► Capital Deployment at Historic Scale
A handful of overlapping investors, a dozen key individuals, and three cloud providers now control the trajectory of artificial intelligence — and June 2025 marked the moment the economics stopped being hidden.
The AI industry in 2025-2026 represents perhaps the largest coordinated capital deployment in technology history: over $200 billion poured into a sector where the top three companies control 88% of enterprise usage, a single chipmaker owns 86% of the GPU market, and virtually every major player shares investors, board members, or former colleagues.
This report maps the actual power structure, documents the June-July 2025 inflection point when subsidized pricing began cracking, traces the growing AI-crypto convergence, and evaluates evidence for deliberate market consolidation.
The AI industry's power structure is dominated by a surprisingly small number of entities.
Not Organic Prices: These valuations are not organic market prices — they are set by a rotating cast of the same institutional investors appearing in round after round across competing companies.
October 28, 2025: Converted to a Public Benefit Corporation
February 2026 Series G: Raised $30 billion alone, led by GIC and Coatue
March 2025: Merged with X (formerly Twitter) in all-stock deal valuing xAI at $80B, X at $33B
January 2026: Raised $20 billion at $230 billion valuation (Sequoia, a16z, Fidelity, BlackRock, NVIDIA ~$2B, Tesla ~$2B)
February 2026: SpaceX acquired xAI, creating a trillion-dollar entity spanning rockets, social media, and frontier AI
The Circular Economy: NVIDIA invests in virtually every major AI company (OpenAI, Anthropic, xAI, Cursor, Mistral, Cohere, Perplexity) while selling hardware to all of them — a circular dynamic where investment returns flow back as GPU purchases.
Pattern: A core group of perhaps 15-20 institutional investors rotates through every major AI funding round, creating a web of cross-holdings that makes competitive dynamics partly illusory.
FTC Report (January 2025, Unanimous): Cloud service providers gained "significant equity, revenue-sharing rights, consultation/control/exclusivity rights" through AI partnerships.
These aren't just investments — they're structural lock-in mechanisms: OpenAI committed $250 billion in Azure services back to Microsoft.
The AI power network runs through roughly a dozen individuals whose board seats, investments, and personal relationships create the connective tissue between ostensibly competing entities.
Notably: Several PayPal Mafia members (Musk, Thiel, Hoffman) now find themselves on opposing sides of the AI landscape.
June 16-18, 2025: The most concentrated pricing shock in AI tool history. Near-simultaneous limit tightening across Cursor, GitHub Copilot, and Replit.
Cursor Pricing Shock — Replaced 500-request-per-month Pro plan with $20 compute credit pool, effectively halving available usage to ~225 requests on Claude Sonnet 4. Overage charges hitting credit cards without warning. One Hacker News user reported $350 in overage charges in a single week. A team reported spending over $4,600 in six weeks.
GitHub Copilot Caps Enforced — Monthly premium request caps: 300 for Pro ($10/month), 1,500 for Pro+ ($39/month).
Replit "Effort-Based Pricing" — Replaced flat checkpoint costs with variable compute charges. Some users reporting bills of $350 in a single day.
Not Coordination, but Convergence: This wasn't coordination in the conspiracy sense — it was convergence driven by identical economic pressures. Agentic coding workflows had fundamentally broken flat-rate subscription economics. When AI tools shifted from simple code completion (cheap, predictable token usage) to autonomous multi-step agents that could run for hours, the gap between what users paid and what compute actually cost became unsustainable.
Cursor CEO Michael Truell (July 7): Issued formal apology, admitted rollout was botched, offered refunds. But the underlying message was clear: the era of subsidized unlimited AI usage was ending.
Timeline: Launched as research preview February 2025, generally available May 22. Operated on progressively tighter limits throughout June-July period:
The limits weren't arbitrary — they were rationing mechanisms reflecting the reality that inference at scale costs far more than subscription fees cover.
The funding environment in June-July 2025 was historically unprecedented.
Cursor Series C — $900 million at $9.9B valuation, led by Thrive Capital with a16z, Accel, DST Global. Remarkable for a company valued at $2.5B just months earlier.
Meta-Scale AI Deal — $14.3 billion investment for 49% non-voting stake. Prompted Google, OpenAI, and xAI to immediately cut ties with Scale over data confidentiality concerns.
xAI Funding — Closed $10 billion in debt and equity.
Anduril Series G — Raised $2.5 billion with $1 billion from Founders Fund alone.
DoD Contracts — Department of Defense awarded contracts worth up to $200 million each to Anthropic, Google, OpenAI, and xAI for frontier AI prototypes.
