The AI industry's economic model is splitting into three distinct lanes in a single earnings week: hyperscalers pouring capital into infrastructure, model providers slashing prices to hold market share, and device makers preparing to pass compute costs directly to consumers. ANALYSIS The simultaneous appearance of Amazon's raised capex forecast, OpenAI's aggressive price cuts, and Apple's signal that it will charge for heavy AI usage reveals an industry where the cost of building AI, the cost of selling AI, and the cost of using AI are moving in sharply different directions.
Why it matters
The AI build-out has entered a phase where the economics no longer move in lockstep. Amazon expects capital expenditures to reach $220 billion this year12. Microsoft invested $41 billion in the quarter ended June 30 while keeping its full-year capex outlook steady14,13. Yet at the model layer, OpenAI slashed the price of GPT-5.6 Luna by 80%, bringing API costs down to $0.20 per million input tokens, and cut Terra by 20% to $2 per million input tokens11,1. And at the device layer, Apple CEO Tim Cook said the company "will have some kind of upgrade possibilities on iCloud Plus where people can buy up the stack" — the first confirmation that heavy Siri AI users may need to pay8,3. ◆ Infrastructure spending is accelerating, model pricing is collapsing, and consumer-facing AI is being metered — three vectors that cannot all be sustainable at the same pace.
The big picture
The hyperscaler earnings this week painted a picture of demand outrunning supply. AWS sales expanded 37% year over year, its fastest growth since 2021, and CEO Andy Jassy raised Amazon's full-year capex forecast to $220 billion10. Microsoft's Azure cloud revenue grew 43% on a constant-currency basis, beating the FactSet consensus of 40.26%. "We are advancing the frontier on the cost-to-outcome curve, ensuring every customer can turn tokens into business results," said Satya Nadella4. Microsoft 365 Copilot reached over 30 million paid seats, and GitHub Copilot now has 50 million users9.
But the model providers selling into that cloud are racing downward on price. OpenAI is "facing pressure to cater to a more cost-sensitive customer base, where enterprises have been less inclined to deploy expensive models without a clear picture of the return on their investments". The company is also working to fend off competition from Chinese startups and tech giants. OpenAI CFO Sarah Frier told employees at an internal all-hands meeting that the company added more net new ARR in July than it did in all of the second quarter, according to a source who was in the room. ANALYSIS The juxtaposition is striking: OpenAI is cutting prices aggressively while simultaneously claiming record revenue growth, suggesting that volume is being traded for margin.
Meanwhile, Apple reported a 22% surge in handset sales but issued weak guidance for the current quarter, blaming "supply constraints"5. Cook's comments about paid iCloud Plus tiers for AI power users came in that context. Apple is relying in part on Google Gemini models to help power the new Siri.
Between the lines
OpenAI's price cuts are not purely competitive — they are partly technical. GPT-5.6 Sol was "actively used to analyze production traffic, tune load balancing, and autonomously rewrite production kernels," reducing end-to-end serving costs by 20%2. Improved speculative decoding increased token-generation efficiency by over 15%. ANALYSIS The recursive self-optimization loop — where a model reduces the cost of serving itself — introduces a deflationary dynamic that pure capital spending cannot replicate, and it helps explain how OpenAI can cut prices while claiming revenue acceleration.
Goldman's analyst noted that Microsoft is pulling "varying levers on the monetization side" that are "perhaps not nearly as obvious before," pointing to GitHub's shift to consumption-based billing for power users. ◆ That shift mirrors Apple's move toward usage-based AI pricing — both companies are converging on a model where heavier AI consumption triggers higher charges, even as the underlying model costs fall.
Microsoft's 30 million Copilot paid seats sit against an estimated 450 million Microsoft 365 commercial customers. ◆ That ratio underscores the gap between AI availability and AI adoption that the entire industry is still navigating: infrastructure is being built for a demand curve that enterprise purchasing has not yet fully validated.
What's next
Nadella teased that a Copilot "super app" is coming later this quarter. Apple will broadly launch its Siri AI with iOS 27 this fall. Amazon's raised capex forecast to $220 billion signals that infrastructure buildout will intensify through the rest of the year. ◆ The central tension — capital expenditures rising at the infrastructure layer while prices collapse at the model layer and usage-based billing emerges at the consumer layer — will define whether the AI build-out produces returns commensurate with its cost, or whether the industry is building cathedrals and selling admission tickets at a discount.