85B in their IPO and also floated with unusually high cash on the balance sheet.
In short this huge debt deal is the most recent signal of the huge spending on chips and other infrastructure. The high frequency of these capital injections is a statement of priorities: the cash is already being spent, and the company intends to keep spending faster than operations can fund. To demonstrate the insane capex needs, in the second quarter alone, spending hit $18.4B against $7.8B of revenue. At the pace they’re going, the post-IPO cash pile covers only a few quarters of current spending before they need to pump more cash into the machine.
SpaceX’s planned chip financing sits on top of an already aggressive capital cycle and this reported financing ($10 billion in bank loans and $30 billion in investment-grade debt) would be the next tranche. Multi-gigawatt terrestrial compute targets over the next few years are estimated to require on the order of $1.5 trillion in external financing by 2028.
The logic is that compute capacity, once installed and contracted, can convert capital expenditure into high-margin revenue faster than just rockets or satellites. Some estimates already project tens of billions in annual AI-related revenue from facilities like Colossus within a few years. If utilisation holds and power and chips arrive on schedule, the debt can be serviced by the assets it finances. The risk here of course is that the payback depends on continued demand, timely delivery, and power availability, none of which are fully under the company’s control.
Adding $40 billion of debt while free cash flow is still very negative lengthens the period where equity holders are funding a build-out whose returns are still prospective. As for the destination of the money Elon has said SpaceX will build its data centres exclusively on Nvidia hardware.
We’ll see in the next few quarters of capex, contracted capacity, and the gap between operating cash flow and investment. This new $40B debt deal just makes the bet larger.
— Luke
etn. Insight
General Intuition co-founder Pim de Witte joined the boys on the show and gave his thoughts on whether we are actually living in a simulation.
Even if we were, “it still wouldn't answer the fundamental question of why.” Then there’s the compute required. “Think about how much energy somebody would have to be spending to simulate us. It is absurd.” Ultimately, “the complexity of the world we live in is so much bigger than anything any simulation today could do credit for.”
Watch the full clip below.
Continental Breakfast
