Platformonomics TGIF #77: January 24, 2025

Welcome to the Khakiocracy

I calmly and analytically examine our new system of government.

Stargate: A New Hope?

Stargate “works not because ridiculousness is concealed, but because ridiculousness on this scale becomes something else.”. Oh wait, that is a review of the movie with same name. But it also nicely captures this week’s White House announcement of The Stargate Project.

The announcement is ridiculous with its $500 billion headline, the implication the participants know how to spend that kind of money on frontier-scale data centers, and the lack of explanation from whence electrons will flow when they flip the switch.

As many have noted, this cast of characters is nowhere close to being able to come up with $100 billion, much less $500 billion (which in PR terms means you might as well announce $500 billion. They might even have had to talk Masa down from a trillion). Elon was quick to troll them for not having the money, while Satya’s response to the cash question was more nuanced but funnier.

Stargate is Sam Altman announcing a fund (because VC gotta VC) and hoping an early first close juices further investor interest (the White House announce venue suggests those pesky SEC fundraising rules are in abeyance). But it really is a private equity fund, as most of the money will be debt. Those PE deals always work well, especially for endeavors with risks that don’t show up in the cash flow model.

Our protagonists lack the skills and experience to build data centers at scale, much less efficiently and effectively deploy the industry’s largest budget. Aspiring cloud provider Oracle literally defines a region as six racks, and we’ve chronicled their ongoing difficulties ramping CAPEX. Oracle finally managed to spend $10 billion over the last four quarters (Microsoft spent $20 billion last quarter, and spent more on CAPEX in just 2023 than Oracle has in its entire nearly 50 year history).

The Abilene site where “Stargate 1” popped into existence this week (this piece explores the various Stargate double counting) is a Crusoe Energy facility (they’re another ex-crypto miner turned AI company). Essentially, Open AI is subleasing the Crusoe facility from Oracle.

Open AI just hired their first data center employee, so they’re early to creating an eleven zeroes CAPEX deployment machine. Softbank did fund WeWork, which certainly did spend some money (the story would be stronger if Softbank hadn’t sold NVIDIA to buy WeWork).

Building infrastructure at scale is complex, multi-variate, with multi-year lead times. It is physically impossible to grow infrastructure spending exponentially. Stargate has an immense learning curve to climb, much less build a system that systematically and efficiently turns cash into capacity.

And the last sign this initiative isn’t serious is the lack of any discussion about power. They need gigawatts of energy and need it really quickly by energy timescales. Power is the biggest constraint to scaling AI.

One bright spot from this announcement might be evidence we live in a multi-crony system of crony capitalism. Not just Elon can make half-baked White House announcements.

Even if Stargate is mostly hyperbole, who am I to complain? We welcome “ridiculousness on this scale” as it pushes the CAPEX game to another order of magnitude. Stargate has already precipitated others to raise their bets.

But the biggest question is where does the name come from? The cheesy movie or the 1970s CIA psychic research project?

Spending Larry’s Money

Larry the Lawnmower is shrewdly reaping the PR benefits of Stargate without contributing much cash. He might be saving his simoleons for Tik Tok and/or Salesforce.

My theory is the Tik Tok deal would mostly be for his son “Bronny”, for whom Larry purchased listless legacy media company Paramount. Oracle could continue to provide infrastructure (remember when Oracle was also in charge of ensuring the Tik Tok algorithm was well-behaved?) and Paramount would be able to earnestly explain how Quibi was just early.

I talked to The Information about why Larry would be better off buying Salesforce than trying to catch the hyperclouds who are $200+ billion of CAPEX ahead. Benioff seems to want out, so this deal wouldn’t secure a succession strategy for Oracle. (But the database vampire doesn’t need a succession strategy, as the decades old joke goes, because Larry is going to live forever). After failing to sell Salesforce to either Amazon or Microsoft, and facing real AI disruption risk, Oracle might be the best landing pad for Amway Salesforce. Oracle’s future is in applications anyway.

Some argue Larry should just spend his money on a tailor.

It’s Not Just Me: Germany Edition

A Different Kind of Tariff

Mark has come a long way, but might still have further to go.

I’ve suggested a “countervailing nuclear umbrella service fee” equivalent to EU fines on American tech companies. But why not make it an escalating multiple for each offense?

Attack of the EuroStack

EU government officials call for more government action to address a problem of too much government action?

Would this be more like Quaero or GAIA-X?