For months, the technology world has been waiting for the SpaceX IPO, an event poised to be one of the largest public offerings in history. We expected revelations about rocket launch economics, Starlink’s global reach, and Elon Musk’s interplanetary ambitions. What we got was something far more immediate and, for those of us tracking the artificial intelligence arms race, far more revealing. Buried within the hundreds of pages of the S-1 filing is the first real, audited look into the financial furnace of a frontier AI lab: xAI. And the numbers are staggering. The filing peels back the curtain on a company burning through cash at an astonishing rate, making enormous bets on private power generation, and striking unprecedented deals with its direct competitors just to fuel its colossal ambitions.
This isn’t just a financial disclosure. It is the most transparent document we have ever seen detailing the raw, brutal economics of building a next-generation large language model. It tells a story not of elegant algorithms, but of billion-dollar losses, controversial energy sources, and a high-stakes, symbiotic relationship between bitter rivals. The SpaceX IPO filing has inadvertently become the Rosetta Stone for understanding the true cost of competing with OpenAI and Google.
The Shocking Burn Rate of a Trillion-Parameter Dream
Let’s start with the headline figure. In 2025, xAI posted an operating loss of $6.4 billion on just $3.2 billion in revenue. To put that in perspective, the company spent nearly three dollars for every dollar it brought in. While losses are the norm for ambitious tech ventures, this level of cash incineration is in a league of its own, dwarfing the reported burn rates of its peers. The filing reveals that losses ballooned from $1.56 billion in 2024, showing an acceleration of spending that is both breathtaking and, in the context of its goals, entirely logical.
The reason for this financial vortex is clear: Musk’s stated goal to scale Grok, xAI’s flagship model, to “multiple trillions of parameters.” This isn’t a simple software update. It is a declaration of war against the laws of physics and economics. A multi-trillion parameter model requires a city-sized cluster of GPUs, a data center infrastructure of unprecedented scale, and, most critically, an almost unimaginable amount of electricity. The $6.4 billion loss isn’t just abstract accounting. It’s the tangible cost of thousands of Nvidia GPUs, the concrete and steel of data centers, and the megawatts of power needed to bring them to life.
While OpenAI and Anthropic remain private, their financials are shrouded in investor updates and carefully curated press releases. We hear of multi-billion dollar funding rounds and massive compute deals with Microsoft and Google, but we don’t see the line items. The SpaceX filing gives us the line items. It shows that the race to artificial general intelligence is, for now, a race to the bottom of the balance sheet.
A Surprising Lifeline: Selling Compute to the Competition
How does a company losing over six billion dollars a year survive? The S-1 filing reveals an answer that sent shockwaves through the industry. In a stunning strategic pivot, xAI has become a major compute provider to one of its biggest rivals, Anthropic.
According to the filing, Anthropic has signed a deal to pay xAI a staggering $1.25 billion per month for 300 megawatts of compute capacity from xAI’s new Colossus 1 data center near Memphis, Tennessee. This is not a typo. The deal, which runs through May 2029, could be worth over $40 billion to xAI, providing a critical and massive revenue stream to offset its development costs.
This arrangement fundamentally changes the competitive landscape. The AI industry has historically been divided into three camps: the model builders (OpenAI, Anthropic, Cohere), the hyperscalers who provide the compute (Amazon Web Services, Microsoft Azure, Google Cloud), and the chipmakers (Nvidia). With this deal, xAI becomes a hybrid entity, a rarity in this high-stakes game. It is simultaneously building Grok to compete with Anthropic’s Claude, while also selling Anthropic the very power and infrastructure it needs to do so.
This emerging model, where rivals become each other’s landlords, highlights the extreme scarcity of one resource: AI-ready data center capacity. There simply are not enough power-dense, GPU-filled facilities to meet the insatiable demand of model developers.
For xAI, it’s a brilliant, if potentially fraught, way to monetize a gargantuan capital investment before its own models can fully utilize it. For Anthropic, it secures a massive block of much-needed compute outside the orbit of the big three cloud providers. It is a marriage of convenience born from a shared crisis of scarcity. The filing notes that xAI expects to “enter into additional similar services contracts,” signaling that this is not a one-off deal but a core part of its business strategy moving forward. Elon Musk isn’t just building an AI, he’s building a power broker.
The Dirty Secret of AI’s Power Thirst
The deal with Anthropic only makes sense if xAI has a surplus of power to sell. The IPO filing reveals exactly how Musk plans to secure it: by doubling down on a controversial, brute-force solution. The company has committed to spending another $2.8 billion on natural gas turbines to power its data centers.
This massive investment comes even as xAI faces a lawsuit from the NAACP over its existing, unpermitted use of dozens of gas turbines in Memphis, a community already struggling with poor air quality. The turbines are essentially mobile, jet-engine-powered generators, a way to get massive amounts of electricity online quickly without waiting for grid upgrades. They are a shortcut around the single biggest bottleneck in the AI boom: the creaking electrical infrastructure of the United States.
This is the dirty, physical reality behind the ethereal concept of cloud computing. Every query to Grok or Claude, every image generated, consumes a real amount of energy. As models grow to trillions of parameters, that energy consumption becomes astronomical. By investing in its own gas-fired power generation, xAI is ensuring its path to scale is not dictated by utility commission timelines or grid capacity. It is also, critics argue, externalizing the environmental cost of its ambitions onto local communities.
The $2.8 billion commitment shows that for Musk, the speed of AI development is paramount, and securing a private power supply is a non-negotiable part of the strategy, regardless of the regulatory and reputational blowback.
Risk, Regulation, and “Spicy” Mode
The candor required in an IPO filing also forces a company to publicly acknowledge its risks. For xAI, one of those risks is its own product. The document explicitly warns investors that features like Grok’s “Spicy” and “Unhinged” modes, which operate with fewer safety filters, could expose the company to regulatory scrutiny, lawsuits, and reputational damage.
This isn’t just boilerplate legal language. The company has set aside a reserve of $530 million for potential litigation losses, with some of that fund specifically earmarked for complaints related to sexualized imagery generated by Grok. Musk’s mission for a “truth-seeking” and less-censored AI is now a quantified financial risk listed in a public document. It puts a price tag on the ongoing debate over AI safety and alignment. While competitors like Anthropic build their brands around constitutional AI and safety, xAI is codifying its edgier, libertarian approach as a multi-million-dollar liability.
The Price of Ambition
The SpaceX S-1 filing has done more to demystify the AI industry than a dozen research papers. It lays bare the astronomical costs, the unorthodox strategies, and the uncomfortable compromises required to compete at the very highest level. We now see that Elon Musk’s xAI is a company defined by a series of audacious, high-risk gambles.
It is gambling that it can sustain over $6 billion in annual losses. It is gambling that it can build a trillion-parameter model that justifies that spend. It is gambling that it can pioneer a new business model as both a competitor and a landlord to its rivals. And it is gambling that it can build its own private energy empire, sidestepping the grid and its environmental regulations.
For years, the AI arms race has been discussed in terms of model capabilities and benchmark scores. This document reframes the entire conversation. It is a war of capital, of energy, of logistics. The winners may not be those with the cleverest algorithm, but those with the deepest pockets and the most secure access to power. The road to AGI, it turns out, is paved with debt, powered by gas turbines, and funded, in a strange twist of fate, by the very rivals you seek to surpass.
