The S&P 500 rejected SpaceX and signaled it won't bend rules for OpenAI or Anthropic either. Here's what that profitability wall means for AI companies building at scale.

Why AI Giants Like OpenAI Can't Get Into the S&P 500

The most valuable AI companies in the world — OpenAI, Anthropic, SpaceX — are locked out of the most tracked stock index on the planet. Not because they lack market cap. Not because they lack brand recognition. Because they do not turn a consistent profit, and the S&P 500 index committee is not in the mood to negotiate.

That is a more consequential signal than it first appears.

The Rule That Walls Out Billion-Dollar Brands

The S&P 500 has a profitability requirement baked into its inclusion criteria: a company must report positive as-reported earnings for the most recent quarter and positive cumulative earnings over the four most recent quarters combined. It is a straightforward filter that has held for decades.

SpaceX reportedly sought an expedited path to inclusion. The committee declined to waive the rule. The implicit message to OpenAI and Anthropic — both of which are burning hundreds of millions annually on compute, safety research, and talent — is the same: earn your way in, or stay out.

This is not the committee being hostile to tech. Amazon, Google, and Meta all eventually cleared the bar. The S&P 500 has absorbed plenty of high-burn companies. The requirement is simply that the burn eventually stops.

What It Reveals About the AI Business Model

The rejection draws a useful line between two very different types of AI companies:

Infrastructure and platform AI (GPU cloud providers, API-first model companies) — where unit economics are brutal early but defensible at scale once adoption locks in.
Application-layer AI (vertical SaaS, AI copilots, workflow automation tools) — where margins can become healthy relatively quickly because the underlying model cost is someone else's problem.

OpenAI and Anthropic sit uncomfortably in the first category. Training frontier models requires capital expenditure that rivals national infrastructure projects. Inference at scale is expensive. Despite charging for ChatGPT Plus, enterprise API access, and operator agreements, neither company has publicly demonstrated that revenue comfortably exceeds spend.

SpaceX is a different case — it is genuinely profitable in some segments — but its corporate structure, with multiple share classes and Elon Musk's control provisions, creates additional complications beyond the earnings rule alone.

The Profitability Conversation Nobody Wants to Have

There is a polite silence in much of the AI industry around unit economics. Conversations center on benchmark performance, parameter counts, and context windows. The harder question — what does it actually cost to serve one million queries, and does the revenue from those queries exceed that cost — gets less airtime.

For software founders and product teams building on top of these models, that silence carries real risk. Pricing from the major model providers can shift significantly as their own cost structures evolve. A feature that is economically viable today at current API prices may look very different in 18 months if provider pricing changes to reflect true compute costs.

The S&P 500 exclusion is, in a roundabout way, the capital markets making that concern visible and formal.

Index Exclusion Has Downstream Effects

Being outside the S&P 500 is not merely a prestige issue. Inclusion in the index triggers automatic buying from index funds and ETFs — a massive, price-insensitive capital inflow that boosts valuation and liquidity. Exclusion means that pool of capital remains unavailable.

For companies that are still private (OpenAI, Anthropic), this is somewhat moot today. But as both firms move toward public offerings or structured liquidity events, the profitability question will move from theoretical to fiercely practical. Institutional investors who benchmark against the S&P 500 will apply the same lens the index committee just applied.

The companies that get ahead of this — by demonstrating a credible path to positive unit economics, not just impressive revenue growth — will have a materially easier time in the public markets.

What This Means for Software Teams Building With AI

If you are a SaaS founder or engineering lead integrating AI into your product, the index committee's decision is a useful prompt for your own financial hygiene:

Model your AI costs as a cost of goods sold line item, not as an R&D or infrastructure abstraction. Treat inference costs the way you treat cloud hosting — with usage-based accountability.
Track gross margin per AI-assisted feature, not just overall product margin. Some features will be heavily subsidized by current model pricing. Know which ones.
Diversify model dependencies where possible. If one provider's pricing reflects true costs before another's does, you want optionality.
Build efficiency into the product loop early. Prompt compression, caching frequent outputs, choosing smaller models for lower-complexity tasks — these habits are easier to build before scale than after.

The frontier model companies are fighting a cost battle at enormous scale. Your job is to make sure their economics do not become your problem.

A Profitable AI Industry Is Still Being Built

None of this means AI is a bad bet. It means AI — like cloud infrastructure before it, like mobile platforms before that — is in a phase where the dominant players are investing heavily to establish positions they expect to monetize for decades. That is a reasonable strategy. It is also one that requires patient capital and a tolerance for index exclusion in the interim.

The S&P 500 committee did not rule on whether AI is transformative. It ruled on whether the current crop of AI leaders can generate sustainable earnings. The answer, for now, is not yet.

Why this matters for your project: If you are building or scaling a software product that relies on AI APIs as a core dependency, the financial instability of frontier model providers is a legitimate architectural and business risk. Designing for model portability, monitoring inference costs against revenue, and staying close to the unit economics of every AI-powered feature are not optional hygiene — they are the difference between a product that scales profitably and one that quietly bleeds margin as it grows.

Source: Ars Technica / Hacker News — https://arstechnica.com/tech-policy/2026/06/sp-500-blocks-fast-spacex-entry-wont-waive-rule-for-unprofitable-ai-firms/