OpenAI's Landlord Problem

The $115 Billion Burn

The artificial intelligence business continues to be a story of landlords and tenants. The landlords—Microsoft, Amazon, Google—own the vast, expensive data centers. The tenants—AI labs like OpenAI and Anthropic—rent immense amounts of computing power to train and run their models. For a while, this was a perfectly symbiotic relationship. But the tenant’s rent is now spiraling out of control.

Last week, The Information reported that OpenAI has revised its projected cash burn through 2029 to an eye-watering $115 billion, a figure more than triple the original estimate. Here are the details, via Reuters:

The new forecast is $80 billion higher than the company previously expected, the news outlet said, without citing a source for the report...The company's cash burn will more than double to more than $17 billion next year, $10 billion higher than OpenAI's earlier projection, with a burn of $35 billion in 2027 and $45 billion in 2028, The Information said.

These are the numbers of a tenant paying an exorbitant and ever-increasing rent. The partnership with Microsoft, once a multi-billion-dollar coup that gave OpenAI the infrastructure to scale, has also become its single greatest cost center. The AI landlord problem, it turns out, is that your most important partner can also be your most expensive dependency.

So what does a tenant do when the rent gets too high? It tries to buy the building. OpenAI’s solution is to stop being a tenant and become its own landlord. This involves two monumental and capital-intensive initiatives. First, it is partnering with semiconductor giant Broadcom to produce its own custom AI chips, a move to escape its reliance on expensive Nvidia processors. Second, it is building its own data centers. It has diversified its cloud suppliers to include Oracle and Google, but the real play is the "Stargate" initiative with SoftBank to construct its own massive AI factories.

It is a fascinating and terrifyingly expensive gamble. Designing cutting-edge chips and building global data center networks are two of the most complex engineering challenges in the world. But the move is also a quiet declaration of independence, a strategic pivot to reduce its reliance on Microsoft. It pits OpenAI not just against other AI labs, but against the very infrastructure giants it once depended on. The brutal economics of the AI frontier mean you are either a landlord or a tenant. OpenAI has decided the only thing more expensive than building the factory is continuing to rent it.

Tesla's Trillion-Dollar Narrative

One way to think about a CEO pay package is that it is a mechanism for a company’s owners to reward its top employee for doing a good job. Another, more modern way to think about it is that it is a narrative-setting device, a tool for telling a story about what the company is and where it is going.

Anyway, Tesla’s board is asking shareholders to approve a new pay package for Elon Musk that could be worth as much as $1 trillion over the next decade. This number is, it is safe to say, quite large. It is roughly the same as Tesla’s entire current market capitalization. If Musk hits all his goals, he will be paid one whole Tesla. The board says this is necessary to "retain and incentivize" its visionary leader. Here is Tesla Chair Robyn Denholm on the plan:

Simply put, retaining and incentivizing Elon is fundamental to Tesla achieving these goals and becoming the most valuable company in history. [The package is designed] to align extraordinary long-term shareholder value with incentives that will drive peak performance from our visionary leader.

This seems like a reasonable thing for a board chair to say, but it is worth asking: what exactly are they incentivizing him to do? The company’s core business of selling cars is facing a bit of a slump, and its brand has been dinged by Musk’s political adventures. The board, according to reports, has been nervous enough to start looking for a potential successor.

So the new pay package is not really about the car business. It is about a different, much more exciting business. The milestones Musk has to hit are not about, say, improving panel gaps on the Model Y. They are about things like growing the company’s market cap to $8.5 trillion, putting one million robotaxis on the road, and deploying one million humanoid robots.

This is the real point of the package. It is a tool for fundamentally reframing the company. The message to Wall Street is: please stop worrying about our boring, slightly struggling car company, and start thinking about our amazing, futuristic, world-conquering robotics and AI company. The trillion-dollar number is a useful distraction from a series of disappointing quarterly delivery numbers.

Of course, a Delaware judge already threw out Musk’s last record-breaking pay package, calling the process "deeply flawed". So the board is trying again, this time with a number that is almost 20 times bigger. This is a bold legal strategy. But it is also a bold business strategy. If you can’t fix your current narrative, you might as well try to sell everyone on a much more exciting, and much more expensive, new one.

The Scoreboard

• AI: ASML becomes Mistral AI’s top shareholder after leading latest funding round, sources say (Reuters)
• Self-driving: Uber, Momenta to begin self-driving testing in Munich next year (Reuters)
• Semiconductor: Broadcom stock jumps 9% on new $10 billion customer that analysts say is OpenAI (CNBC)
• Semiconductor: India's manufacturing states fight to attract chip investments (Nikkei Asia)