Nvidia Allies With Six Wall Street Giants to Finance $500 Billion in AI Infrastructure

By establishing an independent compute financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, Nvidia is moving beyond hardware supply into the business of subsidizing the demand for its own products.

On August 12, Nvidia announced it had assembled a coalition of six of Wall Street's largest asset managers — Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs, and KKR — to establish what the company describes as an independent compute financing platform targeting more than $500 billion in AI infrastructure deployment. The announcement crystallizes a strategic pivot that goes well beyond hardware sales: Nvidia is now in the business of manufacturing demand, not just manufacturing chips.

The financing mechanism addresses a structural friction that has quietly threatened to cap the AI buildout. Even the largest cloud providers and enterprises face capital constraints when committing to GPU clusters at the scale that Nvidia's hardware increasingly demands. Kyber racks — Nvidia's next-generation compute enclosure — are reportedly priced at $41.6 million per unit, equipped with 340.4 terabytes of DRAM and HBM4E memory priced at roughly $19.76 per gigabyte, according to Wccftech. By channeling institutional capital directly to GPU customers at competitive rates, Nvidia effectively subsidizes the demand side of its own market, creating a reinforcing loop: cheaper financing lowers the barrier to deployment, which accelerates orders, which sustains the revenue trajectory.

That trajectory has already reached a scale few foresaw just three fiscal years ago. Nvidia's data center segment crossed $75 billion in Q1 FY2027, up 92 percent year-over-year, with Q2 guidance set at $91 billion — a figure that would have equaled the company's entire annual revenue several times over in the early AI era. The company posted roughly $26.9 billion in fiscal year 2022 revenue; it has since compounded at a rate that has reshaped the entire semiconductor industry's self-understanding. Vera Rubin, the current-generation architecture, is now in full production, and the roadmap extends through Rubin Ultra and Feynman toward 2030. Beyond the financing program, this week's news underscores how comprehensively Nvidia is rewiring its strategic relationships: SpaceX has reportedly committed exclusively to Vera Rubin for orbital AI data centers and plans to scale to 10 gigawatts of compute by the end of 2027; Nvidia is committing up to $3 billion to Lancium, a Texas energy developer central to the Stargate infrastructure, addressing power as a binding constraint; and the Financial Times reported Nvidia as the mystery anchor tenant of Hut 8's Beacon Point campus in Texas under a 15-year lease valued at up to $50.2 billion — a figure that, if accurate, would rank among the largest real estate commitments in corporate history and warrants independent confirmation. IBM and Together AI separately formalized a $240 million multi-year agreement to build a large-scale NVIDIA HGX B300 inference cluster on IBM Cloud, extending Nvidia's customer base into regulated enterprise workloads.

Yet the same week that Nvidia's alliance with Wall Street capital reinforces its dominance in Western markets, the structural headwinds in China are sharpening. Multiple analysts now forecast that Nvidia's AI chip market share in China could collapse from roughly 40 percent to as low as 8 percent within two years, as Huawei's domestically developed accelerators gain scale under sustained U.S. export controls. That is not a rounding error: China was a meaningful revenue segment before controls tightened, and its accelerating displacement by indigenous hardware represents a durable, policy-driven headwind rather than a cyclical one. HBM supply constraints add a second layer of pressure: Nvidia is reportedly testing Rubin Ultra configurations with as little as 192 gigabytes of memory — stepping back from HBM4E to HBM4 — as soaring memory costs force design flexibility even as Samsung reaches 80 percent yield on HBM4. Consumer GPU prices rose as much as 39 percent across the RTX 50 series, a signal that Blackwell-era supply tightness has not resolved. On the competitive horizon, Microsoft is expected to unveil its Maia 300 inference chip in September, Anthropic has announced plans for custom silicon, and Broadcom's strong market performance this week reflected the market's recognition that hyperscaler custom chip programs represent a credible long-run alternative to merchant GPU procurement for specific workloads.

Nvidia is simultaneously moving up the software stack. The Nemotron 4 model at one trillion parameters and the open-weight Nemotron 3.5 Lightning represent the company's most direct assertion yet that it intends to compete at the model layer, not merely supply the compute substrate. Whether this expansion deepens ecosystem lock-in or creates friction with the major AI labs that are simultaneously its largest GPU customers remains an open question. Three signals will be particularly informative in the months ahead: the Q2 FY2027 earnings call on August 26, which will test whether the $91 billion guidance holds and provide first color on Rubin Ultra ramp dynamics; the pace of Huawei's Ascend shipments into Chinese enterprises through the remainder of the year, which will calibrate how quickly the China displacement scenario accelerates; and whether the $500 billion financing platform converts into genuinely incremental GPU deployments rather than refinancing of existing ones — a distinction that matters significantly for determining whether Nvidia's demand-creation strategy represents real market expansion or simply pull-forward.