Nvidia Invests $1.5 Billion in SoftBank Data Center Developer as Its Balance Sheet Becomes a Competitive Weapon
Nvidia's investment in a SoftBank-backed data center developer to secure OpenAI Stargate capacity reveals a company that has moved well beyond selling chips — and that now carries the systemic risks of the infrastructure cycle it is helping to finance.
Nvidia's $1.5 billion investment in a SoftBank-backed data center developer — executed to secure compute capacity for OpenAI's Stargate project — is less a capital allocation decision than a strategic declaration. For a company that began as a graphics card maker and spent decades competing on silicon performance, the move encapsulates a dramatic broadening of competitive scope: Nvidia is now in the business of financing the infrastructure that runs its chips.
The Stargate deal is not isolated. Nvidia has reportedly backed as much as $105 billion in credit for OpenAI's Ohio data center campus and announced a $500 billion AI financing plan that credit markets received as stabilizing. The company co-invested with Coatue in Firmus, a $2 billion raise targeting Asia-Pacific compute capacity, and took a strategic equity stake in Cloverleaf Infrastructure, a developer focused on power-enabled data center land. According to reports, it is also in advanced talks for a roughly $3 billion equity investment in SB Energy to secure long-term renewable power, while AI cloud operator IREN has disclosed a $3.4 billion multi-year contract with Nvidia. Viewed together, these moves reveal a company engineering demand certainty into its supply chain — locking in customers through capital commitments, not product specs alone.
The financial transformation underlying these moves is stark against Nvidia's own history. Capital expenditure stood at $138.7 million in FY2011; FY2027 Q1 revenue alone reached $81.6 billion, with data center revenue up 92% year-over-year. The Vera Rubin platform, the next-generation architecture succeeding Blackwell, is in active H2 2026 production ramp: Microsoft has received first systems, CoreWeave is provisioning Rubin hardware for HRT, and IBM with Together AI signed a $240 million multi-year deal for Nvidia B300 inference clusters. The earnings call expected around August 26 — where analysts project roughly $91 billion in quarterly revenue — will test whether that demand can sustain the current infrastructure investment cycle, or whether early saturation signals are beginning to surface.
Against this expansion, several structural constraints are tightening simultaneously. The global memory shortage is forecast by industry analysts to persist into 2028, and TrendForce has flagged the possibility that Nvidia may reduce HBM specifications on Rubin Ultra due to supply constraints — a material risk if confirmed. Nvidia has reportedly locked multi-year exclusive supply agreements with major memory manufacturers, suggesting management shares this concern. The company is also reportedly exploring the acquisition of Korean AI chip startup Rebellions at approximately $2.3 billion, a move that would deepen vertical integration in custom silicon. On the competitive perimeter, Qualcomm has unveiled data center AI chips targeting Nvidia and AMD directly, and Cerebras claims its CS-4 server outperforms Nvidia GPUs by 30x on chatbot inference workloads — a figure that invites independent scrutiny but underscores that the competitive landscape is not static.
The China variable adds a distinct and unresolved layer of uncertainty. President Trump stated this week that China has not yet approved purchases of the H200, noting Beijing's preference for domestic alternatives. While H200 units are reportedly entering China under case-by-case import licenses, domestic accelerator makers have already secured meaningful market share, and Nvidia's ability to serve what was once a significant revenue source remains policy-constrained. Nvidia formally denied reports of a year-end China-specific LPU and a separate claim about a dedicated China chip roadmap. Separately, customers have reportedly been warned of AI-related server price increases exceeding 15% — a move that protects margins but creates friction at a moment when competitive alternatives are proliferating. Nvidia shares traded at $214.72, down roughly 1%, reflecting investor caution ahead of the guidance update.
The picture that emerges is of a company in a self-reinforcing cycle of demand creation, but one increasingly exposed to the systemic risks that cycle produces. The core opportunity remains substantial: hyperscaler and sovereign AI buildouts are accelerating, the Vera Rubin ramp is proceeding, and Nvidia's allocation power — demonstrated again by SpaceX choosing it over AMD — remains formidable. But the financing model carries concentration risk: balance sheet exposure in infrastructure deals means Nvidia's fortunes are growing more tightly bound to the success of its customers' projects. Three signals merit close attention in the weeks ahead: the August 26 earnings guidance and any commentary on China-specific revenue trajectory; the resolution of H200 import policy and potential LPU approvals; and HBM4 supply availability, which will determine whether the Rubin Ultra roadmap proceeds on schedule or requires the specification adjustments that TrendForce has flagged.