Another POV we included is @levie
’s point about how AI costs ≠ lower AI bills, which reminded me of what we wrote in last week’s newsletter: that new focus on AI efficiency could just be the latest phase of Jevons Paradox. Excerpt from last week’s Monday even before K3:
ECONOMY
-

Cheaper AI Can Increase Usage Instead of Lowering Total Bills
By
–
-
Model Routers Need to Measure Quality Alongside Cost and Latency
By
–
I ignore most “smart routing” pitches. Picking the cheapest model is trivial, proving it actually met your quality bar is the hard part, and nobody shows their work.
— Nico (@nicos_ai) 20 juillet 2026
Router’s beta is aimed at the hard part: it logs model, usage, latency, and estimated cost, then tests semantic… https://t.co/LJjboATvfHI ignore most “smart routing” pitches. Picking the cheapest model is trivial, proving it actually met your quality bar is the hard part, and nobody shows their work. Router’s beta is aimed at the hard part: it logs model, usage, latency, and estimated cost, then tests semantic
-
Why AI’s Benefits May Be Hard to Capture in GDP
By
–
Most of AI’s benefits won’t show up in GDP, just like most of computers’ didn’t.
-
Training Open Frontier Models Raises Questions About US Value Capture
By
–
But can any US firm capture the value of open frontier LLM investment? We have a decent number of weak open models that trained for millions, but nothing close to the frontier, which would cost much more.
-

AI Compute Growth Depends on Building Data Centers and Power Capacity
By
–
THE AI RACE ISN’T ABOUT CHIPS I thought it was a capital and GPU game. Wrong. The US added 18GW of datacenter capacity in 2025. It needs ~100GW in 2028 alone. The race is about who can physically build.
Fluidstack: $830M at a $7.5B valuation, leading Anthropic’s $50B compute -
AI Supply Chains Are Becoming a Matter of National Prosperity
By
–
Recently in the U.K. I gave a speech on how we are manufacturing non-biological Intelligence. I said this is a geopolitical event: the most important of the 21st century; and that owning the supply chain of Intelligence will be foundational to every nation's future prosperity
-
Kimi and GLM Raise Questions About US LLM Providers’ Profitability
By
–
This won’t age well, at all. Kimi is going to have massive *immediate* impact on the already shaking profitability of American LLM providers. And (along with GLM 5.2] it’s decisive evidence that the DeepSeek moment was not a fluke.
-

AI Roundup Covers ChatGPT Sites and AI Budget Planning
By
–
Top stories in AI today: – Anthropic ends Fable 5’s access questions
– The Rundown Roundtable: Our AI use cases
– Deploy a mini-SaaS in minutes with ChatGPT Sites
– OpenAI CFO’s new math for AI budgets
– 4 new AI tools, community workflows, and more -

Meta and Anthropic Discuss a Reported $10 Billion Compute Deal
By
–
Meta and Anthropic discuss $10 billion AI compute deal https://
cloudcomputing-news.net/news/meta-anth
ropic-ai-compute-deal/?utm_source=dlvr.it&utm_medium=twitter
… #Cloud #Automation #Data #MLOps #Innovation #AgenticAI #ExecutiveLeadership #DataArchitecture -

Apple’s Reported Valuation Overtakes NVIDIA as AI Distribution Gains Attention
By
–

Apple quietly passed Nvidia as the world's most valuable company (~$5T). For 2 years, Nvidia was the AI boom's biggest winner. Now the market is rewarding distribution + ecosystem over raw compute supply. AI is becoming a consumer story.