Scaling the Hype: NVIDIA × OpenAI × Oracle
Data centers won’t cure cancer; they’ll scale enterprise AI
This is a very special article - a two-part collaboration with
by
!The sister post by
is OpenAI: a method to the madness?It’s not just some post! It’s a post that presents the opposite take to mine on this topic and advocates for OpenAI and their friends. Read this post as well for the full picture -and subscribe to my opponent:
Recently, OpenAI, Oracle, and Nvidia announced deals that involve paying each other hundreds of billions of dollars. For someone not connected to finance, the logical question is: how in the world can a company like OpenAI, whose actual revenue is in the single to low double-digit billions, commit to paying $200 billion? Where does this money come from? This is the mystery of the AI economy. The trend is concerning — especially given that Deutsche Bank has warned of an AI bubble:
What Happened
Around three weeks ago, Oracle’s stock surged after it announced that customer bookings, essentially contracted obligations for future services, grew by over 300% compared to last year. It turned out that most of this growth came from one single customer: OpenAI, which signed a deal with Oracle to provide cloud computing services worth $300 billion. Why did OpenAI choose Oracle? As Fortune writes:
Part of the reason Oracle was able to strike the deal with OpenAI at all is due to Ellison’s courting of Nvidia CEO Jensen Huang, which has allowed his company, despite previously trailing behind other cloud providers, to secure a large stockpile of top-of-the-line Nvidia GPUs and position itself as a significant player in the AI infrastructure space.
In short: Nvidia agreed to sell (or lease) Oracle a massive stockpile of GPUs, which remain extremely scarce and valuable.
While many wondered how a company that, even under the most optimistic projections, has only about ~$12 billion in annual revenue could afford all this, Nvidia announced a multi-year plan to deploy around 10 gigawatts of systems — equivalent to millions of GPUs and worth up to $100 billion — for OpenAI. However, this does not solve the mystery of how OpenAI is going to pay Oracle, since Nvidia is investing systems, not cash. The benefits of Nvidia are also not that clear:
Under the deal, Nvidia will be taking a financial stake in one of its largest customers, but without receiving any voting power in return, according to a person close to OpenAI.
And as we know Oracle is buying chips from Nvidia. It bought chips for just one OpenAI dataset in Texas for $40 billion and the plans are to build at least five.
This is a simplified circle of money. The important part is that Nvidia’s $100 billion is not cash but Nvidia systems. This investment cannot be used to cover the costs of the Oracle deal, which is valued at $300 billion over the next five years. Oracle has confirmed it will pay $40 billion to Nvidia for chips for a single data center in Texas and there are plans to build at least five such facilities.
I cannot validate whether this is financially feasible, but what I can do is look at what OpenAI promises to deliver once these investments are realized and assess the feasibility of those promises.
Further in the post:
What the money is to be spent on and what deliverables are promised
How feasible those deliverables are
What it is most likely all about
Bold predictions about why big innovations will not come out of this initiative and who will be the key innovators instead
Paid subscribers get access to the full archive, quarterly roundtables, and priority individual chats to consult on AI topics.
Founding subscribers receive a 45-minute one-on-one call to discuss the AI topic of their choice—whether strategy, boosting developer productivity, identifying meaningful use cases, etc.—plus top-priority access in individual AI chats.
Also buy an item in AI realist shop and get promoted to paid subscription for a month for each purchased item:
https://airealist.myshopify.com/