Nvidia's $105 Billion Guarantee Anchors OpenAI's Ohio Data Center Buildout
Nvidia's decision to backstop up to $105 billion for OpenAI's Ohio campus marks a structural shift: the chip giant is now simultaneously supplier, financier, and equity stakeholder in the infrastructure buildout it profits from.
When Nvidia agreed to backstop up to $105 billion in construction and equipment financing for OpenAI's Ohio data center campus — a commitment confirmed by Reuters and corroborated across multiple outlets on August 18 — the move did more than formalize a large capital allocation. It repositioned Nvidia as simultaneous supplier, financier, and strategic anchor for the company that has most visibly defined the commercial trajectory of large language models. Alongside the guarantee, Nvidia made a $1.5 billion equity investment in SB Energy, SoftBank's clean-energy subsidiary developing the site, tying the chip maker's balance sheet to both the hardware and the landlord in a single transaction.
The campus sits at the PORTS-Pike technology park in Ohio and is designed around approximately 8 gigawatts of compute capacity, a figure corroborated across multiple reports; one account placed total site capacity at up to 10 gigawatts. OpenAI has signed a 20-year lease with SB Energy as landlord, while Nvidia secures exclusive chip-supply rights for the installation. Jensen Huang stated alongside the announcement that the broader AI infrastructure buildout could generate potential revenues of $600 billion by 2030 for companies positioned within it — a figure that represents Nvidia's own market framing rather than an independent projection. The guarantee itself moved during negotiations: early reports cited $250 billion, later accounts brought it to $120 billion, before Reuters confirmed $105 billion — a compression that reflects active risk negotiation between the parties rather than any reduction in the project's scope.
OpenAI arrives at this commitment from a position of genuine revenue momentum. Annualized revenue has crossed $40 billion, and CFO Sarah Friar has been holding investor meetings in preparation for an IPO that could target a $1 trillion valuation, according to Benzinga. In August the company completed a $7 billion secondary share buyback that allowed current and former employees to liquidate holdings at an implied $852 billion valuation. On the compute side, OpenAI is simultaneously diversifying: GPT-5.6 Sol Ultrafast, launched this month, runs on Cerebras WSE-3 wafer-scale engines rather than Nvidia GPUs, achieving 750 tokens per second — one report cited approximately 14 times the pace of standard inference — and the disclosure sent Cerebras shares up 15 percent. The juxtaposition of a long-dated Ohio commitment built on Nvidia hardware exclusivity with active Cerebras deployment at the inference frontier suggests OpenAI is deliberately managing supply-chain concentration even as it deepens its largest infrastructure bet.
Nvidia's rationale is arithmetically coherent. By guaranteeing the financing, the company is in effect subsidizing the infrastructure through which it will sell GPUs, with hardware margins that should substantially exceed the guarantee's cost exposure — provided the campus reaches its operating targets on schedule. The $1.5 billion SB Energy stake (The Information had earlier reported talks involving up to $3 billion) gives Nvidia a direct financial interest in the site developer, aligning hardware demand with delivery execution. Broadcom has moved in parallel: the company has locked in multi-year AI chip agreements with OpenAI, Google, and Meta for custom silicon and networking, signaling that the broader chip ecosystem is now consolidating around hyperscale infrastructure commitments of this duration and scale.
The ambition of the deal, however, coexists with a set of tensions the announced numbers do not resolve. OpenAI itself stated on August 19 that it is deliberately moderating its model development pace because of cybersecurity risks, warning that forthcoming systems such as Astra may soon develop meaningful offensive capabilities — an acknowledgment that the lab's own product roadmap is subject to self-imposed constraints. A company committing to decades of compute investment while voluntarily slowing frontier model deployment faces a capital-intensity paradox that the Ohio lease makes concrete. Power is a parallel constraint: the campus's appetite — reports variously cite 8 to 10 gigawatts, with Crypto Briefing placing an initial AI-factory phase at approximately 4.25 gigawatts — requires a pace of grid integration that U.S. utilities have not previously demonstrated at this scale or speed. Anthropic, for its part, is reportedly approaching an IPO in profitable condition, adding competitive pressure to the enterprise and premium-inference revenue that underpins OpenAI's valuation projections.
Three indicators will determine whether these commitments are vindicated over the near term. First, the rate at which SB Energy advances PORTS-Pike from announced to operational gigawatts: U.S. data-center buildout history has consistently shown a wide gap between headline capacity and grid-connected reality, and power-interconnection queues remain a structural bottleneck. Second, the revenue trajectory implied by OpenAI's IPO pricing: closing the gap between $40 billion in current annualized revenue and a $1 trillion target valuation demands a sustained growth rate across exactly the period when the Ohio capital commitment is most exposed. Third, the competitive durability of Nvidia's hardware exclusivity at the campus, set against the Cerebras precedent at the inference layer: if the next generation of AI workloads shifts structurally toward fast, low-latency inference — where wafer-scale and custom-silicon architectures are proving competitive — the economics of an 8-gigawatt, GPU-centric campus will face architectural headwinds that no guarantee, however large, can preempt.