The Bleeding Edge Weekly — W44: AI goes to the bond market, and the labs start selling the help

Debt is now paying for part of the AI build-out. Global bond sales hit a record on an "AI borrowing binge," and Microsoft is back in the bond market for the first time since 2017. Meanwhile the labs are moving into services and distribution to make sure all that capacity gets used.

The week in models

The build-out goes to the bond market. Microsoft started a $25B US high-grade bond sale, its first since 2017, and peak orders reportedly reached $115B. That sits inside a bigger number: AI borrowing drove global bond sales to a record. In the same week, Alphabet's revenue and income disappointed and its shares fell, and Nasdaq futures slid as mixed Big Tech results tested the AI trade. Multiple outlets confirm the Alphabet miss and the record issuance. The Microsoft order-book figures come from Bloomberg alone, and we haven't confirmed them independently.

Anthropic goes after the business buyer. The Financial Times reported on October 23 that Anthropic is getting into consulting. We are relying on The AI Opportunity's write-up of that report and haven't verified it ourselves. The product side is better documented. Claude Duet got its first hands-on reviews: Lenny's "How I AI" and ChatPRD walked through building a sales deck and a sales pipeline in it, and AI Search listed it among the week's launches. The examples are sales and go-to-market work, not code. If the consulting report holds, a lab selling implementation work competes directly with the integrators that resell its models.

Nvidia and Hugging Face: a big number, few details. The Neuron reports that questions are being raised about a $12.93B Nvidia deal involving Hugging Face, the largest public hub for open-weight models and datasets. The excerpt gives no deal terms and doesn't say who is asking the questions. The deal is unverified, and the missing details are the point. If the dominant chip supplier gains influence over the main distribution point for open models, the most vendor-neutral part of AI becomes less neutral.

29,000 findings, 516 patches. AI Dev Signals (MarkTechPost) reports that Anthropic's open-source vulnerability scanner has logged 29,000 findings, and 516 of them have been patched. In a separate report, The Neuron counted 62 critical or high-severity Chrome bugs found this week. We haven't verified either figure independently. The ratio is still the story: AI now finds bugs much faster than maintainers can fix them, so the bottleneck has moved to triage and remediation.

Devices & robotics

Applied Intuition, the autonomy company hiding in plain sight. The AI Opportunity profiles Applied Intuition, which builds the simulation and software layer for autonomous vehicles and machines and sells it to the automotive and defence industries. The newsletter calls it a $15B company "nobody knows." Drawing on a 52-minute conversation with CEO Qasar Younis, it argues the company could eventually be worth $100B. That's one newsletter's thesis and the valuation figures are unverified. The business model is still worth a look, because it doesn't depend on which frontier model leads this month.

What it means for leaders

For CEOs: The cost of AI capacity is no longer just a technology curve. It now depends partly on credit markets. Microsoft is borrowing, and one Alphabet miss can move the whole AI trade, so expect your board to ask whether your AI plans assume prices will keep falling. Your model vendor may also start pitching you consulting. The question to be ready for: if our AI supplier also wants to run our transformation, who is checking its advice independently?

For CIOs/CTOs: Two exposure points moved this week. Your supply of open models may soon run through a hub that sits closer to Nvidia. Mirror the weights you depend on and keep the artifacts in your own registry instead of pulling them live from Hugging Face. If you run AI code scanners, expect findings to outrun fixes, as the 29,000-to-516 ratio shows. Our read: keep your current scanners, and spend the next budget on remediation and triage capacity, not more detection.

For AI transformation leads: Duet's first reviews are about sales decks and pipelines, which makes it a good candidate for a contained pilot outside engineering. On governance, the consulting report raises a procurement question worth settling now: should the company that supplies your model also design your rollout? Write the policy before the pitch arrives. The experiment to run this month: give one sales team two weeks with Duet, and score its output against the same deck built the old way, measuring both time and quality.

The pattern

This week's borrowing and this week's business moves look like two halves of one strategy. Borrowing at record scale only makes sense if the capacity it pays for gets used, and the labs are now reaching downstream to guarantee that. Anthropic is moving toward consulting and business-user products like Duet. Nvidia is reportedly moving toward the hub where open models are distributed. Alphabet's miss shows how quickly investors punish doubts about demand. Here is our speculation: the scanner numbers hint at what that downstream business could look like. A tool that produces 29,000 findings and 516 fixes leaves a large remediation backlog, and clearing that backlog takes people, which is what consulting sells. We have no evidence that Anthropic plans to connect the two. Still, when suppliers carry debt and also sell the services that create demand, buyers should ask who benefits from each problem that gets found.

---

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.