Simultaneous Launches:
The four companies that spend their days competing for developer mindshare were simultaneously signing up as defense contractors — all funded by the same investors, all using the same NVIDIA hardware.
Dismembered in 72 Hours: The Windsurf saga crystallized every power dynamic operating in the industry. A company with 1 million developers and 350+ enterprise clients was dismembered in 72 hours — not by market failure, but by Big Tech's acqui-hire gravitational pull.
May 2025: OpenAI agreed to acquire Windsurf for $3 billion (largest-ever acquisition), securing exclusivity window through July 11.
Deal Collapses: Microsoft's 2023 partnership agreement gave it IP rights over anything OpenAI acquired.
Friday Evening (Exclusivity Expires): Google executed a $2.4 billion reverse acqui-hire, hiring CEO Varun Mohan, co-founder Douglas Chen, and ~40 senior R&D staff into DeepMind's Gemini coding team.
Monday Morning: Cognition acquired Windsurf's remaining 210 employees, IP, product, brand, and $82 million ARR business.
NVIDIA Milestone — Market cap hit $4 trillion (first company to reach this milestone), driven by AI infrastructure demand and lifting of export restrictions on H20 chips to China.
EU AI Act — Enforcement date. EU Commission published guidelines for General Purpose AI provisions.
Media Shift: Goldman Sachs and Sequoia published reports questioning whether AI could generate the ~$600 billion in annual revenue needed to justify current spending. "AI bubble" narrative intensifying through Q3 and Q4.
The most direct AI-crypto connection runs through Sam Altman himself.
World (formerly Worldcoin): Co-founded by Altman in 2019, reached 25 million app users by February 2025. Launched in US on May 1, 2025 with Orb iris-scanning verification in seven cities.
The Thesis: AI makes proof-of-personhood essential — directly connecting OpenAI's existence to World's value proposition.
Forbes Report (January 2025): OpenAI building biometric-verified social platform. WLD token jumped 27% overnight.
Countries That Have Suspended/Banned/Investigated: Kenya, Indonesia, Philippines, Thailand, Colombia, Spain, Portugal, Germany
Token Concentration: Over 90% of WLD tokens held by top 100 wallets, with one wallet controlling 20.71% of supply. Token unlocks for early investors continue through July 2028.
Legal Separation: Entities remain legally separate — no direct revenue connection between OpenAI and World — but Altman's simultaneous chairmanship creates an implicit narrative bridge that benefits token holders.
Physical, Not Just Financial: By October 2025, bitcoin miners had announced $65 billion in AI/HPC conversion contracts. The largest documented intersection between AI and crypto is infrastructure.
CoreWeave Example: Originally a crypto mining operation. By mid-2025, generated over $1.2 billion revenue at ~$48 billion valuation running AI GPU data centers.
Crusoe Energy: Sold mining assets entirely to focus on AI.
NVIDIA's Dual-Use Hardware: Same GPU architecture (CUDA) underpins both industries. RTX 5090 (Blackwell architecture) dominates mining and AI benchmarks alike. NVIDIA paid a $5.5 million SEC fine in 2022 for failing to disclose crypto mining as major revenue source — dual-use nature has always been a feature, not a bug.
Other Dual-Focused Funds: Paradigm, Polychain Capital, Multicoin Capital, Hack VC (allocated 41% of latest fund to Web3 AI startups).
x402 Protocol: Open standard using HTTP 402 "Payment Required" code for stablecoin micro-payments. Processed over 50 million transactions.
AgentKit: Enables crypto wallets inside AI agents.
Agentic Wallets (February 2026): Purpose-built infrastructure for autonomous AI agents to hold and spend money.
x402 Foundation: Co-founded with Cloudflare (September 2025) to standardize how AI pays for services.
Payments MCP: Enables LLMs including Claude and Gemini to access blockchain wallets via natural language.
Players: Bittensor ($3.7-4.4B market cap), ASI Alliance (Fetch.ai, SingularityNET, Ocean Protocol), Render Network.
Reality: Remains niche relative to centralized labs. Grayscale filed for a Bittensor Trust (October 2025), signaling institutional interest, but Ocean Protocol withdrew from ASI Alliance same month.
No mainstream AI lab views decentralized AI as a serious near-term threat.
Tesla/SpaceX Holdings: Tesla holds 11,509 BTC ($1.3B), SpaceX holds ~8,285 BTC. X Money's crypto integration remains unconfirmed despite 40+ state money transmitter licenses.
The Bottom Line: The AI-crypto connection is real and growing — through shared infrastructure, shared investors, shared individuals, and emerging payment protocols — but the two industries' fusion remains early-stage and uncertain.
