Welcome!

Welcome to thirty-two eleven (”3211”). This newsletter will entail: 3 Reads, 2 Rabbit Holes, 1 Watch, and 1 Listen — 3/2/1/1. Detailing what I have been up to, on the internet.

introduction:

An interesting read to kick things off regarding electricity demands from AI, written by Scott Galloway:

AI’s Electricity Needs Are Growing Faster Than the Grid Can Handle

Tech companies are racing to build new data centers to handle the surge in AI computing demand. These facilities use massive amounts of electricity — just this year, electricity demand from data centers in the U.S. is estimated to rise 22%.

The problem is that the electrical grid isn’t built for that kind of growth. Power shortages are already delaying data center projects by two to six years. Which raises two questions: Can the U.S. actually produce and deliver enough electricity to sustain this boom — and if it can, who’s paying for it?

Nowhere is that question more extreme than with OpenAI. OpenAI wants to build out a chip network that would consume 250 gigawatts (GW) of energy by 2033. That is equivalent to a fifth of America’s entire electric generation capacity.

• To put that in perspective: If OpenAI had 250 GW of capacity today, it would rank as the seventh-largest country in the world by electricity production capacity — ahead of France, South Korea, Brazil, and Saudi Arabia.

The U.S. added a record 56 GW of new power capacity last year. Even if that pace continued annually, OpenAI’s plan would require the company alone to use more than half of all the new electricity the country adds over the next eight years.

Where would that power come from? Nuclear is the popular talking point, but to generate 250 GW from nuclear alone, you’d need roughly 250 new reactors at an estimated cost of $12.5 trillion. Each one can take 15 years or more to build, so the math and the timeline don’t add up.

Natural gas isn’t a realistic option either, as much as Trump wants it to be. Just 12.5 GW of natural gas capacity is expected to come online this year, one of the biggest additions ever. OpenAI’s target is 20x that.

The obvious answer is solar. Renewable power like solar and onshore wind is the least expensive and quickest power generation source to deploy in the U.S., even without government subsidies. Of the 56 GW of capacity added last year, more than half of it came from solar.

But politics are getting in the way. The One Big Beautiful Bill Act scraps key tax credits and adds new permitting hurdles, including a requirement that new projects be personally approved by Interior Secretary Doug Burgum. These changes could cut U.S. solar capacity projections by roughly 20%.

Energy isn’t the only challenge. Local communities are increasingly pushing back against new data centers. About $64 billion worth of projects have been blocked by residents who don’t want them in their communities. Their criticism is simple: These projects raise costs without giving much back.

In major data center hubs, electricity prices have jumped as much as 267% over the past five years, according to Bloomberg. And despite the huge price tags, they don’t create many jobs.

• Apple’s data center in North Carolina was a $1 billion investment, but resulted in fewer than 100 permanent jobs.

For many communities, the math just doesn’t work: higher electricity bills, environmental impacts, and little economic benefit — all to power the infrastructure behind an AI boom that may be running faster than the grid, the economy, or the public can keep up with.

https://www.profgmarkets.com/subscribe

3 reads:

1 - Comparing LLM models at GenAI image revisions / creations.

2 - Pretty cool to see IC360 played a part in the recent bust of insider trading of sports betting in the NBA.

IC360 is a portfolio company of SeventySix Capital (based in Conshohocken, PA).

3 - For those planning on hosting parties as we approach the holidays…

2 rabbit holes:

1 - My X (twitter) feed has become flooded with Polymarket and Kalshi market betters, some turning $15 into $96,000 betting on the weather in London for a specific day (wild).

2 - Bullish on Wispr AI - the best dictation AI software out there on the market. Currently using the free version, but potentially going to dive into Pro eventually.

1 watch:

1 - Casey Neistat watch shopping.

1 listen:

1 - Lady and Man - Khruangbin.

Cheers,