Nvidia's $105 Billion Guarantee Redraws the Map of AI Infrastructure Finance
Nvidia's residual-value guarantee for OpenAI's Ohio campus is unprecedented in corporate infrastructure finance, arriving precisely as Anthropic overtakes OpenAI on revenue and operating costs accelerate.
On August 18, Reuters reported that Nvidia will provide up to $105 billion in financial guarantees backing OpenAI's Ohio data center development — an arrangement that, in both scale and structure, has no clear precedent in the history of corporate infrastructure finance. Crucially, the guarantee is not tied to OpenAI's balance sheet. Instead it is anchored to the residual value of long-term infrastructure and deployed hardware — a mechanism that makes Nvidia simultaneously the principal hardware creditor and the financial backstop for what may become the world's largest purpose-built AI compute campus.
The Ohio facility, developed by SoftBank's SB Energy subsidiary at the PORTS-Pike site, is anchored by a 20-year OpenAI lease covering 8 gigawatts of compute capacity. Nvidia has separately invested $1.5 billion directly in SB Energy to lock in land, power, and shell capacity. The $105 billion guarantee reached its current form after a notable negotiation arc: early reports cited a figure as high as $250 billion, which Reuters and other outlets then tracked falling to approximately $120 billion before settling at $105 billion by August 18. Embedded in the deal are exclusive hardware supply rights for Nvidia — a detail that transforms the guarantee from passive credit support into a vertically integrated supply chain commitment. Jensen Huang has projected that the partnership could generate up to $600 billion in revenue for Nvidia by 2030, a forecast that frames just how central this single bilateral relationship has become to Nvidia's long-range revenue thesis.
The deal reflects how far OpenAI has traveled since its 2015 founding as a nonprofit research organization. The company now reports annualized revenues exceeding $40 billion, a figure CFO Sarah Friar has been presenting to investors as OpenAI accelerates toward an IPO targeting a $1 trillion valuation. IBM's $240 million joint commitment to OpenAI and Together AI, announced this week, signals how deeply the enterprise technology ecosystem is integrating OpenAI's stack. Yet the competitive backdrop is shifting in ways that complicate the narrative: Anthropic has reportedly passed OpenAI in annual revenues for the first time — a reversal that, if sustained, would mark a meaningful reordering in a market OpenAI long defined on its own terms. Anthropic is also said to be approaching its own public offering in a profitable position and on a timeline that appears to lead OpenAI's.
Even as Nvidia consolidates its role as OpenAI's infrastructure guarantor, OpenAI has been diversifying its silicon supply chain in ways worth noting. The company's GPT-5.6 Sol Ultrafast mode, launched this month, runs at 750 tokens per second on Cerebras WSE-3 wafer-scale engines — roughly 14 times standard inference throughput — routing its highest-speed tier entirely around Nvidia GPUs. Cerebras shares surged 15 percent on the disclosure. This is not a defection from Nvidia's ecosystem; at the scale OpenAI is building toward, no single chip architecture is cost-optimal for every workload type. The Cerebras arrangement suggests OpenAI is deliberately assembling the optionality to route inference by economics and latency profile rather than by single-supplier dependence, a prudent hedge for a company committing to a 20-year hardware lease anchored to one vendor.
The financial engineering of the Ohio deal carries structural risks that deserve measured scrutiny. A residual-value guarantee depends on hardware retaining meaningful value over the life of a long-term agreement — a demanding assumption in a sector where GPU generations cycle rapidly and successive architectures systematically compress the resale prices of predecessors. OpenAI faces additional cost headwinds: the company disclosed that implementing multi-stage chain-of-thought security monitoring will raise overhead on certain workloads by approximately 20 percent, a cost it has accepted to address AI cybersecurity risks it describes as growing too dangerous to leave unmonitored. OpenAI has also acknowledged deliberately pacing model development, citing concerns that near-term systems like the forthcoming Astra may acquire meaningful offensive cyber capabilities. Rising fixed infrastructure costs, intentional restraint at the capability frontier, and a rival now outpacing it on revenue combine to make the arithmetic supporting a $1 trillion IPO valuation more demanding than headline revenue figures alone would suggest.
Three concrete signals will determine whether this infrastructure bet delivers on its promise. First, whether OpenAI's revenue growth can reliably outpace the capital commitments now accumulating around the Ohio campus — fixed costs are indifferent to AI market cycles, and a 20-year lease provides little flexibility in a demand downturn. Second, how hardware residual values evolve as Nvidia's next-generation architectures arrive at scale; if depreciation accelerates beyond the guarantee's structural assumptions, the arrangement's risk profile shifts materially toward Nvidia. Third, whether Anthropic's reported revenue lead proves durable or represents a momentary inversion in a market still sorting its long-term winners — a confirmed profitable IPO from Anthropic ahead of OpenAI's own offering would narrow the window for the Ohio campus to begin returning capital before the public markets render a verdict on both. For now, the $105 billion guarantee is the clearest signal yet that Nvidia and OpenAI have decided to bind their institutional fortunes together at a scale that leaves little room for either to walk away.