Austin walks through six AI stories: OpenAI's claim that an unreleased model produced 372 new math results, a reported $40 billion debt package for SpaceX to buy Nvidia chips, a new open standard from Meta, Walmart, Stripe and Sierra for AI shopping agents, Mistral's trillion parameter open weight model, Lambda's reported pre-IPO raise, and Google's Nano Banana 2.1 image model. A recurring thread is compute: who is paying for it, who is renting it, and whether the build out is ahead of demand.
OpenAI says an unreleased model produced 372 new math results
Watch from 2:15
- What happened
- OpenAI posted 722 manuscripts and 372 result families to GitHub on Tuesday evening, including a claimed solution to a four dimensional version of a named conjecture. The model was given about 4,000 problems, and the average result used about three hours of ChatGPT Pro level thinking. Many proofs are checked in Lean, a proof checking language, which Scientific American says makes them all but certain to be correct. OpenAI has not released the model, the exact prompts, or the compute per problem. Fields medalist Terence Tao has criticized the pace, and MIT's Andrew Sutherland says claims of single agent one shot solutions should be treated as unverified until humans test them.
- The claim
- OpenAI says each result resolves or makes real progress on an open question in math or theoretical computer science, that nearly every result came from one prompt to one agent, and that the model is coming soon.
- Austin's take
- Austin calls it a phenomenal advance, wonders if the timing helps offset OpenAI's recent security troubles, and points to three things: the volume and difficulty of novel results, far less compute than last month's Navier Stokes effort that took about 10,000 agents, and how fast things moved in roughly a month. He says there is little direct everyday impact yet.
- Why it matters
- If the results hold up, AI is producing research faster than mathematicians can check it, which changes how science gets done and verified.
- What to do
- Treat the headline number as a claim for now. Watch for independent human review and for the model's actual release before drawing big conclusions.
- Not confirmed yet
- The model, prompts, and per problem compute are unreleased, and outside experts say the single agent claims are unverified until humans check them.
SpaceX reportedly lining up $40 billion in debt for Nvidia chips
Watch from 7:42
- What happened
- The Financial Times reported that SpaceX is arranging roughly $40 billion of debt, about $10 billion in bank loans and $30 billion in investment grade bonds, organized by Apollo with PIMCO among firms in talks, to buy Nvidia chips for AI data centers. It may not close until 2027. Reuters reports SpaceX plans to use Nvidia chips exclusively in future data centers; SpaceX shares fell 1% and Nvidia rose half a percent in extended trading. Morgan Stanley estimates AI infrastructure needs about $1.5 trillion of outside financing by 2028.
- The claim
- Elon Musk has said xAI's Colossus data center could double its Nvidia chips by December.
- Austin's take
- Austin sees the backing from Apollo and big financiers as a real signal of demand, but compares it to the Cisco fiber build out, where overbuilding ahead of demand left the stock taking about 20 to 25 years to recover. He thinks demand is there for now.
- Why it matters
- AI is increasingly built on borrowed money, and debt has to be repaid whether or not the AI pays off.
- What to do
- If you follow AI stocks or infrastructure, watch whether this deal actually closes and how much of the build out depends on debt.
- Not confirmed yet
- The deal is reported, not confirmed, and may not close until 2027. Austin's figures on the data center build time were flagged by him as approximate.
An open standard for AI agents that shop on your behalf
Watch from 14:16
- What happened
- Meta, Walmart, Stripe, and Sierra published the Personal Agent Protocol, an open standard built on OAuth, the tech behind sign in with Google. Users grant an agent specific read or write access to an account, and businesses decide what agents can do and can see what they did. Shopify and several others signed on, and a v0.1 spec is due later this month. Google, Amazon, Apple, and OpenAI are not named partners.
- The claim
- Backers say it lets a business verify a bot is acting for a real person. Sierra cofounder Bret Taylor, who also chairs OpenAI, told CNBC it is kind of chaos until such a standard exists.
