OpenAI’s CFO Just Explained Why Nvidia Is No Longer the Only Option
OpenAI's CFO just revealed how $122 billion in cash fundamentally changes the power dynamic between the world's most important AI lab and its chip suppliers, and the implications for Nvidia's premium pricing are only beginning to surface.
Sarah Friar, chief financial officer of OpenAI, sat with CNBC’s Jim Cramer and described her chip procurement in the language a corporate treasurer would use. “We also have a strategy for diversifying our supply chain. And that is just good CFO risk mitigation. You never know when someone’s supply chain is going to get gummed up. So you need to have multiple providers.”
That framing is the clearest confirmation yet that NVIDIA‘s (NASDAQ:NVDA | NVDA Price Prediction) most important private customer now treats the company as one qualified supplier among several rather than the default option.
For shareholders, the consequence hits price and margin long before it hits revenue. OpenAI is still buying enormous quantities of NVIDIA silicon and has committed to roughly 12 gigawatts of NVIDIA compute through 2030.
What has changed is negotiating posture, and posture is what compresses a premium multiple.
Training Versus Inference Is Splitting Into Two Markets
Friar drew a clean line between two workloads that used to be treated as one. “Nvidia is still an incredible platform of accelerators for training. But in jalapeno’s case, that is a chip very focused on inferencing because it is set up exactly for our models. So therefore it is very efficient for you.”
Training a frontier model happens episodically and rewards general-purpose silicon. Inference runs on every user prompt afterward, and its economics reward chips built around one model’s operations.
A chip designed around one lab’s transformer stack can remove unused circuits, dedicate more die area to the operations that matter, and reduce memory movement. Lower cost per token at comparable throughput follows.
Friar’s point is that NVIDIA owns the half of the workload that ends when training ends, while purpose-built silicon owns the half that compounds with every query.
Balance Sheet Is Where the Leverage Sits
OpenAI is a private company, and every figure Friar cites about it is her own and has not been independently verified. That said, the number she volunteered is the point.
Sarah Friar said OpenAI raised $122 billion in the first quarter of this year, describing the balance as “cold, hard cash sitting there.”
Cash of that magnitude removes the two levers a supplier normally uses to hold pricing: urgency and dependency. A buyer without a funding cliff will wait a quarter for better terms.
She added that “An IPO is just a milestone in the journey. I’m going to keep reiterating that.” No listing deadline forces OpenAI to lock in supply at any price it can get.
Five Sessions Have Already Sorted the Three Names
Trailing five sessions tell three different stories. NVIDIA is down 5.9%, AMD (NASDAQ:AMD) is off 0.3%, and Broadcom (NASDAQ:AVGO) has dropped 7.95%. A single week is a small sample.
Broadcom fell hardest because its custom-accelerator revenue is the most concentrated by customer. Q3 AI semiconductor revenue reached $16.7 billion, up 221% year over year, and any pause in one lab’s ramp lands directly on the forward guide.
NVIDIA’s decline matters more because it follows a quarter in which revenue reached $96.22 billion, up 105.8% year over year, with a guide of $108 billion for the current quarter. Pricing power inside that growth is now the question.
AMD held roughly flat as its Instinct roadmap and gigawatt-scale deals with Anthropic and Meta get validated one contract at a time.
Bull and Bear Case for NVDA Stock
The bull case rests on the platform argument the NVIDIA chief executive made on the most recent earnings call. NVIDIA is the only architecture that runs every frontier model in every cloud; its opportunity per gigawatt has expanded from roughly $18 billion with Hopper to $40 billion with Vera Rubin, and demand exceeds supply through at least fiscal 2028. On those inputs, a forward earnings multiple of 24x is defensible, and the underlying figures are laid out in the most recent 8-K.
The bear case is what Friar described. Large AI customers are commissioning inference-specific silicon while still buying training capacity from NVIDIA and AMD, compressing NVIDIA’s share of the fastest-growing workload category even as unit volumes rise. Gross margin, guided to bottom in the 71% to 72% range in the fourth quarter, is where that pressure surfaces first.
The variable that decides between them is inference mix. If custom accelerators prove out at roughly half the cost of a GPU across more than one lab, NVIDIA’s premium contracts before revenue growth ever slows.
Contact [email protected] for any questions or corrections.
Omor Ibne Ehsan is a writer at 24/7 Wall St. He is a self-taught investor with a focus on growth, cyclical, and dividend equities that have strong fundamentals, value, and long-term potential. He also has an interest in high-risk, high-reward investments such as penny stocks.
