OpenAI’s recent surge has redefined the stakes in the global AI race, turning ambition into a tangible, multi-trillion-dollar infrastructure push. Over the course of a single week, the company announced a series of landmark partnerships that underscore its transformation from a model developer into a full-scale AI infrastructure powerhouse. The momentum began with Nvidia committing up to $100 billion to help OpenAI construct data centers powered by millions of GPUs. Just a day later, OpenAI expanded its Stargate project with Oracle and SoftBank, scaling the commitment to $400 billion across multiple sites and phases. On Thursday, the company sealed a strategic integration with Databricks, embedding its upcoming GPT-5 model directly into enterprise data tools and signaling a major leap in commercial adoption. These moves reflect CEO Sam Altman’s vision of OpenAI evolving into a hyperscaler—akin to Amazon Web Services or Microsoft Azure—capable of delivering AI at global scale. The company is betting big on the idea that future AI advancement hinges not just on smarter algorithms, but on access to vast computing power, energy, and physical infrastructure. Altman has repeatedly emphasized the scale of the challenge. He told reporters in San Francisco that OpenAI may eventually spend trillions on data center construction. At current plans, building 17 gigawatts of capacity would require the energy output of roughly 17 nuclear power plants—each of which takes over a decade to build. With U.S. grid capacity strained, gas turbines in short supply, and renewable projects mired in regulatory delays, the path forward is fraught with hurdles. Yet Altman remains undeterred. He highlighted the Abilene, Texas Stargate site as just 10% of what’s planned, with ten such facilities in the works. “This is 10% of what the site is going to be,” he said, gesturing at the sprawling complex. “We’re doing ten of these.” Investors are divided but largely impressed. Gil Luria of D.A. Davidson called the strategy “fake it ‘til you make it” but noted it’s working—so far. Deedy Das of Menlo Ventures said he doesn’t see the scale as crazy, but rather existential. “This is the race to superintelligence,” he said. “Access to compute and data is the new oil.” The financial reality, however, is complex. OpenAI is a non-investment-grade startup with no positive cash flow. While Nvidia’s $100 billion will come in $10 billion tranches over years, and Oracle’s $400 billion commitment is phased, the bulk of funding will come from private markets. Equity is expensive, so OpenAI is exploring debt financing, aided by Nvidia’s long-term lease structure that could improve lending terms. CFO Sarah Friar said OpenAI is also building some infrastructure in-house—not to replace partners, but to become a more efficient operator. By managing more of the stack internally, the company aims to challenge vendor pricing, reduce costs, and gain better control over delivery timelines. Monetization remains a critical challenge. While ads are not ruled out, Altman has expressed preference for affiliate-style fees—like a 2% cut when users buy something discovered via ChatGPT—without compromising model rankings. That model could help reduce burn rate and support future fundraising. Enterprise demand is surging. Accenture CEO Julie Sweet said her firm signed 37 new clients this quarter with over $100 million in bookings. “Every CEO board recognizes AI is critical,” she said. “The problem is most companies aren’t AI-ready yet.” Databricks CEO Ali Ghodsi echoed that sentiment. He stressed that while demand is growing rapidly, the world isn’t yet using AI at full capacity. “There’s going to be much more AI usage in the future than we have today,” he said. That’s why Databricks is integrating OpenAI’s models while maintaining flexibility—hosting all three major foundation models to avoid vendor lock-in. Still, the execution risk is real. The entire ecosystem—Nvidia supplying chips, Oracle building sites, OpenAI driving demand—depends on sustained coordination. Delays in energy infrastructure, regulatory approvals, or supply chains could stall progress. As Friar put it: “There’s not enough compute to do all the things AI can do, and so we need to get it started. And we need to do it as a full ecosystem.”