Thursday links: bubbles, datacenters, and pickleball in prison

Byrne Hobart offers a timely reminder: investors are good at recognizing breakout technologies and terrible at investing in them. 

He cites the classic example of RCA. In 1929, a holder of its shares would have been thrilled to learn how ubiquitous radio would soon become — and befuddled to hear they were about to lose 98% on their investment in radio.

The pattern repeats, Hobart says, because it’s so difficult to foresee who will ultimately capture the value created by a world-changing technology. 

Hobart argues that overinvestment often ends up subsidizing whoever controls the real bottleneck: “If there's any given piece of infrastructure that's a useful complement to some other layer that has more pricing power, then overinvestment in that infrastructure acts as a subsidy for whoever controls the real value lever.”

He cites the examples of railroads, Gilded Age trusts, and fibre-optic cable: in each case, investors financed the infrastructure and most of the value accrued somewhere else. 

I think we can add crypto to that list.

Crypto investors financed the infrastructure: layer-1 blockchains, DeFi protocols, and zero-knowledge proofs. But the value is accruing elsewhere: to Robinhood, Stripe, and Kalshi, for example.

Worse still, the biggest beneficiaries might turn out to be banks — the very villains crypto set out to disintermediate.

Which raises an obvious question: is AI next?

It seems to fit the pattern: the technology is obvious but the business models are not. Hobart cautions it requires “a pretty hefty upfront due diligence commitment” to have any hope of predicting where value will accrue in the AI ecosystem.

I suspect he’s understating it. 

Jeremy Grantham on SpaceX and bubbles 

Hedge fund manager, market historian, and caller-of-bubbles Jeremy Grantham says we are now in rarified air. “In 50 years,” he told Odd Lots, “market historians will look back and talk in awe of SpaceX, read its prospectus as a kind of novelty/joke, and compare it to the South Sea bubble.”

He likens the AI capex race among hyperscalers to a cage match worthy of the White House lawn. With seven giant players “all deciding at the same time to fight it out in one market,” Grantham warns, “It could be a very messy, blood-curdling game.”

He couldn’t be more delighted.

“These are extraordinary times,” the 87-year-old says. “It’s seldom been more interesting. I’m thrilled to be alive when all the major issues I’ve been studying have come to a head at the same time.”

He concludes with a concise bit of investing advice: “Try to avoid the hype and check the numbers.”

Datacenters in space

Physicist Samir Varma has some advice for anyone trying to check the numbers on SpaceX.

Contrary to popular belief, he says, datacenters in space “is not a cooling bet, and barely a launch bet.” Instead, “it's a bet on how slowly Moore's Law moves.”  

Critics of SpaceX are right that it’s counterintuitively easier to cool datacenters on Earth: “The freezing cold vacuum of space is actually a terrible place to cool things down.” But they’re wrong to conclude this makes the project uneconomic. The solution is simply a very large radiating surface — which is feasible because radiators are light and therefore cheap to get into orbit.

On launch costs, Varma tests the math on SpaceX’s most optimistic projections and concludes it’s at least plausible for building in space to become cheaper than building on Earth.

Assuming that hurdle is cleared, the investment case for SpaceX comes down to a single question: how quickly will the semiconductors they send into orbit become obsolete?

If the performance per watt of GPUs and other AI chips returns to its historical pace of doubling every two years, Earth-bound datacenters will have an insurmountable advantage over space-based ones. The chips in space would become obsolete before they could earn back the cost of their voyage.

If, however, performance per watt doubles at its current pace — about every four or five years — it can make sense. Especially if construction of Earth-bound datacenters gets snarled in shortages of turbines and other scarce parts (or by politics, I’d add).

“That’s the number to watch,” he concludes. “Not the temperature of space, and not even the price of a rocket. The rate at which silicon improves.”

I’m sure Grantham and Hobart would both find that fascinating. And impossible to predict.

SBF incarcerated

Journalist Simon van Zuylen-Wood has an update on Sam Bankman-Fried, who he speaks to twice a week. Mostly it’s interesting for the details of low-security prison life.

(He narrowly avoided high-security prison due to a technicality of time-served, Zuylen-Wood says).

SBF played Shattered Pixel Dungeon roughly 6,000 times for the year or so he had access to a prison-gated tablet, we learn. He reads a lot of escapist fiction (contrary to his previous stance against books). He writes a serialized, fictionalized prison memoir (“allegorical and surprisingly humorous,” Zuylen-Wood says). He’s written a Vegan Prison Cookbook, although there’s no cooking involved. To stay vegan, he avoids the meal plan, assembling what he can gather from the commissary (beans and rice, mostly).

He plays pickleball, teaches a chess class, plays a trading-card game (an hour or two a day), advises inmates on legal strategy, and has a job at the rec department (which pays $5.25 a month).

He still does some trading: in one case, two highly regarded muffins for a pillow rigged from ripped clothes and mattress innards (while in a facility that didn’t otherwise allow pillows).

He also spends a lot of time trying not to change.

“Not only does he not wish to exit prison as a changed person, it is more or less his greatest fear.”

Contrary to the trope that everyone in prison says they didn’t do it, a former inmate tells Zuylen-Wood that 99% of the incarcerated admit they did. 

SBF remains stubbornly in the 1%. 

He continues to argue that the FTX collapse "was mostly the result of an innocent accounting glitch," as Zuylen-Wood paraphrases his position, and that Alameda was “legally allowed to borrow as much as it wanted from FTX.”

He has not had much luck convincing people of that. His petitions for a new trial have been rejected and his rebranding as a Trump supporter has failed to win him a presidential pardon.

Things could be worse, though.

“Bankman-Fried looked well,” Zuylen-Wood reports after a visit, “about 30 pounds lighter than he was on the outside and tanned from months of pickleball in the yard.”

— Byron Gilliam