Will OpenAI and SpaceX find seats?

Musical chairs in AI — what's emerging through the fog

Anthropic confidentially filed for its IPO yesterday at a $965 billion valuation, giving it substantial competitive advantage in the AI category IPO race. OpenAI’s confidential filing is expected to follow this fall. The analytical environment processes them as rivals competing for AI category leadership at increasingly remarkable valuations. The competition operates within a different structural picture. The chairs have been filling, and not the way the visible competition suggests.

The visible competition

The AI category has produced sustained coverage at the application layer. Anthropic moved from roughly $9 billion in annualized revenue at the end of 2025 to $45 billion in May 2026, a $30 billion Series G at $380 billion post-money in February, and yesterday’s $965 billion confidential filing. The confidential filing carries structural significance beyond the valuation — first-to-file produces competitive advantage in the IPO race against OpenAI, which is expected to follow this fall at higher valuations.

The competition produces visible drama: which lab is leading, which model is better, which IPO will be larger. Underneath that, the established players have been operating their AI category positioning through extension of the markets they already control. The conventional analytical environment processes each move on its own without integrating across them.

The chairs that filled

The consumer-as-customer chair sits with Apple and Google.

The combination holds the consumer position through hardware-and-search. Google has been absorbing AI into the search market it has operated for two decades — AI Overview at the top of search results, Gemini integrated into Workspace, search itself repositioning as the AI-mediated information layer. Apple operates the consumer hardware platform that distributes that experience, with Apple Intelligence handling device-side integration and the longstanding $20 billion annual search arrangement keeping Google as the default that monetizes AI-mediated information. The combination handles AI as evolution — extension of an existing defensible market through coordinated existing infrastructure and channels.

Google announced an $80 billion equity raise this week, funded in part by a new-to-the-market deep pockets player. Berkshire Hathaway participated for $10 billion, alongside its largest portfolio holding in Apple. The Berkshire methodology — long-duration positioning in businesses that control category foundations — recognized the AI infrastructure investment as the kind of allocation it has always made in established positions. The institutional environment now reads the consumer-as-customer position as the value-methodology allocation, not as defensive maneuvering against the labs.

The enterprise-as-customer chair sits with Anthropic, on the model Palantir established.

Palantir built the analog over twenty years. Foundry and Gotham deployments became operationally load-bearing for enterprise and government customers. Customer expansion through integration depth rather than logo acquisition. Capabilities that consumer-focused competitors don’t operate at scale. Anthropic has been building the same chair in AI systematically. 300,000+ business customers driving 80% of revenue. 1,000+ customers spending over $1 million annually, doubled from 500+ in under two months. 100,000+ accounts running Claude on Amazon Bedrock as of April 2026. Claude Code reaching $2.5 billion in annualized revenue by February 2026. The Glasswing security partnership announcement May 28. The Constitutional AI and alignment research positioning that establishes Anthropic in the safety-and-security category.

Anthropic’s $965 billion valuation rests on the chair occupation. The Bedrock partnership depth. The enterprise expansion revenue model. The safety capabilities producing institutional and government customer relationships. The position holds because it operates in territory the consumer-as-customer holders don’t directly occupy.

Microsoft is building another chair — productivity-and-infrastructure. Windows, Microsoft 365, Copilot integration across the enterprise productivity stack. The N1X laptop SoC partnership announced through NVIDIA’s Taipei keynote yesterday extends Microsoft’s hardware reach into the AI PC category. The chair sits between consumer and enterprise, operating across productivity software that runs on consumer devices for enterprise customers. Microsoft is building, not occupying inherited position — the chair is being constructed. The staged backing-off from OpenAI exclusivity — October 2025 ending compute right of first refusal, April 2026 ending model exclusivity — reads as recognition that the consumer-AI chair wasn’t winnable through the OpenAI partnership, with Microsoft refocusing on the chair it can build from its own established channels.

Amazon holds infrastructure through AWS, with Bedrock as the AI-customer connection point. Anthropic runs through Bedrock as a major customer; the partnership connects Amazon’s existing AWS position to the enterprise-AI demand Anthropic generates. Amazon’s chair extends the AWS infrastructure position into AI through the partnership architecture.

NVIDIA holds chip-and-architecture underneath all of them, with the Vera launch yesterday positioning NVIDIA as the chip foundation for Anthropic, OpenAI, SpaceXAI, and the broader AI infrastructure category. NVIDIA also operates as an active investor in large niche AI players — companies building category-defining positions in specific domains like CoreWeave at cloud infrastructure, Wayve at autonomous vehicles, and Figure AI at humanoid robotics. These participants are pursuing the Palantir-template approach in their own categories. NVIDIA’s chair extends from chip supply through equity participation in the application-layer participants its chips support.

The chairs are filled. Consumer-as-customer with Apple and Google. Enterprise-as-customer with Anthropic. Productivity-and-infrastructure under construction by Microsoft. Infrastructure with Amazon. Chip-and-architecture with NVIDIA. The visible competition between Anthropic and OpenAI for AI category leadership operates within a structural picture where the consumer-as-customer chair is held, the enterprise-as-customer chair is held, and the other chairs are held by participants that haven’t been competing for AI-leadership recognition.

No chair for OpenAI?

OpenAI built itself for the consumer-as-customer chair. The ChatGPT product, the consumer pricing tiers, the consumer-developer API positioning, the brand-recognition strategy. The category positioning that produced the historical valuations OpenAI carries.

