GPT-5.6 found optimizations that "reduced end-to-end serving costs by 20%" for OpenAI to serve that model Presumably that's billions of dollars a month in savings at this point?
ECONOMY
-
Post Questions Anthropic Revenue Projections Based on Heavy Token Spending
By
–
This post on Anthropic by @dkthomp sounds compelling but I don’t think it actually is. 1. The projections of Anthropic’s revenue are likely overoptimistic. Their big Q2 was projected to be $10.9B (after $4.8B in Q1) but importantly that was at the height of tokenmaxxing, and
-
GPT-5.6 Sol Cuts Model-Serving Costs by Optimizing GPU Kernels
By
–
Interesting: OpenAI says GPT‑5.6 Sol helped cut its end-to-end model-serving costs by 20%, by autonomously rewriting and optimizing production GPU kernels. Sol also improved its own speculative decoding model:
– Designed and ran hundreds of architecture experiments
– Launched -
AI Competition Increasingly Depends on Electricity and Physical Infrastructure
By
–
AI is becoming a physical infrastructure industry. Every model and agent depends on data centres, cooling, substations, grids and electricity. The next AI race will not be won by software alone. It will be won by those who can build the infrastructure beneath it.
-
Longer AI Sessions Increase the Cost of Switching Models
By
–
It gets worse because the tax grows with how much context you're carrying. Double the session, and you double the cost of every switch inside it. Cost per 1,000 prefix tokens barely moved across a six-fold change in session size: https://
alphasignal.ai/news/model-rou
ting-layers-and-the-prompt-cache-tax
… -
Changing Models Mid-Session Can Invalidate the Prompt Cache
By
–
Your model router charges you for context that you've already paid for, eleven times over. Prompt caches belong to one model. Change the model during a session, and the new model will read everything all over again, at the full price. We priced it with three different session
-
Freehand Uses AI to Automate Back-Office Invoice Processing
By
–
the entire BPO industry must be losing
— Nico (@nicos_ai) 29 juillet 2026
for 30 years
India and the Philippines owned the world's back office because checking invoices was cheaper done by people offshore
that's over.
Freehand came out of stealth doing all of it with AI:
• reads every invoice a company gets
•… https://t.co/K9m9x1RUSO pic.twitter.com/0kNfCJEKWrthe entire BPO industry must be losing
for 30 years India and the Philippines owned the world's back office because checking invoices was cheaper done by people offshore
that's over. Freehand came out of stealth doing all of it with AI: • reads every invoice a company gets
• -
Cerebras Inference Could Change GPT-5.6 Speed and Cost Trade-Offs
By
–
GPT-5.6 on Cerebras will usher in a new era. Many are still unaware of how significant this change will be. I am well aware that the costs for inference on these inference chips are very high, but these too will decrease in the future, just as all costs are falling. Imagine,
-
AI-Linked Stocks Expose Differences Between Focused and Broad Product Portfolios
By
–
A little bit of what we are seeing among AI-linked names is the difference of a very, very focused product set versus a very broad portfolio. The broad portfolios of Corning and Vertiv have not insulated them. The very, very focused offerings of Bloom and Seagate are really
-
Electric Power Could Become the Next Bottleneck for AI Infrastructure
By
–
AI is not only a software revolution. Every model, agent and inference request depends on data centres, cooling, substations, grids and electricity. The next AI bottleneck may not be chips. It may be power. My new article: AI’s Next Bottleneck Is Power The AI race will be