Try Pro free

Latent Space · 4 min · Nov 7, 2026

AI data-center investment could approach $1 trillion a year

Jamin Ball estimates that 100 gigawatts of new capacity at $50 billion per gigawatt implies $5 trillion of spending over five years.

In this episode

Full analysis
  1. 1:08:24Anthropic’s $80.7B loss came from convertible accounting8 min
  2. 17:34Anthropic’s serving efficiency rose sevenfold in a year3 min
  3. 1:22:58GPU lenders underwrite the AI lab paying the bills5 min
  4. 9:00An IPO can improve access to credit for compute buildouts3 min
  5. 23:26Enterprise AI spend is still a small share of company budgets3 min
  6. 26:55Revenue growth and efficiency can support Anthropic's compute commitments6 min
  7. 51:58Frontier labs need a repeatable model factory, not one winning model5 min
  8. 1:28:11AI data-center investment could approach $1 trillion a year4 min
Up next · The Tim Ferriss ShowUse the cost of inaction to make change feel urgent5:13

Takeaways

  1. The estimate assumes 100 gigawatts of capacity added over five years.
  2. At $50 billion per gigawatt, that implies $5 trillion in spending.
  3. Power, land, silicon and supply chains can constrain the buildout.

About this segment

Jamin Ball estimates that 100 gigawatts of new capacity at $50 billion per gigawatt implies $5 trillion of spending over five years. That scale makes power, construction and supply-chain constraints central to forecasts.

Jamin Ball

Jamin Ball is a partner at Altimeter and author of Clouded Judgement.

Invests in enterprise software and AI companies, and writes Clouded Judgement on software metrics.

  • Shawn Wang (swyx), Host · Co-founder, AI Engineer

    Wrote “The Rise of the AI Engineer” and founded Smol AI, maker of AI News.

From “Why Would Anthropic Go Public Now? — Jamin Ball, Altimeter”, Nov 7, 2026. Watch the full episode

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 1:28:16

    that or something more or less than that. Are they about equal? I don't know. that's interesting because I think OpenAI was ahead correct by a lot yeah exactly but I think Anthropics closed the gap a lot and OpenAI at least started like Stargate was a plan the first stage in a plan to get to 30 30 gigawatts yeah back to 2030 yeah and so I was like okay I haven't heard that plan from Anthropics but again let's do some fun math really quick on the fly let's see if I can do it getting tired OpenAI let's say I think when they came out with that it was maybe a year ago and they were what, like one and a half gigawatts, something like that.

  2. 1:28:52

    So whatever, round that down to zero for just the sake of the argument. Let's say open an anthropic, each one go from zero to 30 over five years. So you're gonna add 60 gigawatts of AI capacity, just the two labs. What's a good assumption for their total capacity? I don't know, 40, 50%. Let's say it's 40%. And I don't know, now we're getting too tough in math in my head, but you know, let's say that gets you to, I don't know, 110, 120 gigawatts of total capacity brought online over the next five years.

  3. 1:29:20

    Let's say 100, just so the math is easier. And let's say it costs $50 billion to build out a gigawatt. That's $5 trillion. $5 trillion over the next five years, right? And so call it a trillion a year. You're talking about a non-trivial percent of GDP, you know, real GDP. That's like 5% plus or minus of GDP. Like that is a massive, massive build out, you know, rivals, the real world build out. You know, I think I've seen some charts somewhere that kind of compare those expected numbers.

  4. 1:29:58

    But like, you know, when you're measuring things in like whole number percents of GDP, that's huge. And so like the credit markets, debt markets need to be super, super, super deep and super, super open for that to happen. you need global build out. You know, it's kind of funny, you know, semi-analysis number for next year is like 50. And, you know, I think Brad at the All-In Summit said, oh, maybe it's lower.

  5. 1:30:25

    And I think there was some trolling going on back and forth. And I think the end result of that was like, hey, maybe the 50 gigawatt number, I don't want to speak for anyone, was more of like what's constructed. You know, Brad was more of what's constructed and live generating tokens because there can be a gap in what's constructed and ready to go versus what's actually live and generating tokens.

  6. 1:30:42

    but I think it's like a real debate in the world, you know, and, and you could see a pretty wide range of outcomes. You know, do we really add 50 gigawatts of capacity next year as an industry? Um, or is it 20 or 30, you know, like who knows? I just know everyone you talk to in the supply chain is saying, you know, we, we have, we have orders or indications for orders that are well over a hundred percent of our capacity.

  7. 1:31:06

    And like, I don't have line of sight to that slowing down for the next few years. Yeah. And. Yeah. And the limit, like LAN, shell power, all these, but also silicon, like straight up. Totally. Like compute memory. Totally. Every part of the supply chain. Totally. That's the work. That's why Semi-Alios is so powerful right now. I mean, look at these recent Kimi models. Kimi K3. You know how much memory?

  8. 1:31:26

    You need gigabytes of memory just to serve that. You're not running that on your laptop. You know, you need like a rack to serve one model. Again, those are just the big models, right? Like you're also going to have tons of small models. But I don't know, the big models are getting bigger, which only require more memory. and more compute to run. I used to say they were capping out. I used to say the max size is 1 to 2 and then we would just optimize from there.

  9. 1:31:49

    I've been wrong. What's interesting is it does feel like the open-weight models are capping more around 100 billion active parameters. Yes, that's right. It's mostly an inference speed thing. Totally. But I think it's actually really interestingly tied to the generations of GPUs that they have. Yeah, exactly. Because you wait for the very rubens to roll out and whatever China has. Yeah, exactly. Well, then they'll get it.

  10. 1:32:12

    you know? Yeah, totally. So they're really memory constrained at the infrastructure and that actually shapes the model architecture which is like kind of cool. Totally. So that is Anthropic. Yes. Powerhouse tour. Congrats. I don't know how you felt coming into this but you did great. Oh, this is fun.