The chair has been held by Apple and Google. Not removed, not made unavailable — held, through coordinated existing positions that didn’t require new construction. OpenAI built itself for a chair that was always occupied. The visible competition between OpenAI and the established players framed the question as one of competitive replacement; the structural picture frames it as competition for a chair that wasn’t available in the first place.

OpenAI’s enterprise pivot is late. Anthropic established the enterprise-as-customer chair through years of focused development. OpenAI’s enterprise positioning — ChatGPT Enterprise, the API enterprise tier, the partnership-driven enterprise customer acquisition — competes for customers Anthropic has already integrated into. The chair was occupied before OpenAI began pursuing it at scale.

The Microsoft partnership has been the primary enterprise distribution channel — and Microsoft has been backing off in stages. October 2025 ended Microsoft’s right of first refusal as OpenAI’s compute provider. April 2026 ended Microsoft’s exclusive license on OpenAI models and capped the revenue share. OpenAI can now use any cloud and license to any company; Microsoft retains the equity stake and IP license through 2032 but the exclusivity that defined the partnership is gone. The partnership structure that legitimized OpenAI’s enterprise valuation has been weakening from the Microsoft side. The independent enterprise channel Anthropic built through direct Bedrock partnership, Cowork desktop product, and the API direct relationships sits at depth OpenAI hasn’t reached.

The condition produces the running-around behavior the analytical environment has been observing. Capital raises to fund the continued search. Partnership modifications to expand positioning. Product launches across categories. Communications register pivoting to enterprise framing. The behaviors are individually rational and collectively demonstrate what’s happening: a participant operating without a defensible chair at the implied valuation.

When the music stops — when the analytical environment processes the broader picture — OpenAI’s valuation requires a chair the picture doesn’t include. The condition isn’t a competitive ranking. The chair isn’t there.

Postscript: SpaceX arrives at the game

The largest IPO in history is approaching the AI category. SpaceX brings substantial currency to the game. Supply-constrained equity through the dual-class controlled-company structure. Forced-inclusion mechanics through Nasdaq’s accelerated index inclusion rules. Day 1 trading appreciation through the institutional demand absorbing the offering. The currency is real and substantial.

SpaceX-xAI merged in February 2026 at a combined $1.25 trillion valuation, bringing existing AI positions into the entity. Grok reached approximately 64 million monthly active users by early 2026 — third behind ChatGPT and Gemini, with US chatbot market share growing from 1.9% to 17.8% over twelve months. xAI generated approximately $500 million in annualized revenue at the end of 2025 with substantial operating losses. X provides the social distribution and user-data resource Grok trains on. The existing positions are real but operate below chair-occupying scale — behind Anthropic’s $45 billion annualized revenue at the enterprise-as-customer level, and against the Apple-Google held consumer-as-customer chair. Grok also operates against regulatory friction: UK Ofcom and EU investigations, national bans in Indonesia and Malaysia, content-safety incidents that constrain enterprise and government customer acquisition.

SpaceX needs a chair the current circle doesn’t include. Apple and Google hold consumer-as-customer. Anthropic holds enterprise-as-customer. Microsoft is building productivity-and-infrastructure. Amazon holds infrastructure. NVIDIA holds chip-and-architecture. The chairs available to a player with this currency don’t include a chair SpaceX can sit in despite the existing positions.

SpaceX has no publicly identified operator running the AI category positioning at the scale the established players and Anthropic operate. The IPO documents present an AI segment generating substantial projected revenue. The actual AI capability the segment requires hasn’t been built yet — SpaceX’s largest current AI customer is Anthropic, leasing SpaceX’s compute capacity. Yesterday’s principal clarification: “the short deal length was our request, not theirs, as I thought we might need the compute back at some point.” The compute runs to the chair-occupant because SpaceX hasn’t yet built capability that consumes it internally.

The options are build or buy or both.

Build requires capability development on a timeline the IPO valuation doesn’t accommodate. SpaceX has potential position through orbital deployment — orbital data centers operating beyond Earth-bound infrastructure constraints — that would sit at a different layer from the existing established players. The development requires years. The IPO sells participation in the AI TAM now.

Buy requires acquisition of chair-occupant participants. Anthropic won’t come available — the enterprise-as-customer chair depends on institutional independence that absorption would dissolve, and the founders’ positioning around AI safety as institutional commitment runs counter to absorption into a controlled-company structure. OpenAI can be pursued — SpaceX attempted acquisition at $97.4 billion in February 2025 — but OpenAI doesn’t bring a chair. Acquiring OpenAI solves the operator question without solving the chair question.

Cursor matters to either path. SpaceX’s 30-day option to acquire Cursor operates as the first AI category move with operational substance. Cursor sits at the application layer rather than the chair layer, but the acquisition brings AI capability, customer relationships, and team that begins category presence. The acquisition matters whether SpaceX pursues build or buy as the broader strategy.

Two players are now running. OpenAI looking for a chair that doesn’t exist at the implied valuation. SpaceX approaching the game with currency and constraints, looking for both a chair and an operator at substantially larger scale. Both are new to the structural picture. Both need a chair to support what they’re selling.

The conventional path is to keep running. SpaceX brings the currency to attempt both build and buy. The constraints — time, availability of targets, capability development requirements — operate against the timeline the IPO valuation has built.

The music is still playing. The chairs are filled. Two players are running. Whether new chairs will be set up is the question the structural picture leaves open.

---

This piece extends from the body of work, particularly NVIDIA’s Three Pillars, One Operator and the broader framework reading catalog.