WALL STREET HAS always understood that winning a lucrative mandate means making concessions to the client. Banks sponsor galas, lend at razor-thin margins, suffer through awkward golf outings. But Elon Musk's latest demand of his IPO advisers ventures into altogether different territory: buy enterprise subscriptions to Grok, his AI chatbot, and integrate it into your IT systems — or risk losing a seat at the table for what may be the largest initial public offering in history.
SpaceX filed confidentially with the SEC this week and is now floating a valuation above $2 trillion to prospective investors, according to Bloomberg. With 21 banks assembled under the codename "Project Apex," the five lead bookrunners — Bank of America, Citigroup, Goldman Sachs, JPMorgan Chase, and Morgan Stanley — stand to split fees that could exceed $500 million. Several have already agreed to spend tens of millions on Grok enterprise licenses and have begun embedding the chatbot into their internal systems, according to the New York Times. The cost is a rounding error next to the fee pool. The real question is what Musk gets in return beyond the revenue.
The data of the deal
Yet the most consequential part of this arrangement may not be the subscription dollars — it is the integration itself. Musk has been explicit that data is xAI's primary competitive advantage. At the Baron Investment Conference last November, he called X's real-time conversational firehose "by far, the best source of real-time data in the world." The Musk data ecosystem is already formidable and unlike anything a rival AI lab can replicate: X contributes billions of daily social interactions for training language models; Tesla's global fleet generates continuous sensor data — video, depth estimation, driving behavior — across millions of miles of road each day; SpaceX feeds orbital operations data into the stack. Now add Wall Street.
When five of the world's largest banks integrate Grok into their IT environments, xAI gains something it cannot acquire on the open internet: a window into how the most sophisticated financial institutions actually use AI. What queries do traders run at 6 a.m.? What documents do compliance officers feed into the system? What workflows break, and where? Even if xAI never trains directly on the banks' proprietary data — and its enterprise FAQ promises it will not, by default — the usage patterns, query distributions, and failure modes constitute an invaluable signal for product development in the most lucrative vertical in enterprise AI.
The fine print, however, deserves a closer read. xAI's own enterprise FAQ notes that while business data is not used for model training by default, the company "may offer free credits in exchange for you agreeing to permit us to train on your business data." It is a clause buried deep in the legalese, but it reveals the architecture of the incentive: a built-in mechanism to flip the switch from data isolation to data ingestion, one willing customer at a time. Whether any of the banks would agree to such terms is another matter — regulated financial institutions tend to be exquisitely sensitive about data governance. But the option exists, and xAI has made it contractual.
The broader context makes the data play even more legible. Grok's enterprise product launched only in late December; its security certifications and compliance infrastructure still lag behind ChatGPT Enterprise and Claude for Work, both of which have spent years building the trust architectures that regulated industries demand. Grok's U.S. chatbot market share has surged from under 2% to roughly 18% in a year, according to Reuters — impressive consumer growth, albeit one fueled partly by controversy around its less-restrictive content policies and the sexualized-image scandals that have gotten the chatbot banned in several countries. Enterprise adoption remains nascent. By compelling the IPO banks to adopt Grok, Musk is short-circuiting the slow, trust-building sales cycle that enterprise AI normally requires — and doing it with the most prestigious reference customers imaginable.
The strategy has a second layer. If Goldman Sachs and JPMorgan are Grok customers, that becomes a reference case xAI can cite in every subsequent pitch to a hedge fund, asset manager, or insurance company. The distinction between choosing Grok on its merits and choosing it because the alternative was forfeiting a once-in-a-generation fee opportunity is one the sales deck need not belabor. Whether those institutions continue using the product after the IPO prices in June will determine if Musk has bootstrapped a genuine enterprise business or merely rented prestige for the roadshow.
Musk, in effect, is extending his data flywheel into its most valuable new domain. The AI race is increasingly a data race, and the companies with proprietary, real-world data streams — not just internet text — will have the structural advantage in building models that actually do useful work. OpenAI has Reddit; Google has Search and YouTube; Anthropic has its constitutional approach to safety. Musk has X, Tesla, SpaceX, and now — however reluctantly — Wall Street. The bankers may think they are buying a seat at the IPO of the century. They may also be handing Musk the keys to the AI model that eventually replaces them. ■
For more, join 75,000 subscribers getting tech's favorite brief.
