Nvidia investing $1.5B in SoftBank data center developer behind OpenAI project | TechCrunch

Nvidia announced on August 17, 2026 that it will invest $1.5 billion in SB Energy — the SoftBank subsidiary developing OpenAI's forthcoming Ohio data center — while simultaneously extending up to $105 billion in credit to help build the facility. The deal, confirmed across reporting from TechCrunch, The Verge, and Nikkei Asia, includes a sole-supplier clause that positions Nvidia as the exclusive hardware provider for the project. At that combined financial exposure, Nvidia is no longer just a chip vendor to the AI industry — it is becoming a co-owner of the infrastructure that runs on its own chips.

What's Converging

The last eighteen months have seen every major AI lab race to lock in compute at scale, but the architecture of those deals has quietly shifted. Early rounds of hyperscaler buildouts — Microsoft's OpenAI partnership, Google's TPU expansion, Amazon's Trainium push — were primarily about deploying existing capital into new capacity. What's different now is that the chip makers themselves are moving upstream into infrastructure finance. When a semiconductor company starts writing nine-figure checks to the entity building the building, the dynamic between supplier and customer collapses into something harder to categorize.

Energy has become the second bottleneck that nobody announced a press release about. Data center developers are now competing directly with LNG export terminals for natural gas supply in the same regional markets, and construction costs for new gas generation capacity have risen sharply — by some estimates more than 60% in two years. That's not an abstract macroeconomic signal; it shows up as real dollar figures in deals like this one, where the power generation infrastructure attached to a single AI facility runs into the tens of billions before a single GPU is racked. The compute spend gets the headlines, but the energy spend is quietly approaching the same order of magnitude.

SoftBank's posture throughout this period has been worth tracking. SoftBank and OpenAI have had a deepening financial relationship since at least the Stargate initiative earlier in 2026, and SB Energy — the SoftBank entity at the center of this new deal — is not a startup. It is an established developer with existing infrastructure relationships, which is part of why OpenAI signed a lease with it rather than building directly. Nvidia choosing to buy into that entity, rather than simply negotiate a supply contract, signals something about how chip makers now think about demand stability: equity is a stronger lock than a purchase order.

The Specific Development

The facility at the center of the deal sits near Cincinnati, Ohio, and is being designed for a total capacity of eight gigawatts — a number large enough that multiple reports felt the need to contextualize it against national grid comparisons. The Verge's coverage confirms the 8 GW figure and notes that the first 800 megawatts are expected to come online in 2028. That's a two-year runway before the first phase goes live, which matters for understanding Nvidia's financial exposure: the $105 billion isn't a payment today, it's a credit commitment structured to fund construction over time as milestones hit.

The $1.5 billion equity stake in SB Energy is separate from that credit line and works differently. Equity means Nvidia participates in SB Energy's upside — or downside — as a business, not just as a debt holder. Multiple reports confirm both numbers, though neither Nikkei Asia nor The Verge independently verified the sole-supplier clause in the same detail as the primary TechCrunch reporting. That clause is the piece worth watching most closely: it means OpenAI cannot, within the terms of this deal, source competing hardware for this facility. AMD, Google TPUs, and any future entrant are structurally locked out of what is shaping up to be one of the largest single AI compute deployments ever built.

The energy story attached to this project adds a layer that the chip-focused headlines tend to skip. The natural gas plant co-located with the data center carries a reported price tag of roughly $33 billion for 9.2 gigawatts of generation capacity — a figure that is, by the primary reporting's account, approximately 66% higher than comparable gas generation would have cost two years ago. The driver is competition: AI data center developers and LNG export operators are drawing from the same gas supply pool in the same regions, and once a facility of this scale comes online, it could materially shift regional gas prices.

Our read is that the energy angle is underweighted in the current coverage cycle. Developers and AI practitioners focused on API access and model capability tend to treat power infrastructure as an ops abstraction. But when a single data center requires the equivalent of a medium-sized city's power generation, and when that generation is getting more expensive faster than model efficiency gains can offset it, the economics eventually flow downstream into API pricing and availability. The Ohio facility is a useful forcing function: it makes the energy constraint concrete in a way that abstract GW projections do not.

What's Likely Next

The immediate question is regulatory. A $105 billion credit commitment from a chip maker to a data center developer that is also its captive customer is exactly the kind of vertical arrangement that draws antitrust scrutiny, particularly in an environment where competition regulators in the US and EU have both signaled heightened attention to AI market concentration. Whether the sole-supplier clause specifically becomes a focal point for review — or whether the scale of the credit line does — is worth watching over the next 30 to 60 days as the deal completes its disclosure cycle.

The second thing to track is whether the 2028 first-phase target holds. Eight hundred megawatts in 2028 requires breaking ground on both the data center and the gas generation plant on a timeline that leaves limited margin for permitting delays, materials shortages, or grid interconnection disputes. Any slip in the energy infrastructure side would push out the compute availability side regardless of how many GPUs Nvidia produces. For developers whose roadmaps depend on OpenAI's API capacity scaling to support increasingly agentic and inference-heavy workloads, the Ohio facility's construction schedule is a more meaningful signal than any model release date announced between now and then.

Sources