The numbers are unambiguous.
GitHub Copilot: Microsoft was losing more than $20 per user per month on Copilot at $10/month price. Average compute cost ran ~$30/user, with some heavy users costing $80.
Anthropic: Burn rate ran at roughly 70% of revenue through 2025, with internal documents projecting breakeven by 2028.
ChatGPT Subscription Reality: Only ~5% of 700+ million weekly active users pay for subscriptions. Even by 2030, OpenAI's most optimistic projections show only 8.5% paying.
The Math Is Clear: The vast majority of AI usage today is subsidized by venture capital and strategic investment, not by user payments.
GPT-4o Free Tier — Available to all users.
GPT-4o Restricted — Restricted to $20/month subscribers. Free API rate limits dropped from 20 requests per minute to 3.
Microsoft 365 Price Increase — First increase in four years. Frontline worker subscriptions jumped 33%, explicitly framed by industry analysts as "the moment that AI's free lunch finally gets a bill."
The subsidize-then-squeeze playbook from ride-sharing is being replicated at a scale orders of magnitude larger.
FTC Unanimous Report (January 2025, 5-0): Identified competition risks in AI partnerships. Cloud service providers gained equity, revenue-sharing rights, and control mechanisms through AI investments. Former Chair Lina Khan called it evidence of "lock-in."
Trump-Appointed Successor Andrew Ferguson (March 2025): "Big tech is one of the main priorities of the Trump-Vance FTC" — launched broadest antitrust investigation of Microsoft since the 1990s, demanding nearly a decade of data on AI operations, costs, and licensing.
The Acqui-Hire Loophole: Licensing technology plus hiring key employees rather than acquiring companies has proven remarkably effective at circumventing merger review thresholds. Over 2024-2025, Big Tech deployed $40+ billion through these structures.
Covariant CEO (said the quiet part aloud): A full acquisition would have been "killed by antitrust authorities."
Explicitly Drawn Comparison: The comparison to Uber's trajectory is explicitly drawn across multiple serious analyses and is the most instructive historical parallel.
Uber's Evolution:
Oxford University Research: After Uber introduced dynamic pricing algorithms, fares rose for passengers and fell for drivers while Uber's revenue share grew.
The AI Version: Replaces rider subsidies with API and subscription subsidies — and the scale is dramatically larger.
Explicitly documented circular dynamic:
Cloud Lock-In: Cloud providers offer $100,000+ in free credits to AI startups, creating lock-in.
FTC Finding: One cloud partner described information-sharing arrangement as "a multiyear crystal ball into the future needs of AI infrastructure."
Genuine competition remains possible:
But the structural forces — capital concentration, infrastructure dependencies, talent absorption, and investor overlap — are all moving toward oligopoly.
Not a Competitive Market: The AI industry's power structure is not a competitive market in any traditional sense. It is a capital deployment operation where roughly 15-20 institutional investors fund both sides of every competitive dynamic, three cloud providers control the infrastructure layer, one chipmaker commands 86% of the hardware market, and a network of perhaps 50 individuals make the decisions that determine which companies survive.
June 2025 was the inflection point where the economic reality became visible to end users. The near-simultaneous limit tightening across Cursor, GitHub Copilot, and Replit on June 16-18 wasn't coordinated — it was inevitable. Agentic AI workflows broke flat-rate economics, and the companies couldn't hide the gap between subscription revenue and compute costs any longer.
The question is no longer whether AI pricing will increase, but how quickly and how completely the subsidization era will end.
The AI-crypto convergence is real but early. The physical infrastructure overlap ($65 billion in mining-to-AI conversions), the shared investor base (a16z, Founders Fund), and the emerging payment infrastructure (Coinbase's x402, Agentic Wallets) represent genuine structural connections — not speculative hype.
Sam Altman's simultaneous leadership of OpenAI and World embodies the thesis that AI and identity verification are co-dependent. But no major AI lab has issued tokens, and the "decentralized AI" sector remains marginal.
The consolidation evidence is strong enough to call it a pattern:
This is the Uber playbook executed at 100x scale. The FTC's bipartisan 5-0 vote on AI competition risks and the Trump-era expansion of the Microsoft investigation suggest regulators see it too.
The Three Most Important Numbers in AI Today:
Everything else is narrative.
How usage limits quietly tighten after adoption across the AI industry.
READ REPORTInvestigation reveals 3-10x pricing markup and API dependency on Anthropic.
READ REPORTHidden prompt injections, behavioral controls, and the June-Sept 2025 transformation.
READ REPORT