The bigger story is not the headline.

On June 4, 2026, S&P Dow Jones Indices declined to bend its rules for SpaceX. SpaceX had asked for unusually swift entry into the S&P 500 on the strength of an “unprecedented market capitalization.” The answer was no: you are not yet profitable.

Read in isolation, this is just a SpaceX IPO story. The interesting part is who got caught in the same net. By refusing to carve out an exception, S&P also closed the fast-track lane for OpenAI and Anthropic. There will be no six-month seasoning period, no profit waiver, no IWF shortcut for unprofitable AI mega-caps. The “use your size to skip the line” playbook is now off the table.

Then, the same week, CNBC reported that Google is paying xAI $920 million a month for compute capacity inside xAI’s data centers. That is roughly $11 billion a year flowing from one AI giant to a direct competitor.

One door closes. Another one opens, with a price tag.

Public markets are telling AI companies: we will not reprice you through index inclusion. Private compute markets are telling them the opposite: we will pay you, in cash, for the infrastructure you control. That gap is the actual 2026 story for anyone building, funding, or competing with frontier AI companies.

What S&P was actually protecting

It is worth being precise about what got rejected, because S&P did not refuse SpaceX because of rockets or Elon Musk. The June 4 decision was a formal “no” to a set of rule changes the index committee had been openly considering for months.

The proposed changes were aimed at “MegaCap” companies with unprecedented market capitalizations and were specifically:

  • Shortening the seasoning period for newly public companies from 12 months to 6
  • Waiving the four-quarter profitability requirement for select mega-caps
  • Adjusting the Investable Weight Factor (IWF) to account for concentrated free-float structures

S&P said no to every one of them. The reasoning was about precedent, not SpaceX: if a company can use its size to skip the line, the next application of that precedent would almost certainly come from a frontier AI lab. S&P chose to write the rule once, for everyone, and let SpaceX, OpenAI, Anthropic, and whoever else wants in, walk through the normal door.

That matters because the S&P 500 is not just an index. It is a passive-investor machine. Once you are in, trillions of dollars in index funds, ETFs, 401(k) plans, and target-date funds buy you mechanically. Skipping the seasoning period, or entering profitlessly at scale, is not a technicality. It is one of the most powerful re-rating events in modern markets. S&P just made clear that path is closed to unprofitable AI.

Why OpenAI and Anthropic are the real collateral damage

It is tempting to treat this as a SpaceX-only story. The math says otherwise.

S&P’s rules are written for categories of companies, not specific names. The MegaCap consultation was the first time in years the index explicitly considered an exception for “unprecedented market cap + persistent losses.” Once that door opens, the queue behind it is well known and well funded.

OpenAI is widely reported to be preparing an IPO at a valuation north of half a trillion dollars. Anthropic is closing in on its own listing. Both are pre-profit by design; both are betting that growth, compute, and distribution will compound faster than their losses. The same structural facts apply to xAI, should it ever file publicly.

The S&P committee received that signal in its consultation feedback and chose to deny it. The downstream effect is not subtle:

  • Going public gets harder, because the market knows index inclusion is not on the table
  • The post-IPO re-rating gets slower, because passive flows are gated
  • The narrative arsenal shrinks — “the next Tesla” is no longer available as a price-discovery story

In other words: SpaceX was the name on the rejection letter. OpenAI and Anthropic are the companies that lose the most optionality from it.

The $920M/month counterweight

Now flip the lens. The same week, CNBC reported Google would pay xAI $920 million per month for compute capacity.

Pause on that number for a second. Roughly $11 billion a year. More than the annual R&D budget of most public software companies. More than the annual revenue of most AI startups. This is not a procurement contract. This is a structural cash flow line item in the budgets of two of the largest players in the industry.

And look at the shape of the relationship:

  • Google is one of the largest cloud and AI infrastructure operators on Earth
  • xAI is a direct competitor to Google DeepMind
  • Google is paying xAI, in cash, to run workloads inside xAI’s data centers

The only way that makes sense is if Google’s own capacity is not enough. The frontier-AI compute squeeze has gotten severe enough that a hyperscaler will pay premium dollars to a competitor to keep its own roadmap on schedule. That is the most honest signal we have had in years about how tight the supply side really is.

It also tells you exactly why SpaceX was asking for the S&P fast track in the first place. The xAI data center buildout is being financed, in part, on the back of an implied future valuation. The faster SpaceX gets into the S&P 500, the faster xAI gets a balance sheet to match its compute ambitions. S&P just said no to that financing arc — at least on the public side.

The structural shift nobody is naming

Put the two events next to each other and a real shift comes into focus. AI company valuations in 2026 are migrating away from the old SaaS-style “future free cash flow discounted to today” framework. The new anchor is compute: how much you own, how much you can sell, and how durable the customer relationships are.

The signals:

  • xAI is monetizing compute before it monetizes models
  • Google is paying a competitor to fill a compute gap it cannot close fast enough internally
  • SpaceX is treating its data center buildout as a core revenue line, not a cost center
  • S&P is signalling that this category of company is too risky, or too opaque, for index inclusion

The compounding implication: the “AI infrastructure company” identity is now financially distinguishable from the “AI model company” identity, and the market is starting to price them very differently. A company that can credibly sell compute at hyperscaler-grade reliability and at scale is starting to look like an infrastructure business with AI optionality. A company that can only sell models is starting to look like a software business with unusually high capital costs.

That distinction is going to drive a major part of the 2026–2027 re-rating across the sector.

Three practical takeaways

If you are building, investing in, or competing with frontier AI companies, three things are worth locking in.

1. Compute cash flow is the new valuation floor

The next time you see a private AI company raising at a stunning number, the first question is no longer “how good is the model” or “how fast is revenue growing.” It is: how much of the balance sheet is backed by compute contracts? Long-dated compute offtake agreements, capacity reservations, and infrastructure pre-payments are the closest thing the AI sector has to recurring revenue. Treat them that way.

2. Compute resale is about to become a real product line

xAI selling capacity to Google will not stay an isolated deal. The moment Anthropic, OpenAI, and the second-tier frontier labs finish their data center builds, “compute as a service to other labs” is going to be a first-class revenue line. This is not a side business. It is the new high-margin cash cow, and it will be one of the strongest defenses against the public-market chill S&P just signalled.

3. IPO timelines drift right, and “pre-IPO” gets longer

OpenAI and Anthropic were already going to face a difficult IPO tape. S&P’s June 4 ruling makes it harder. Expect the next 12–18 months to be defined by:

  • Continued reliance on primary markets and structured private debt
  • More creative financing — compute pre-payments, offtake-backed notes, hybrid instruments
  • A longer “pre-IPO” state where companies are economically public but legally private

The label “public” is going to mean less than it used to. The label “compute-backed” is going to mean more.

Closing thought

The question for 2026 AI companies is no longer “will we get to IPO?” It is something sharper and more uncomfortable:

Are we building a company that is meant to become public, or are we building a company that can fund itself indefinitely on private infrastructure revenue?

SpaceX tried the first path through the S&P 500 and got told no. xAI is already running the second path, and it is being paid $920M a month to do it.

The rest of the industry is going to have to pick a side — and stick with it.

References