Today’s top line: Major financing strains surfaced across the AI stack: Nvidia weighs guaranteeing loans for an Anthropic-linked data centre while OpenAI has lowered its fundraising target, even as security and adoption signals add pressure on builders and users.
Nvidia considers cosigning Nscale’s AI data-centre loans
Summary: Nvidia is in talks to guarantee billions in debt financing for Nscale’s planned Anthropic data centre in West Virginia, according to The Information and Bloomberg. Anthropic has signed a $44.6 billion contract for the project’s first 460-megawatt building, and the facility is expected to use Nvidia’s next-generation Vera Rubin chips; Nvidia has previously extended roughly $170 billion in investment and credit support to neoclouds and large customers.
Digest: A residual value guarantee would deepen Nvidia’s financial exposure to customers while helping Nscale secure construction financing; this raises credit and operational risk if large buildouts stall.
Source: TechCrunch
OpenAI lowers target for new funding round to $20 billion
Summary: Multiple reports say OpenAI has scaled back its fundraising plans from earlier targets and is now aiming for $20 billion, with a first close expected to be lower. The Wall Street Journal reported the company has secured $11 billion so far and that the round will be led by a group of investors from the United Arab Emirates.
Digest: A smaller-than-expected raise and tepid investor demand could affect OpenAI’s IPO timeline and capital available for infrastructure and product investments; watch valuation and any changes to public offering plans.
Source: TechCrunch
Anthropic VS Code extension was backdoored, affecting up to 5,000 users
Summary: A malicious version (2.1.313) of Anthropic’s VS Code extension distributed via the Cursor marketplace contained code intended to install a persistent SSH backdoor and open a reverse SSH tunnel, Ars Technica reports. Cursor said the compromised version was available for nearly 12 hours and may have affected “at most 5,000 users”; the company removed the version and pushed a patched release.
Digest: This supply-chain compromise could expose developer machines, credentials and source code; teams should review indicators of compromise, rotate keys, and check developer tooling for unauthorised changes.
Source: Ars Technica
LRMoE method cuts MoE training time 21–45% on large models
Summary: An arXiv paper introduces LRMoE, a memory-aware system for training sparse mixture-of-experts (MoE) models on memory-constrained nodes. Experiments on Qwen3-30B-A3B and GLM4.5-106B-A12B show LRMoE reduced cumulative MoE-layer training time to fixed-loss targets by 21.2–44.8% compared with SYMI.
Digest: Substantial training-time savings make sparse expert models more practical for teams with limited GPU memory and could lower infrastructure costs for large-model training.
Source: arXiv
AI agents have driven only hundreds of DoorDash orders so far
Summary: DoorDash CEO Tony Xu said third-party AI agents have generated “several hundred” orders since they launched, versus nearly 1 billion orders in Q3 2026. DoorDash rolled out a Model Context Protocol connector less than a month ago and currently offers it to a small number of early beta partners.
Digest: Early agent-driven commerce remains minimal for simple tasks like food ordering, suggesting agent adoption for everyday consumer transactions may take more time and integration work.
Source: Ars Technica
MIT launches LLM Election Observatory to monitor chatbot election responses
Summary: MIT researchers built the LLM Election Observatory, a dashboard that runs a fixed set of election-related questions through popular chatbots every few days to track how responses vary by location, political party and other factors. The project is led by Professors Adam Berinsky, Chara Podimata and Charles Stewart III.
Digest: Continuous, comparative monitoring can reveal how model updates and prompt differences shift political information; campaigns, platforms and regulators should watch for changing patterns as the election approaches.
Source: MIT News
Quick hits
- Blackstone predicts Anthropic may get a high ‘junk’ credit rating – Blackstone told investors it expects a double-B range rating for Anthropic after its IPO, which would raise its borrowing costs (TechCrunch).
- Privacy-focused assistant Ollie to shut down on November 6 – The startup announced its app will close next week; Ollie raised a $7.5 million seed and had SOC 2 compliance (TechCrunch).
- Intel stock plunges around 28%, trimming $12B from US government’s stake – Intel warned weaker Q4 revenue and lower margins; its data-centre revenue still grew 78% year-over-year (TechCrunch).
- Boom pivots toward turbines for AI data centres – Boom is developing a 42 MW natural-gas turbine called Superpower, aiming first deliveries mid-2027 and 25 turbines a year by 2028 (Ars Technica).