Stargate represents an immense power move.

OpenAI CEO Sam Altman has spent the past year attempting to secure an unprecedented amount of computing power to train the company’s AI models. Reports suggest that Japanese officials even laughed at the sheer volume of electricity he requested. The stakes were clear: without vast computing resources, OpenAI risked falling behind major tech players like Google and Meta, who have spent years building robust AI infrastructure.

However, last week, this seemingly unattainable goal turned into an official announcement. Altman managed to secure a staggering $500 billion commitment to develop OpenAI’s vast data center network, known as Stargate, with financial backing from SoftBank, Oracle, and Abu Dhabi’s MGX fund. The White House endorsed the venture in a press conference, where President Donald Trump stood alongside Altman, Oracle cofounder Larry Ellison, and SoftBank CEO Masayoshi Son. (The name references the 1994 sci-fi film where a stargate functions as an ancient transport portal under the control of a supreme ruler.)

If realized, Stargate could become the most extensive private computing infrastructure project in history. It would involve a network of enormous computing complexes, each covering hundreds of acres and consuming as much electricity as a small city. These facilities would require their own electrical substations and transmission lines, making the project not just about AI dominance but also about controlling a substantial portion of America’s energy infrastructure.

The deal depended on a precise sequence of events: Microsoft relaxed its grip on OpenAI after Altman’s brief ousting, Trump signaled to Silicon Valley that he was open for business, and tech’s major players eagerly sought involvement in high-profile AI projects. The agreement also highlights Altman’s strategic acumen—he, a progressive San Francisco tech leader, navigated an administration that opposes many of his publicly held views and managed to secure a major victory. By leveraging Trump’s hunger for successes, SoftBank’s enthusiasm for hype, and fears surrounding the AI arms race, he orchestrated an unparalleled computing deal for OpenAI. The financial structure behind it is intricate, making it difficult to trace the money flow.

Yet, despite the grand vision, Stargate’s foundation appears shaky. Altman has secured only a fraction of the funds he initially sought, particularly from investors like SoftBank, known for prioritizing hype over actual results. Meanwhile, a Chinese startup, DeepSeek, recently developed an OpenAI-level model using significantly fewer resources, stealing the spotlight from Altman’s announcement. Even among those typically eager to support his ventures, skepticism is growing.

“We’ll need significantly more compute,” Altman reportedly told policymakers this week after DeepSeek’s breakthrough dented Nvidia’s market value, raising fresh doubts about Stargate’s feasibility. “There is real competition in the world, and we’re eager to take the next step.”

The financials don’t quite add up.

Though Stargate has been marketed as a hallmark Trump-era project, it actually originated last year as a Microsoft-OpenAI collaboration for a massive data center. However, their partnership became strained as Microsoft grew wary of its reliance on OpenAI and OpenAI struggled with rising computing costs. In response, Altman sought funding from a broader array of investors rather than relying solely on Microsoft’s deep pockets.

OpenAI and Microsoft insist their partnership remains intact. “Absolutely not!” Altman declared on X when asked if they had split, adding, “Very important and huge partnership, for a long time to come. We just need more compute.” Microsoft CEO Satya Nadella echoed this sentiment in a CNBC interview, describing it as an “adjustment” made to support Altman’s vision while ensuring Microsoft’s strategic interests remained protected—a corporate way of saying they reduced their financial risk.

Nevertheless, Microsoft either couldn’t or wouldn’t fund the infrastructure at the scale OpenAI desired, prompting Altman to seek alternative backers. However, industry analysts remain unconvinced by the financial feasibility of the project. Stargate requires $100 billion in its first year alone, yet its investors’ liquid capital falls far short. SoftBank holds $30 billion in cash reserves, MGX manages a $100 billion fund (part of which is already allocated to OpenAI and could soon include competitors like xAI), and Oracle possesses only $11 billion in cash.

Meanwhile, OpenAI, possibly the best-funded startup in history, continues to burn through cash at an alarming rate. Financial records reviewed by The New York Times indicate that the company expected to spend around $5 billion last year, including major expenses like employee stock compensation. While Altman’s exceptional dealmaking skills may secure the initial $100 billion, it remains unclear how he intends to multiply that fivefold.

Moreover, SoftBank’s investment history raises concerns—just look at the Vision Fund’s disastrous bet on WeWork.

“All I know is I’m good for my $80 billion.”

