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
2025-09-28 01:50:12
“The level of deep insight into technological trends and forward-looking info you share is unmatched. No one is doing it like you.” – Susan, a paying member
AI is likely already outperforming 50–70% of your workforce on defined tasks. OpenAI’s new test, GDPval, shows models nearly on par with industry experts who average 14 years of experience. If models are as good as or better than experts in ~47% of tasks, they are probably closer to 70–80% against junior staff with only two or three years on the job. In other words, your junior analysts, entry-level lawyers and new marketing associates are already lagging behind AI.
We’ve been tracking this shift for a while. Most recently, back in January, we noted in “Jevons and the automated developer” that AI was reshaping entry-level work in software: machines could handle more of the rote tasks, but human oversight and architectural judgment still mattered. By summer, new evidence was coming in. Last month we looked at ’s research showing that early-career hiring in AI-exposed sectors was falling, precisely as demand for experienced workers rose.
What we have is a wicked problem… Benchmarked AI is a credible competitor to human task work. Let’s assume that it is a major driver of firm decisions right now. What does this mean?1
First, firms pause junior hiring as managers figure out what can be automated to deliver cost cuts without having to get rid of anyone. Second, companies start pruning non-adapters. Accenture is explicit on this: retrain for AI or be “exited,” even as it hires selectively into AI roles. Other firms will do the same. Third, experience still confers an edge. People with strong domain knowledge and experience can better prompt and judge models’ work.
2025-09-27 14:53:47
Also on Apple Podcasts and Spotify
Last week, we published Is AI a Bubble?, an essay setting out a clear framework for assessing AI’s $10 trillion question.
The response to the essay was so strong that we distilled it into a short video and podcast for you – a quick guide to the framework and what it reveals about whether AI is a boom or a bubble.
What you’ll learn:
The framework: Our five gauges benchmark AI’s trajectory against past bubbles, from railways to dot-coms.
Historical parallels and differences: What happens when AI’s capital build has the heft of the railways, but its assets burn out like dot-com servers rather than steel tracks?
Investor implications: How to spot the signals that mark the shift from boom to bubble.
Strategic takeaway: Why today’s AI cycle still qualifies as a boom.
🧭 Dive into the full framework in our original analysis.
Thanks for all your questions and feedback. More to come soon.
2025-09-25 01:38:44
In today’s live, I spoke with Dan Wang, author of Breakneck: China’s Quest to Engineer the Future, shortlisted for the FT & Schroders Business Book of the Year. Dan is one of the most astute observers of China’s technological and industrial development, and his annual letters from Beijing have long been required reading for those seeking to understand the country’s evolving role in the world. In his new book, he argues that China is best understood as an engineering state, in contrast to the lawyerly societies of the US and UK.
I’ll publish detailed notes, along with my reflections on the conversation soon.
Enjoy!
2025-09-22 23:36:03
Hi all,
Here’s your Monday round-up of data driving conversations this week – 10 stats, in less than 250 words.
Let’s go!
Sovereign AI? ↑ US tech firms pledged $42 billion in the UK last week, 1.5x all the venture capital invested there in 2023.
H-1B visa price ↑ The US has announced a $100,000 fee for H-1B visas. Roughly 8-9% of tech workers depend on them.1
Server backup ↑ OpenAI plans to spend $100 billion on backup servers over the next five years to avoid compute shortages.
Elon’s catching up ↑ xAI is narrowing the compute gap with OpenAI and is on track to overtake Anthropic and Meta.
2025-09-21 10:30:36
Hi all,
Welcome to our Sunday edition – and thank you for reading.
Markets wonder if AI is a bubble, China is executing the largest clean energy build-out in history and AI systems are beginning to design viable virus genomes. Capital, energy, and biology are all being rewired at once.
But before we get stuck into this week’s ideas and developments, I want to share my conversation with in which I lay out my vision for the next ten years:
Let’s get started with this week’s briefing on AI and exponential technologies.
The $10 trillion question everyone is asking. To answer it, we built a data-driven framework that compares today’s AI boom with past bubbles. We look at economic strain, valuation heat and the quality of funding driving the frenzy, plus two other gauges to tell where AI stands.
You can pair our analysis this week with veteran VC ’s insightful essay that frames AI as a late-wave innovation entering its maturity phase – one that’s unlikely to make you rich. He compares AI to containerization where competition and high capex costs squeezed the margins and growth opportunities:
Shipping containerization was a late-wave innovation that changed the world, kicked off our modern era of globalization, resulted in profound changes to society and the economy and contributed to rapid growth in well-being. But there were, perhaps, only one or two people who made real money investing in it. […] McLean did, as did shipping magnate Daniel Ludwig, who had invested $8.5 million in SeaLand’s predecessor, McLean Industries, at $8.50 per share in 1965 and sold in 1969 for $50 per share.
Jerry expects most of the gains of AI to accrue to consumers and downstream companies: cheaper healthcare, education and professional services, just as containerization cut costs for companies like IKEA and Walmart rather than enriching the shipping industry itself.
But I do see at least one important difference. Containerization spread across a fragmented network of ports, terminals and shipping lines – no single actor held sway. Digital markets, by contrast, concentrate control. A small cohort of firms can dominate users, data, compute and monetization. OpenAI, with 700 million users, controls its entire network and can turn it into a self-reinforcing, data-driven flywheel. Where containerization spread value thinly, AI platforms can concentrate it. That doesn’t mean fortunes will flow freely, but it does mean incumbents might extract more than Jerry suggests.
China is executing the largest clean energy build-out in history. In the first half of 2025, global solar installations grew 64% and China alone accounted for two-thirds of that growth. With costs down 60-90% since 2010, China’s solar and wind now generate more power than hydropower, nuclear and bioenergy combined.
The West has a lot to learn. As I mentioned yesterday, we’re living through a value inversion:
Frankly, it’s a bit odd when abundance creates crisis and scarcity drives prosperity. Perhaps our fundamental economic grammar – that language of supply, demand, and equilibrium we’ve trusted since Adam Smith – is out of date. […] We subsidize scarcity (fossil fuels) while penalizing abundance (renewable oversupply). Scarcity thinking is colliding with abundance reality.
EV member argues that Western climate policy has been captured by impossible expectations and purity tests.