Nvidia reportedly may provide up to $105 billion of credit support for the land, power, and buildings behind OpenAI’s planned Ohio facility. What changes when a chip supplier begins helping finance the infrastructure required to use its chips?
Demand increasingly looks real. But Nvidia and Amazon are making commitments that extend far beyond the next product cycle, just as long term interest rates raise the hurdle for every dollar invested.
Nvidia Is Moving Beyond Selling Chips
Nvidia reportedly invested $1.5 billion in SB Energy and agreed to provide credit support capped at $105 billion for OpenAI’s planned Ohio facility. The support covers the land, power, and shell.
The strategic logic is clear. If power and buildings constrain compute demand, Nvidia can help remove those constraints.
But selling chips is one business. Investing in energy and supporting credit for a customer’s physical infrastructure introduces materially different risks.
The planned campus is expected to begin adding capacity in 2028, with completion expected by 2032 under a 20 year OpenAI lease. Nvidia is therefore taking exposure to infrastructure whose economics extend far beyond a typical product cycle.
The Missing Terms Prevent a Return Judgment
The lease payments, financing triggers, ownership terms, and Nvidia’s potential loss exposure have not been provided. Without them, the economics cannot yet be judged.
The $105 billion cap measures the potential scale of the commitment. It does not reveal the expected return Nvidia will earn for accepting that exposure.
Demand Is Becoming Harder to Dismiss
These investments would be more concerning if demand remained mostly theoretical. The operating evidence is becoming difficult to dismiss.
Anthropic’s reported revenue run rate increased from $9 billion at the end of 2025 to more than $65 billion in July 2026. Preliminary latest quarter revenue exceeded $11.5 billion, compared with $787 million a year earlier, while adjusted operating income was positive.
Fabrinet reported Q4’26 revenue of $1.32 billion, up 45% year over year, and adjusted earnings per share of $4.10, up 55%.
Higgsfield’s annualized revenue reportedly reached $700 million in August, up from $20 million a year earlier. CBRS reportedly contracted 750 megawatts through 2028 as the exclusive compute backbone for OpenAI’s GPT-5.6 Sol Ultrafast mode.
Nvidia guided to $91.0 billion of revenue for its August 26 earnings, up 96% from $46.7 billion, with approximately 75.0% non-GAAP gross margin. That guidance assumes zero China data center revenue. Vera Rubin shipments are expected to begin this quarter under continued supply constraints.
Together, those figures suggest utilization may be catching up with spending. Software revenue is accelerating, infrastructure suppliers are growing, and compute capacity is being contracted years in advance.
Real Demand Does Not Validate Every Buildout
Amazon reportedly increased its planned Louisiana data center investment from $12 billion to $18 billion and added a third campus.
Current operating momentum may justify major expansion. It does not prove that every project built to serve AI demand will earn attractive returns, especially when construction, capacity, and lease commitments extend through 2028, 2032, and across 20 years.
That distinction matters because the cost of capital is moving against the buildout. The 30 year Treasury yield reportedly reached 5.3%, its highest level in more than 19 years or since 2007. The 10 year yield rose above 4.7%.
Higher long term rates raise the hurdle for power generation, data centers, multiyear construction, and assets whose value depends heavily on distant cash flows.
The Risk Has Shifted to Capital Allocation
The central question is no longer whether more compute will be consumed. Investors also need to understand who owns the infrastructure, who finances it, what activates the guarantees, how durable utilization will be, and who absorbs the loss if projected demand arrives later than the capital.
AI demand looks increasingly real. What remains unproven is whether financing that demand through long leases, credit support, power generation, and data centers will produce durable shareholder returns at today’s cost of capital.