OpenAI and SoftBank will each contribute $19 billion to Stargate, according to reports from The Information. While they will be the largest stakeholders, their exact ownership structure remains unclear. Altman has described them as general partners (GPs), implying they will wield substantial control rather than merely holding equity stakes. OpenAI and SoftBank are expected to each hold around 40% of the project, with Oracle and MGX taking smaller shares. The total initial investment amounts to $45 billion.

The remaining funds are expected to come from external investors and debt financing, potentially in the form of bonds. SoftBank is no stranger to debt—it already carries $150 billion in liabilities. Additional capital from the United Arab Emirates is also anticipated. However, the gap between $45 billion and $500 billion remains vast, even with aggressive capital-raising strategies.

Elon Musk, who cofounded OpenAI but left after a leadership dispute, wasted no time in voicing his skepticism. “They don’t actually have the money,” Musk posted soon after the announcement. “SoftBank has well under $10B secured. I have that on good authority.”

Altman pushed back, countering that Musk’s claim was “wrong, as you surely know,” and asserting that Stargate’s first site is already under development. Yet, doubts persist within tech circles. Nadella evaded financial questions in the CNBC interview, stating that he is “not in the details” of Stargate’s funding. Pressed further, he only confirmed Microsoft’s annual $80 billion Azure investment, quipping, “All I know is I’m good for my $80 billion.”

As concerns grow, increasingly ambitious promises emerge in Stargate’s promotional materials: AI-powered cancer vaccines designed in 48 hours, over 100,000 new American jobs, and US dominance in artificial intelligence. The venture aims to merge AI with biotechnology to revolutionize medical research while bolstering national security. The vision is vast—positioning AI as a transformative force for human progress.

“This is the beginning of a golden age,” Son proclaimed during the press conference.

Not easy, fast, or cheap

While $500 billion may seem like a substantial amount, it’s relatively modest when considering the ambitious goals of Stargate.

Kent Draper, a leader at data center operator IREN, points out that Stargate’s financial targets make sense in an ideal scenario. With a first-year budget of $100 billion, a significant number of data centers could be built. However, the challenge lies in the uncertainty around how quickly and effectively that budget can be utilized.

Many current data center facilities are not designed to accommodate AI infrastructure, especially due to its massive power demands. A typical server rack — resembling a bookshelf of computers — uses about five to 10 kilowatts of electricity, which is equivalent to the daily energy consumption of an average American home. In contrast, a modern AI GPU rack can draw anywhere from 45 to 120 kilowatts, consuming as much electricity in a single day as several apartments, or even an entire small apartment building. Current facilities need at least 100 megawatts of power, and Draper estimates that the next generation will demand 200 kilowatts per rack, equivalent to the daily energy consumption of dozens of homes.

Simply increasing power capacity isn’t enough, as older data centers “can’t handle that sort of power density,” according to Draper. They lack the necessary electrical infrastructure and cooling systems required to manage the heat generated by AI chips. (Altman, who is heavily invested in nuclear fusion and serves as the executive chairman of Helion, aims to build the world’s first fusion power plant by 2028, potentially powering these data centers with limitless, clean energy, though it’s still a complex challenge. Helion is also backed by Softbank.)

Thus, Stargate may need to retrofit existing sites or find new ones, a process that’s neither quick nor cost-effective. Finding suitable locations to power large numbers of centers within one or two years is difficult, and costs can vary greatly. As Draper notes, “The numbers are pretty widespread depending on how you build it.”

For example, a standard Tier 3 facility could cost $10 to $15 million per megawatt to construct. Even on the conservative side, that would mean around $1 billion for basic infrastructure. With the addition of GPUs and other AI hardware, the price could rise to approximately $3.5 billion for a single 100-megawatt facility. As Draper puts it, “The numbers add up pretty quickly,” highlighting why Stargate’s budget goals are so large.

Stargate is already launching 10 data centers in Abilene, Texas, a location chosen for its potential renewable energy resources, according to Draper. President Trump has promised to expedite the construction process through executive orders as Stargate plans to expand beyond Texas. Unlike traditional data centers, which are concentrated in cities to reduce software lag time, AI facilities prioritize power access over location. However, as Draper notes, when the goal is to use compute for inference, facility placement becomes more complex.

If Stargate fails to secure the full $500 billion in funding, the project will likely scale back. This could result in fewer facilities, smaller builds, longer timelines, or a more regional focus instead of nationwide expansion.

This would fall far short of the original vision to democratize AI computing power by transforming the entire industry. Instead, the goal would be to address the most pressing gaps in the current infrastructure.

