AI Skepticism & Oracle's Big Risk
My jaw is still on the floor from Oracle’s earnings release. In case you missed it, here’s the Wall Street Journal’s recap:
OpenAI signed a contract with Oracle to purchase $300 billion in computing power over roughly five years…
The deal is one of the largest cloud contracts ever signed…
The Oracle contract will require 4.5 gigawatts of power capacity, roughly comparable to the electricity produced by more than two Hoover Dams or the amount consumed by about four million homes.
Two questions immediately come to mind:
How will Open AI pay for it?
If they do find the money, what will the returns be?
How will Open AI pay for it?
The Oracle contract begins in 2027. Payments will begin at $30 billion per year and ramp to $90 billion per year in 2031 as Oracle brings capacity online. Payments will average $60 billion per year over the five year contract.
$60 billion is enough to buy CSX, General Motors, Cencora, Truist Financial, or United Rentals. For $300 billion could buy all five.
These are big numbers in any sense, but especially relative to Open AI’s current business. OpenAI lost $5 billion in 2024 on $4 billion of revenue. They’re projected to lose upwards of $8 billion on $12 billion of revenue in 2025. They don’t expect to be profitable until 2029. $30 billion in 2027 almost triple Open AI’s current revenue. And this is on top of commitments to buy $10 billion of custom AI chips from Broadcom and tens of billions for cloud computing from Microsoft. They’re betting the farm on growth.
Open AI will have to rely heavily on capital markets to finance its shopping spree. That’s always a risky proposition, as funding always seems to dry up at the worst possible time.
Open AI will have to raise so much money, it makes me question whether it is even possible. The global venture capital industry raised about $105 billion last year. Open AI will likely need to raise well over half of that to meet its contractual commitments. And repeat that 5 years in a row.
Are the venture markets deep enough and liquid enough? Open AI is technically a non-profit and is having trouble transitioning to a for-profit enterprise. If it can’t convert, it is going to be even harder to raise the money.
The $500 billion project Stargate unveiled in January 2025 and lead by Open AI, Softbank, and Oracle, “has struggled to get off the ground and has sharply scaled back its near-term plans,” according to the WSJ. They continue, “While the companies pledged at the January announcement to invest $100 billion “immediately,” the project is now setting the more modest goal of building a small data center by the end of this year.”
I expect more of the same with Open AI and Oracle’s latest contract. That probably wont be good for Oracle’s stock price.
It reminds of the Seinfeld scene where a car rental company tells Jerry they don’t have a car for him even though he reserved one ahead of time.
Jerry: You know how to take the reservation, you just don't know how to hold the reservation, and that's really the most important part of the reservation. The holding. Anybody can just take them!
“Anyone” can just say they’re going to spend $300 billion on cloud computing. But the important part is in the doing. While I wouldn’t call Open AI and Oracle “anyone” I would say that it is much easier to promise hundreds of billions of dollars than to raise it and spend it wisely.
What will the returns be?
Its really hard to earn a good return on a huge investment. $300 billion qualifies as “huge” in my book. Berkshire Hathaway is sitting on $314 billion of treasury bills because Warren Buffett can’t find a better return for that amount of cash.
Generative AI undoubtedly has lots of use cases, but it is unclear how it will be monetized. There are several models out there, but none are covering their costs yet. Only 3% of users pay for AI right now, so there’s a long runway to increase penetration.
One risk to Open AI is that a company like Google gives away or subsidizes Gemini as a loss leader to get users into its ecosystem. A price war would be a disaster for the AI companies given the capital they’re investing.
AI is a high fixed cost, low variable cost business. Businesses like these, such as airlines, routinely succumb to the temptation to sell the marginal seat or AI query cheap because it costs them little. The plane is flying anyway, the datacenter is already built.
There’s also a question of technologic change. A spokesperson for Meta told Fortune that it is impossible to say exactly what Meta’s $10 billion Louisiana datacenter will power since it is unclear how AI will have evolved when it opens in 2030. Meta doesn’t know what its datacenter will do, let alone how it will monetize it. That’s a big bet to make with $10 billion.
That Meta’s datacenter won’t be open until 2030 brings up another concern with AI: cyclicality. Industries which are capital intensive, subject to intense price competition, and have long lead times to increase capacity are subject to extreme cyclicality. Shipping is a perfect example. Ships are commodities and have no pricing power. They take several years to build, so there’s a delayed response between high shipping prices and new supply coming online. When supply does come online, it all comes at once because everyone ordered ships at the same time. Prices crash, ships are scrapped, and the cycle repeats.
The AI companies are anticipating huge demand growth and trying to build as much capacity as possible. While I am sure they’re directionally right, there is a chance that they could be wrong and the market swings from undersupply today to oversupply in 2030 when all the new data centers light up. This is what happened during past industrial booms, like railroads and fiber optic cable.
Given the risks of rapid technologic change and questionable monetization, the investors giving OpenAI the money to pay Oracle are probably going to demand equity-like returns. Call it 10%. That would require $30 billion of earnings. Only the top 100 or so companies in the S&P 500 manage this.
Will investors make money in AI?
Open AI aside, its worth asking whether AI in general is good for everyday equity investors.
AT&T operates huge customer service call centers. They’re expensive and they don’t produce revenue. AT&T realizes that generative AI can replace all of its call centers for 5% of the cost. So it does it, and its margins inflate.
Verizon quickly figures out what AT&T did and does the same. Except, Verizon shares the savings with its customers by lowering prices instead of keeping it for itself. Lower prices mean Verizon starts taking market share from AT&T.
AT&T has no choice but to cut prices to match Verizon to stabilize its market share.
In the end, AT&T and Verizon end up with similar market share and margins, expect that their return on capital is lower. They had to make a big investment into generative AI merely to maintain their competitive position. The result was deflationary, so the boost to margins was short lived.
Most industries are competitive, so a scenario like this one is more likely than not. There will be some winners, but most likely the real beneficiaries will be consumers, not investors.
AI could be great for society, deflationary for prices, and akin to inflation or tax for equity investors. The best insulation to this effect, as always, is a strong competitive position that allows a degree of pricing power.
Oracle’s Big Risk
Oracle has significant customer concentration and counter-party risk with Open AI. Open AI, highly dependent on capital markets, is not the strongest counterparty. At 71x earnings, Oracle’s stock price reflects expectations for hyper growth. Any slowdown in Open AI’s funding could derail that.
Oracle used to be capital light and produce a ton of free cash flow. Now free cash flow is negative as the company rushes to build data centers. Unlike its cash-rich Mag7 competitors, Oracle has significant net debt.
It’s also worth pondering what price Oracle is getting for a $300 billion order from Open AI. Open AI has been vocally dissatisfied with capacity constraints at its partner Microsoft, but wouldn’t AWS or Google have been a more logical partner? Shouldn’t Open AI have spread their bets a little more evenly? Unless Oracle offered Open AI an they couldn’t refuse because the price we so low. Liberty points out that Oracle may “bought” its seat at the AI hyperscalers’ table by taking on a low margin contract.
EPC offers separately managed accounts, low minimums, and accepts most account types (including IRAs and 401k rollovers). Qualified Clients can elect a zero management fee structure inspired by the Buffett Partnership. EPC’s clients get complementary access to our premium Substack. To learn more please email info@eaglepointcap.com.
Disclosure: The author, Eagle Point Capital, or their affiliates may own the securities discussed. This blog is for informational purposes only. Nothing should be construed as investment advice. Please read our Terms and Conditions for further details.
Really good article! Thanks!