- Austin's take
- Austin agrees a standard is needed because the web is built to block bots. He notes Amazon has been blocking agents, likely because agent searches undercut its sponsored ad revenue, and doubts everyone will agree on a single protocol. He's excited about the time savings, like having an agent find sofa options.
- Why it matters
- Right now a store can't tell you from a bot acting for you, and whoever writes this standard shapes how AI shopping works.
- What to do
- Before letting an agent shop for you, look at exactly what account access you are granting. Watch whether Amazon, Apple, and Google join.
- Not confirmed yet
- It exists on paper only; the spec is not out yet and major platforms are not on board.
Mistral unveils Large 4, a trillion parameter open weight model
Watch from 20:15
- What happened
- France's Mistral unveiled Large 4, nicknamed Le Chonk, with a little over 1 trillion parameters and 49 billion active per word, using a mixture of experts design. It was trained from scratch on 4,000 Nvidia Grace Blackwell GPUs over two months in Mistral's own European data centers and covers more than 160 languages. It is available through Mistral's API now, with weights due October 27 under Mistral's own license rather than a standard open source one. VentureBeat notes its 62% score on a software engineering benchmark doesn't establish an outright coding lead.
- The claim
- Mistral calls it the most powerful open weight AI system outside China, and its cofounder says it is at the frontier of open weight models. The benchmark scores are Mistral's own.
- Austin's take
- Austin welcomes more competition and more open models from Europe, calls the coding score great for an open model, and ties the huge compute needs back to why companies like SpaceX are borrowing for chips.
- Why it matters
- Two big non Chinese open models in two days means the open weight race is no longer China's alone.
- What to do
- Developers can try it via the API now; check the license terms before planning to self host after October 27.
- Not confirmed yet
- Performance claims rest on Mistral's own benchmarks, and what the custom license allows is still unclear.
Lambda reportedly raising up to $4 billion before its IPO
Watch from 24:31
- What happened
- The Wall Street Journal reported that AI cloud company Lambda is raising up to $4 billion at a $14.5 billion pre-money valuation, led by Blackstone, as its last private round before a planned 2027 IPO. Its order backlog jumped from $15 billion in June to $50 billion in September, anchored by a $35 billion compute deal with Anthropic. It was valued at $5.9 billion last November. Lambda rents Nvidia GPUs to AI labs, and Nvidia is an investor.
- Austin's take
- Austin argues this is unlike the dot com bubble because these companies have real revenue and backlog, though he notes one customer making up most of it is a risk. He believes the binding constraint is money, not compute, and predicts substantial everyday breakthroughs by the end of 2027.
- Why it matters
- Companies renting chips to AI labs are growing as fast as the labs, but a backlog is promised future spending, not cash in hand.
- What to do
- When you see big backlog numbers, check how concentrated they are in one customer.
- Not confirmed yet
- The round is reported, not confirmed, and most of the backlog depends on Anthropic.
Google releases Nano Banana 2.1 image model
Watch from 29:18
- What happened
- Google released Nano Banana 2.1, an upgrade to its AI image generator and editor built on Gemini 3.6 Flash, rolling out in the Gemini app, Google AI Studio, and the Gemini API. Developers pay about 3 cents for a standard 1K image and about 7.6 cents for a 4K image.
- The claim
- Google says it improves on previous versions across the board. Alpha Signal reports it beats the Pro model on editing.
- Austin's take
- Austin jokes about needing more AI slop, but his real excitement is artists using these tools to cut tedious work and become far more prolific. He also thinks purely human, analog art will rise in value.
- Why it matters
- Image editing is one of the AI features regular people actually use, and this lands directly in the Gemini app.
- What to do
- If you use Gemini, try it for everyday image edits, and compare costs yourself if you use the API.
- Not confirmed yet
- Reports disagree on price: The Decoder says about half the cost of Nano Banana 2, others say unchanged. Google's claims are the only benchmarks so far.