Foxconn 2.0

It’s tempting to draw comparisons between Stargate and the Foxconn project.

You might recall the much-publicized $10 billion LCD factory that was ultimately scaled down significantly, long after Trump had his moment of triumph. The promised jobs largely failed to materialize, and the project fizzled out after all the media attention.

It’s entirely conceivable that Stargate could follow a similar trajectory: bold announcements, media buzz, and then the harsh reality. The $500 billion goal could be revised downward, funding could dry up, or the project could quietly shrink into something much smaller. Some data centers will likely be built, but they’ll fall far short of the promised scale. Then again, isn’t that the essence of science fiction? A glimpse of the future that remains just beyond our grasp.

Even if Stargate secures the full $500 billion and builds out its vast AI infrastructure, a more pressing question looms: what if raw computing power isn’t the key to achieving AGI? As Naveen Rao, AI VP at Databricks, argues, “We need a fundamentally new learning paradigm.” More compute alone won’t get us there. His skepticism grew stronger when DeepSeek, a much smaller AI lab in China, reportedly achieved impressive performance using far less computing power by focusing on more efficient training methods.

This raises an unsettling possibility: Stargate could succeed in building its network of power-hungry data centers, only to find that it’s invested in the wrong approach. If better algorithms and more efficient training methods, rather than sheer computational power, are the true keys to next-gen AI, then we may not just be facing a bubble. We could be witnessing one of the biggest technological miscalculations in history — hundreds of billions spent on unnecessary infrastructure, leading to a financial disaster that could reshape tech investing for decades.

As Sequoia, which invested in OpenAI, highlighted last year, AI’s rapid advancement makes long-term data center investments risky. The cutting-edge clusters of today could be outdated before they’re even fully deployed. However, the AI infrastructure race isn’t about rational, long-term planning; it’s about survival. Sequoia partner David Cahn posed a thought experiment: “Imagine you knew for certain that AI would be as transformative as the internet, and you controlled the only AI company in the world. How fast would you build CapEx?” His answer: “You would take your time.”

In short, that’s the massive gamble AI leaders are making right now. Betting on scale isn’t completely foolish, though. The scaling hypothesis — the idea that bigger AI models, with more data and computing power, become smarter — has been the driving force behind the best models we have today. Much of this compute will also be used for inference, so that models can handle millions of user requests simultaneously without crashing the chatbots. When we spoke with the ChatGPT team, the infrastructure lead emphasized that “just keeping it up and running is a very, very big feat.” Stargate could theoretically play a key role in making that happen.

Silicon Valley operates on faith in visionaries

It also serves as a symbol of bragging rights. Mark Zuckerberg, Musk, and Altman appear to be in a data center competition, each trying to outdo the others through social media posts and press releases. Building the largest data center empire has become as fashionable among the tech elite as carrying the latest designer handbag. It signifies a leader’s ability to wield not only computing power but industrial influence — a flex that signals their capability to reshape both the physical and digital worlds.

In Silicon Valley, success often hinges on faith in visionaries, and Altman has expertly cultivated that persona. While Stargate’s investors present it as a solution to major global challenges, the reality is that it’s more about OpenAI securing the computing power needed to compete with rivals.

Paul Graham once wrote that if you dropped Altman on a cannibal-infested island, he’d return five years later as their king. Graham wasn’t referring to Altman’s charm or brilliance; he was emphasizing his adaptability and survival instincts. With Stargate, Altman has reinvented himself once again — now a quintessentially American figure: part industrialist, part tech visionary, part populist advocate. He’s not just promoting AI progress but also advocating for American renewal through large-scale infrastructure and job creation, tapping into Silicon Valley’s anxiety about falling behind and Middle America’s desire for industrial revival.

While a $500 billion moonshot failure wouldn’t be a career-ender — tech tends to forgive these, just ask Adam Neumann — it could hurt something bigger than Altman’s reputation. The AI sector has thrived on a potent blend of technical promises and industrial ambition. If Stargate falters, it won’t just be another startup collapse; it would serve as a wake-up call for an entire industry that has promised to reshape the world through raw computational power.

For now, Altman has cast himself into a classic science fiction tale: the visionary who promises to revolutionize society with technological might. But like many sci-fi plots, the most compelling question isn’t whether the technology works — it’s what happens when human ambition faces the harsh truths of physical reality. In six months or a year, we’ll know whether Stargate was the first chapter of America’s AI revolution or just another techno-optimist fantasy that couldn’t withstand the real world’s challenges.

Advertisement