The trial that could reshape the artificial intelligence industry is now weeks away from opening arguments, and the numbers being thrown around in court filings are staggering.
A pre-trial conference in Oakland, California, this week brought lawyers for Elon Musk and OpenAI together to settle key procedural questions: which of Musk's claims survive to trial, which issues the judge will decide, and what evidence and expert witnesses each side can rely on. The main event begins at the end of April, with Judge Yvonne Gonzalez Rogers presiding.
Rogers is not a judge that either side should expect to push around. She handled Epic Games v. Apple and has a reputation for running tight courtrooms and cutting through legal theatre. Both Musk and Sam Altman are expected to take the stand as witnesses.
The $109 billion figure hanging over OpenAI
The most explosive number in the case comes from an economist hired by Musk's team, a Berkeley Research Group expert with roughly 100 cases on his resume. His formula puts potential damages to OpenAI at $109 billion. The figure is not accidental: it sits uncomfortably close to the size of the funding round OpenAI is currently trying to close.
The basis for the calculation centres on ownership. OpenAI's non-profit parent is estimated to hold between a quarter and a third of the commercial business, and Musk's side argues that 50 to 75% of that stake is attributable to his early contributions to the company. Dr C. Paul Wan is leading the appraisal of how much of OpenAI's value Musk can claim credit for.
OpenAI contests the economist's methodology. That dispute will be one of several the judge must resolve before the jury hears a word.
Settlement is unlikely, but Microsoft might be different
The odds of a deal before April are low. People close to the case describe it as personal and ideological rather than a straightforward commercial dispute, and that dynamic makes compromise harder to reach when both principals are deeply invested in winning.
Microsoft, which has been pulled into the litigation despite Musk saying he has no desire to fight the company in court, may be a different story. The corporate dynamics at Microsoft are less charged, and some settlement discussions involving the company remain possible.
xAI is rebuilding, and Musk says the first version was wrong
While the lawsuit dominates headlines, Musk's AI company is dealing with a significant internal reset. xAI has hired Andrew Milik and Jason Ginsburg, both senior figures from Cursor, a startup that sells AI coding assistants to developers. Milik and Ginsburg were heads of product and engineering at Cursor, and their arrival at xAI stands out because they are joining a company that has recently watched co-founders leave.
Two co-founders departed this week alone, just one month after being elevated in a major restructuring. Several others had already left in earlier rounds of turnover. The attrition is notable even by the standards of an industry where movement between companies is constant.
Musk has acknowledged the problem directly, saying xAI was not built right the first time and is being rebuilt from the foundations up. His stated goal is to catch up to or exceed competitors by the middle of this year. That is an ambitious timeline, given that the company's current coding product is widely regarded as behind the field.
xAI's enterprise business is small relative to Anthropic's, and the company carries a $250 billion valuation following its merger with SpaceX last month, with a combined IPO planned. Burning through $1 billion per month last year, the financial questions around xAI will become more pressing for prospective public market investors.
Meta may be about to license Google's AI model
The most surprising development of the week may be the least expected: Meta is reportedly in talks with Google about licensing the Gemini model for its AI features, at least in the short term.
According to reporting in the New York Times, the two companies have discussed this possibility several times. Meta's own flagship AI model has hit setbacks in development, and the company is weighing its options for keeping its products competitive while internal work continues.
Five companies are currently considered to be building frontier AI models: Google, Anthropic, OpenAI, Meta and xAI. That both Meta and xAI are experiencing difficulties at the same time raises a real question about whether the talent pool is large enough to support five parallel efforts at the leading edge.
Meta is not without strengths in the AI race. Its advertising business continues to grow at roughly 25% annually, driven in part by AI improvements to targeting and delivery. The company's ability to build products that hold user attention, whatever one thinks of how that attention is used, is a genuine competitive advantage, and one that would benefit from a capable underlying model even if that model comes from a rival.
OpenAI quietly walked away from shopping checkout
OpenAI had, at some point, ambitions to integrate a checkout function directly into ChatGPT, giving users the ability to complete purchases without leaving the interface. That plan has been abandoned.
The decision is a telling one. Building consumer products is genuinely difficult, and the gap between a capability that seems obvious in principle and one that works well enough to matter in practice tends to be wide. The exact reasons OpenAI pulled back have not been explained, but the move fits a broader pattern: the hardest part of consumer AI is not the model, it is the product.
The consumer AI wave is still forming
The rough consensus among investors and founders watching this space is that the first companies to be funded in any technology wave tend to be infrastructure and business-to-business companies, and that consumer companies follow once the rails are in place. That wave may now be arriving.
What will separate successful consumer AI companies is not, at this point, which model sits underneath the product. The models are becoming increasingly interchangeable. The differentiator will be taste: the judgment to know what to build, what to leave out, and how to create something simple enough that it becomes genuinely useful to a mass audience.
One example cited in this context is Extra, an email product being built by a team that includes early Pinterest employees. The product applies an agentic layer to email management, and its beta version is already drawing attention. The founders are building with the assumption that models will improve, which means designing for capabilities that do not quite exist yet and staying ahead of what the technology can do.
The limitation of that approach is that models do still have to be good enough, particularly for applications outside the chatbot format. The question of whether they have cleared that bar for most consumer use cases remains open.
The IPO queue is long and the window keeps shrinking
A separate set of pressures is building in the public markets. Multiple major companies are preparing for IPOs this year, but the window for going public is narrower than it appears.
Several enterprise software companies are making the traditional preparatory moves: hiring bankers, meeting investors, and filing paperwork. But the market conditions are difficult. Recent IPOs have not performed well. Uncertainty in broader markets is high. And the expected arrival of very large offerings from SpaceX, OpenAI and Anthropic could pull investor attention and capital toward those names when they do arrive.
SpaceX could go public as early as this summer, potentially at record size. OpenAI and Anthropic are more likely to be end-of-year events. The combined weight of those three offerings could leave less room for smaller companies competing for the same investor dollars.
Among the more unusual IPO candidates is Huddle, a video technology company that sells to professional sports teams. Its niche positioning may actually insulate it from some of the AI disruption concerns weighing on other enterprise software businesses. Cerebras, an AI chip designer, is considered a more natural beneficiary of the current technology cycle and may find more receptive investors as a result.
Private equity-backed companies, including Anaplan and Genesis are also watching for the right moment. The macroeconomic backdrop, including the ongoing geopolitical conflict, is making that timing calculation harder, particularly for companies like SpaceX that need stable markets to support a listing of that scale.
The week ahead includes the GTC conference from Nvidia, where new chip announcements and remarks from Jensen Huang will be closely watched. For the companies building on top of that infrastructure, and the investors trying to work out which ones will still be standing in three years, the signal-to-noise ratio is not getting any better.