OpenAI’s CFO stated that Nvidia is no longer the sole compute option, indicating strategic diversification beyond proprietary accelerators.

OpenAI’s chief financial officer recently revealed how a $122 billion cash reserve fundamentally shifts the leverage dynamic between the world’s leading artificial intelligence laboratory and its semiconductor suppliers. The implications for Nvidia’s premium pricing strategy are only beginning to materialize.

This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.

During an appearance on CNBC with Jim Cramer, OpenAI CFO Sarah Friar framed the company’s chip procurement strategy using the precise language of corporate treasury management. “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.”

This framing serves as the clearest confirmation yet that NVIDIA ( NASDAQ:NVDA | NVDA Price Prediction )—the most critical private customer for the chipmaker—now treats the company as one qualified vendor among several, rather than the default option.

For shareholders, the impact will register in valuation multiples and profit margins long before it affects top-line revenue. OpenAI continues to purchase massive volumes of NVIDIA silicon and has committed to roughly 12 gigawatts of NVIDIA compute through 2030. What has shifted is the buyer’s negotiating posture, and posture alone is sufficient to compress a premium valuation multiple.

Training Versus Inference Is Splitting Into Two Markets

Friar drew a sharp distinction between two computational workloads that were previously treated as a single market. “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 occurs episodically and benefits from general-purpose silicon. Inference, which executes on every subsequent user prompt, operates under entirely different economic rules that reward chips engineered around a specific model’s architecture. A processor designed for a single lab’s transformer stack can eliminate unused circuitry, allocate more die area to high-priority operations, and minimize memory bandwidth bottlenecks. This design efficiency directly lowers the cost per token at comparable throughput levels.

Friar’s underlying point is straightforward: NVIDIA dominates the workload segment that concludes once training finishes, while purpose-built silicon captures the segment that compounds with every query.

Balance Sheet Is Where the Leverage Sits

As a private entity, every figure Friar cites regarding OpenAI is self-reported and remains independently unverified. Nevertheless, the disclosed number establishes the core dynamic. Friar stated that OpenAI raised $122 billion in the first quarter of this year, characterizing the balance sheet as “cold, hard cash sitting there.”

Capital of this magnitude neutralizes the two primary leverage points suppliers typically use to maintain pricing: urgency and dependency. A buyer facing no immediate funding cliff can afford to wait a quarter for more favorable terms. Friar further emphasized that “An IPO is just a milestone in the journey. I’m going to keep reiterating that,” noting that no listing deadline forces OpenAI to lock into supply agreements at whatever price the market dictates.

Five Sessions Have Already Sorted the Three Names

Trailing five trading sessions reveal three distinct market narratives. NVIDIA is down 5.9%, AMD ( NASDAQ:AMD ) is off 0.3%, and Broadcom ( NASDAQ:AVGO ) has dropped 7.95%. While a single week represents a limited sample size, the directional moves reflect shifting investor sentiment.

Broadcom experienced the steepest decline because its custom-accelerator revenue stream is highly concentrated by customer. Third-quarter AI semiconductor revenue reached $16.7 billion, a 221% year-over-year increase, meaning any slowdown in a major lab’s deployment ramp lands directly on the company’s forward guidance.

NVIDIA’s pullback carries greater weight given its recent quarterly results, which saw revenue hit $96.22 billion, up 105.8% year over year, alongside a current-quarter guide of $108 billion. The central question for investors is whether pricing power within that growth trajectory remains intact.

AMD remained roughly flat as its Instinct accelerator roadmap and gigawatt-scale supply agreements with Anthropic and Meta are being validated contract by contract.

Bull and Bear Case for NVDA Stock

The bullish thesis rests on the platform dominance argument articulated by NVIDIA’s CEO during the latest earnings call. NVIDIA remains the sole architecture capable of running every frontier model across every major cloud provider. Its addressable opportunity per gigawatt has expanded from roughly $18 billion with Hopper to $40 billion with Vera Rubin, and demand continues to outpace supply through at least fiscal 2028. Given those fundamentals, a forward earnings multiple of 24x remains defensible, with supporting metrics detailed in the most recent 8-K filing.

The bearish thesis mirrors Friar’s earlier remarks. Major AI developers are actively commissioning inference-specific silicon while maintaining their training capacity purchases from NVIDIA and AMD. This trend compresses NVIDIA’s share of the industry’s fastest-growing workload category, even as total unit volumes rise. The initial pressure point will surface in gross margins, which are currently guided to settle in the 71% to 72% range during the fourth quarter.

The decisive variable between these two scenarios is inference mix. If custom accelerators demonstrate viability at roughly half the cost of a comparable GPU across more than one major lab, NVIDIA’s premium pricing contracts well before overall revenue growth decelerates.

Contact [email protected] for any questions or corrections.