In the News: AI Spend Stalls as Usage Climbs, a Claude Fable Fault and New UK Powers Over AI Agents

Ramp data shows AI spend flat as usage rises, Claude Fable 5.1 errors hit the API, the UK moves on AI agents, Apple adds $33B in memory commitments, and a 50x A

By Avik Ghosh, Managing Founder, illuminis

Five stories from the past week that matter to anyone buying, running or regulating AI, with our take on each.

Companies Use More AI but Spend No More on It

On November 6, TechCrunch reported Ramp data showing token spend grew only 1.5% across August and September while token volumes kept rising, compared with 42% month-on-month growth in July. Spend picked up 8.9% from September 23 to October 23. Ramp's lead economist Ara Kharazian pointed to OpenAI and Anthropic price cuts, and to companies getting better at control: defaults that keep expensive models off long-running tasks, budgets for employees, and routers that move work to cheaper models. Source.

Our take: This is what a buyer's market looks like, and the companies that benefit are the ones already sending each step to the right model. Doing more AI work for the same money is not luck. It comes from measuring the cost to finish each task, not the price per token, and acting on it without asking people to change how they work. That is the case for CompletionPrism™.

Claude Fable 5.1 Errors Hit the API, Claude Code and Claude.ai

On November 8, Anthropic reported elevated errors on requests to Claude Fable 5.1 from 08:18 to 08:32 UTC, affecting Claude.ai, the Claude API, Claude Code and Claude Cowork. A fix was deployed and the incident was marked resolved at 08:54 UTC. Source.

Our take: Fourteen minutes is short, but it reached the API, which means every product built on that one model failed with it. Earlier in the week OpenAI's status page also logged errors in paid ChatGPT plans, ChatGPT Work and Codex Web. A software vendor whose features depend on one provider inherits every one of these incidents. Model-level failover should be part of the product, not a promise. See the Embedded License.

UK Ministers Seek Powers to Investigate AI Agent Incidents

The Register reported on November 6 that an amendment to the UK Cyber Security and Resilience Bill, introduced at third reading in the House of Lords, would let ministers regulate how certain organizations deploy agentic AI and report incidents. The explanatory note covers safeguards, notifying affected organizations, preserving evidence and government-led incident reviews. Any rules would need consultation and parliamentary approval. Organizations that could fall in scope include energy suppliers, managed service providers and cloud providers running AI agents for customers. Source.

Our take: Preservation of evidence is the important phrase. When an agent goes beyond its scope, the company that deployed it will have to show what the agent did and why, and an editable log is a weak answer to an inspector. Records of AI decisions should prove themselves. See InferTrust™.

Apple Adds $33 Billion in Purchase Commitments as Memory Prices Climb

On its November 2 earnings call, Apple said it increased manufacturing purchase obligations by $33 billion from Q3 to Q4 to improve its memory supply. CEO John Ternus called memory pricing "unprecedented" and said Apple partly offset the cost with carry-in inventory, savings on other components and higher product prices. CFO Kevan Parekh said gross margin fell from 50.1% to 47.9% and guided 45% to 46% for the current quarter, mainly because of memory. Ternus said the benefit from carry-in inventory will decrease over time. Source.

Our take: When suppliers hold the upper hand, the choice between longer commitments and spot buying becomes a margin decision, and carry-in stock only buys time. Procurement and inventory teams need to see where component prices are heading and how much cover they actually hold, before it shows up in the margin. That is the work of SourceMind™ and StockBalancer™.

Palo Alto Networks Reported in Talks to Buy Clutch at About 50 Times ARR

On November 5, TechCrunch reported, citing a person familiar with the talks, that Palo Alto Networks is in talks to buy Clutch Security, which secures AI agents' access to company systems, for $500 million in cash plus milestone-based payouts. Clutch was last valued at $85 million and is approaching $10 million in annual recurring revenue, putting the price at about 50 times ARR. It would be Palo Alto Networks' seventh acquisition this year, according to PitchBook. Neither company commented. Source.

Our take: At 50 times ARR, almost all of the price rests on what the target will become, not what it earns today. Milestone payouts help, but the buyer still has to know how solid the revenue is, how the product compares with rivals that have already been bought, and what the integration will cost. Deal teams paying multiples like this need that evidence early. See Deal Intelligence.

Get This Every Week

In the News comes out every week. Subscribe from the Insights page to get it by email.