2025-10-08 23:22:51
I recorded a conversation with Matthew Prince, co-founder & CEO of Cloudflare, a company that sits at the heart of the internet. Few operators have Matthew’s vantage point on how the network is evolving, which makes him uniquely placed to answer: if AI agents read, who gets paid?
Jump to highlights:
(00:46) The web’s currency is dying
(16:19) A mathematical model for knowledge – and its implications on the web
(24:35) What a new business model for the web could be (start here if short on time)
(39:11) How might the agentic web affect content?
Where to find Matthew:
X: @eastdakota
LinkedIn: https://www.linkedin.com/in/mprince/
Members can access a briefing pack I used before our conversation — structured notes on Cloudflare’s AI strategy, crawler economics and governance. These notes will help you get more out of the conversation ⬇️
2025-10-06 21:48:43
“Your global perspective helps me understand my SF Bay Area world” – Cassie R., a paying subscriber
Hi all,
Here’s your Monday round-up of data driving conversations this week in less than 250 words.
Let’s go!
Equity-for-silicon ↑ OpenAI will buy hundreds of thousands of AMD chips, equivalent to six gigawatts of compute capacity or roughly the maximum output of six full-size nuclear reactors. OpenAI will receive an option to acquire up to 10% of AMD.
Gigawatt hunger ↑ Citi estimates that AI computing demand will require an additional 55 GW of global power capacity by 2030 (a little lower than the UK’s installed capacity today), leading to $2.8 trillion in incremental global spending.
Battery storage as a power plant ↑ Batteries supplied over 20% of electricity demand for four straight hours after sunset in California last week. (h/t Nicolas Fulghum)
2025-10-05 09:30:31
In today’s Sunday briefing:
The path to AI completing month-long projects without supervision
OpenAI is building an economy inside itself
China hits the robotics flywheel and again, the West isn’t ready
Plus: gravitational waves, AI and biosecurity, Pakistan’s solar boom, loud equity analysts and the architecture of values.
I had a fun conversation with China expert Dan Wang, whose latest book Breakneck is doing the rounds for all the right reasons. Our research notes on this China conversation are available exclusively to members of Exponential View.
This week, we shared our estimates for the 99% step-length – the number of actions an AI system can take at near-perfect reliability before a human must intervene. Today’s frontier systems reliably manage around 100 steps at that threshold. By our estimates, the number could exceed 10,000 by 2029. A couple of years later, they might have between 3x and 10x that range. At that scale, an AI system could operate for weeks, potentially months, without supervision.
Anthropic’s Claude 4.5 Sonnet, the best-in-class coding model, can reportedly work autonomously for 30 hours on a codebase, a significant jump from the seven-hour ceiling just months ago.
2025-10-04 01:03:31
There is one AI metric that we keep a really close eye on:
How many actions can a system take at 99% reliability before a human must intervene?
We call this the 99% step-length: the number of sequential actions an AI can execute with at least 99% reliability without human help.1
Today’s frontier systems reliably manage around 100 steps at that threshold. By our estimates, the number could exceed 10,000 by 2029. A couple of years later, they might have between three and ten times that range. At that scale, an AI system could operate for weeks – potentially months – without supervision.2
Today, we’ll explain our 99% benchmark and show where we believe the step length of AI could go if the trends continue. Many things could derail this trend, but it’s really important to understand what the world could look like if it does continue.
Earlier this year, researchers at released work showing the length of time that AI can work for on software and coding tasks before failing. It is a great benchmark that we use regularly.
METR’s headline is this: every seven months or so, AI systems are able to undertake tasks twice as long as previously. Their methodology is sound, but they benchmark task length at 50% and 80% success rates. I’ve found that execs often question the usefulness of those levels. A process that works half the time isn’t really one they want to trust.
Even at 90%, one failure in ten attempts would need constant human monitoring. Around 99%, you approach the threshold where autonomous operation becomes viable.
2025-10-03 01:10:59
Every so often, a book rearranges how you think. Dan Wang’s Breakneck is one such book. Its core provocation is that China operates as an engineering state, optimised for building, while the United States has evolved into a lawyerly society, optimised for blocking.
Dan and I have been trying to meet in person for months. At various points we tried to line up meetings in San Francisco, Shenzhen and DC but our travel schedules kept crossing. So we settled for the next best thing: I read Breakneck and we recorded this conversation!
For every major EV conversation, we build a research pack to sharpen the discussion. For the first time, we’ve adapted our research notes for members.
Inside you’ll find:
✅ Dan Wang’s essential ideas, quotes, and strategic claims
✅ Key concepts from Breakneck distilled into usable mental models
✅ Context on AI, energy, process knowledge, export controls, demographics & more
Think of it as a reader’s guide to Dan’s thesis – a way to follow the conversation while also drawing on the deeper research that informs it.
2025-09-30 00:10:14
Hi all,
Here’s your Monday round-up of data driving conversations this week – 9 stats, in less than 250 words.
Let’s go!
GW goals ↑ OpenAI targets a 125-fold energy capacity expansion to 250 GW by 2033. Nvidia is investing $100 billion into OpenAI to roll out millions of Rubin GPUs across 10 GW of new data centers.
Cost of intelligence ↓ Frontier AI costs have fallen. Since o3’s debut, they’ve dropped 64-fold with Grok 4 Fast.
Chinese tech ↑ China is leading in 57 out of 64 critical technology categories, measured by its share of the top 10% of high-quality scientific publications.
AI use in tech ↑ 90% of tech workers use AI at work.1
