The Bleeding Edge Weekly — W42: GPT-6 reaches everyone, and the GPU lenders start to sweat

GPT-6 reached every ChatGPT user this week, and the borrowed money behind AI's compute layer started to look shaky. Demand for AI isn't in doubt. What's in doubt is who carries the debt until the revenue arrives.

The week in models

GPT-6 goes to everyone. OpenAI took GPT-6 from staged access to the entire ChatGPT base and added new "interactive answers," according to the Creators' AI weekly digest and AI Search. Both are newsletters, and our briefing didn't find a primary OpenAI post. Creators' AI puts that base at 1.2 billion users, a number only it reports. If the figure holds, this is the largest model upgrade ever shipped to consumers, and it raises the bar for every other assistant, including the ones inside your company's tools. In the same week, The Neuron reports that eight US states have sued OpenAI over alleged harms to children. We haven't seen the filing, and that claim is unverified.

Anthropic's $16.8B quarter: a big number with a flatter slope. The Neuron reports that Anthropic booked $16.8B in quarterly revenue ahead of a possible IPO. The figure comes from a single source and is unverified. Bloomberg's analyst read is that the results "may aid IPO but not AI trade," and the coverage also flags slowing growth. If the number is right, it's the first frontier-lab quarter large enough to compare with public software giants. The IPO question would then change from whether Anthropic has revenue to how fast that revenue is still growing.

CoreWeave's credit prices in a default, and Huang vouches for it. CoreWeave shares fell 15% in one session, and its CEO blamed a market "acting irrationally". Nvidia's Jensen Huang then said CoreWeave is "improving every day" and needs no backstop. The shares recovered slightly, but credit default swaps still imply a better-than-60% chance of default. These figures come from Bloomberg's reporting, and our briefing hasn't confirmed them independently. Australian neocloud Firmus, whose roadshow used CoreWeave as its closest comparable, pulled its float. A hot US producer-price print also revived Fed rate-hike bets (unverified), which is bad news for anyone financing GPUs with debt.

Claude moves into Google Docs. Claude now works inside Google Docs, Sheets and Slides. Our only source is the Creators' AI weekly digest, so this is unverified. Together with its Microsoft 365 integration earlier this year, that would put Claude inside both major enterprise productivity suites, with Google hosting a direct Gemini competitor in its own flagship product. For buyers, "multi-model" increasingly means whichever model is already in the document.

Devices & robotics

Gemini guards Android Messages, if you turn it on. Google is expanding Gemini-powered scam detection in Android's Messages app, with an enhanced protection mode that users have to switch on themselves. The problem it targets is well suited to a model. A University of Florida study of 175,430 user-reported scam texts found that link-based scams, such as fake toll and postal notices, reuse near-identical scripts (unverified). That kind of repetition is what a model catches well. Because the feature is opt-in, though, it only protects people who go looking for it.

No robotics story in this week's sweep was strong enough to include.

What it means for leaders

For CEOs: If the 1.2 billion figure holds, GPT-6 just changed what your customers expect from any AI assistant, whether or not your product runs on OpenAI. The capital story matters as much as the product one. CoreWeave's credit pricing and Firmus's pulled float show markets deciding who carries AI's debt. The question to be ready for in your next board meeting: if our AI compute provider's financing cracked next quarter, how long could our AI features keep running, and who would pay to move them?

For CIOs/CTOs: If you've committed capacity with CoreWeave or a similar neocloud, review the termination, data-portability and failover clauses now, while credit markets are pricing stress. Claude inside Google Workspace means the choice of model is moving into suite admin settings, so audit which models your users can already call from inside documents, and on what data terms. Our read: keep production workloads on hyperscaler-backed capacity, and treat neocloud capacity as burst rather than baseline until its credit stabilises.

For AI transformation leads: GPT-6 in every employee's personal ChatGPT and Claude in the Docs they already use means unsanctioned adoption probably jumped this week without anyone deciding it should. The eight-state suit, unverified as it is, shows that liability grows with the number of users. The experiment to run this month: pick one department and spend two weeks listing the AI features its staff actually use across Docs, Sheets and ChatGPT. Compare that list with your approved tools, then decide what to sanction, what to block and what to train people on.

The pattern

The two most-quoted moves this week both involved an incumbent giving something up to protect its channel. Huang spent credibility vouching for CoreWeave, a customer whose GPU purchases are part of Nvidia's own demand story. Google gave space in Docs to Claude, a rival to its own Gemini. Neither company was protecting a model. Both were protecting the place where AI gets bought or used. Our read, and it is speculation: as models converge on similar capabilities, lasting leverage goes to whoever owns the channel, whether that's the chip supply, the document or the billion-user app. Those owners will prop up fragile partners and even competitors when the alternative is customers leaving the channel. The limit of that approach also showed this week. Huang's words moved CoreWeave's stock but did nothing to the cost of insuring its debt.

---

This post is also published on our Substack newsletter at edge-ai.forum. Subscribe for the weekly roundup direct to your inbox — fresh AI news, executive context, and devices + robotics every Friday morning.