OpenAI expects $350 billion in revenue in 2030.
It also forecasts $856 billion in compute and infrastructure spending through 2030.
The revenue ambition is enormous. The spending is even bigger.
Huge Growth Still Needs Outside Money
OpenAI projects negative free cash flow of $278 billion from 2026 through 2030.
Anthropic has its own huge target: $100 billion in annual recurring revenue by the end of 2026.
Both companies expect tremendous growth. But growth alone does not pay for the infrastructure needed to support it.
Shay Boloor argued that OpenAI’s plan depends on external financing staying cheap and available for five more years. In his view, even a small rate move could make the math “ugly fast.”
That risk is already part of the market conversation.
The Kobeissi Letter said markets were pricing a 53% chance of another Fed rate increase in October. Last year, markets expected three cuts by October 2026. Another increase would leave rates 125 basis points above those earlier expectations.
Higher rates would make capital-intensive AI plans harder to finance. That matters when a company expects years of heavy spending before producing positive free cash flow.
Meta Is Cutting Costs and Looking for Revenue
Meta is working on both sides of the problem.
Kalshi Finance posted that Meta could reportedly save $8.5 billion by using its own AI chips.
The company is also looking for new ways to make money from Muse, its personal agent. Muse can book appointments, fill out electronic forms, monitor home security feeds, and do more advanced work behind the scenes.
Muse has a free tier and monthly plans costing $20 or $100. Meta is also exploring taking a cut from shopping transactions completed through agents, although it has not settled on a plan.
The logic is simple: lower the chip bill, then find revenue beyond advertising.
But Muse also has to earn trust.
Meta AI chief Alexandr Wang said Muse runs in “its own isolated environment.” He said it never sees actual passwords or payment details, and it asks before taking sensitive actions.
Users must opt out if they do not want Meta using their Muse interactions to train AI models. Meta says it will first remove “critical personally identifying information.”
Meta recently agreed to pay nearly $17 billion to settle a case brought by a coalition of state attorneys general. They had sued the company for misrepresenting the prevalence of harm on Facebook and Instagram.
Meta says Muse has gone through internal testing, red teaming, and a bug-bounty program.
Muse has to become a real business while persuading users to trust it with personal tasks.
The AI Buildout Keeps Spreading
Jensen Huang said $400 billion of venture funding entered AI-native companies during the previous six months, with 80% using open models.
He argued that the world needs both open and closed models. Closed models push the frontier. Open models let companies, researchers, teachers, students, and startups build their own systems.
Huang said Nvidia now runs every model in the world. He described an industry covering models, chips, applications, infrastructure, data centers, construction, electricity, and power generation.
He also said China would reach advanced lithography systems by 2030.
Asked whether AI had reached the AGI moment, Huang said, “I think we’re already there.” He described narrow-domain superintelligence as already present in areas including self-driving and protein synthesis.
On safety, Huang said frontier labs should face extraordinary standards. He supported independent evaluators or auditors.
But he rejected the idea that recursive self-improvement would automatically spiral out of control. Products still need testing, evaluation, and regression checks before release.
The possible uses are expanding too. Watcher.Guru posted that Anthropic is building a biology lab to develop AI-driven treatments for rare diseases.
Big Targets Meet Expensive Stocks
One post said AVGO was trading at its lowest valuation since December 2023.
The same post listed Broadcom AI revenue guidance of $58 billion for FY26, $115 billion for FY27, and $230 billion for FY28. It warned that the later figures contain “a lot of assumptions.”
The stock trades are one thing. The business question is what must go right for Broadcom to reach those later revenue figures.
Market flows can also move stocks without saying much about the underlying business.
Polymarket Money said SpaceX could receive up to $20 billion in passive buying as its Nasdaq 100 weighting more than doubles. That buying does not necessarily mean investors decided the business had improved. Index funds may simply have to follow the new weighting.
A blog speaker said passive investing now represents more than 50% of the market. He put the current dividend yield at 1%, versus around 2% over the last 40 years.
His concern was valuation. He said the market starts from the top 1% of historical valuation and argued that falling valuation multiples could heavily shape future returns.
He preferred good businesses at fair prices and cash when nothing qualified. A separate post applied a “Buffett test” to GOOGL, META, MSFT, NVDA, V, MA, ASML, TSM, COST, and AAPL. It called them strong businesses, but not cheap ones.
AI revenue expectations are huge. New products and lower infrastructure costs still have to turn that growth into cash. In Shay Boloor’s view, OpenAI’s buildout still depends on cheap, available external financing.