AI Spending Crunch Hits OpenAI & Anthropic

EVENT OVERVIEW

Enterprises are slashing AI costs by switching from frontier models to cheaper alternatives like DeepSeek, forcing OpenAI and Anthropic to confront slowing growth just as they race toward IPOs amid rising competition from Microsoft, Amazon, and Google.

The AI hype cycle is encountering its first major reality check. Enterprise customers, burned by ballooning token spend—sometimes exceeding entire annual budgets in months—are pivoting aggressively to cost discipline. Startups like Lindy have migrated 100% of traffic to lower-cost open-weight models (e.g., DeepSeek), delivering dramatic savings while maintaining functionality for many use cases. This mirrors broader trends: Uber implementing spending tiers, finance teams demanding ROI visibility, and model routing becoming essential to match tasks with appropriate (cheaper) models.

OpenAI (run-rate ~$25B) and Anthropic (~$47B annualized) have ridden explosive growth from the “spend-at-all-costs” phase, but pricing power is eroding. Leading models have been slower to cut prices recently, opening the door for hyperscalers. Microsoft, Amazon, and Google—investors in both companies—are now aggressively promoting efficient, lower-cost offerings and full-stack control. Microsoft emphasizes routing via GitHub Copilot and warns against value concentration; Amazon highlights custom silicon advantages; Google touts Gemini 3.5 Flash at 1/2–1/3 the price.

This “spend crunch” arrives at a precarious time. Both AI leaders filed confidentially for IPOs in early June, seeking to lock in peak valuations near $1T before growth moderates. Analyst Gil Luria notes current growth rates are mathematically unsustainable, creating urgency to list while numbers dazzle. However, enterprise controls (usage limits, analytics) rolled out by both firms signal internal acknowledgment of the shift. Mid-sized companies remain early in adoption, but the frontier model premium for routine tasks is under acute pressure.

The transition exposes a classic infrastructure vs. application dynamic: hyperscalers own the stack (compute, distribution, data), positioning them to commoditize model access. OpenAI and Anthropic risk margin compression and slower revenue ramps as customers optimize. Consensus views this as temporary growing pains in an unstoppable AI boom (“toothpaste out of the tube”). The higher-signal view is that unit economics are deteriorating faster than expected, with power shifting upstream to cloud providers and open-source ecosystems. This is not the end of AI investment but the end of the undifferentiated token gold rush.

▸ Immediate revenue growth deceleration for OpenAI and Anthropic as enterprises cap token spend and route routine tasks to cheaper models.
▸ Accelerated adoption of open-weight and mid-tier models (DeepSeek, Gemini Flash) driving rapid cost deflation in AI inference.
▸ IPO timing pressure intensifies; listing now captures peak multiples before growth normalizes and scrutiny rises on path to profitability.
▸ Hyperscalers gain share in enterprise AI budgets through lower-cost, integrated offerings.
▸ Expanded rollout of usage analytics and spending controls across AI platforms, improving customer retention but highlighting prior excesses.

→ Developer behavior shifts from "tokenmaxxing" leaderboards to efficiency metrics, slowing demand for highest-end models in coding and internal tools.
→ Startup funding in AI applications tightens for those without clear unit economics, while efficiency-focused infrastructure plays attract capital.
→ Increased bargaining power for large enterprises and hyperscalers, pressuring OpenAI/Anthropic pricing and margins.
→ Geopolitical angle: Chinese open-weight models (DeepSeek) gain Western traction, subtly eroding U.S. closed-model dominance.
→ Finance/Procurement teams formalize AI as a distinct, auditable spend category, leading to more disciplined procurement cycles industry-wide.

◆ Power consolidates with infrastructure owners (Microsoft, Google, Amazon) who control distribution and hardware, turning frontier model providers into increasingly commoditized component suppliers—consensus underprices this vertical integration risk.
◆ Emergence of a sophisticated multi-model routing layer as a new high-margin bottleneck; companies mastering orchestration (not just raw capability) capture outsized value.
◆ Narrative shift from "bigger models win" to "efficient intelligence wins," creating asymmetric opportunities in specialized/smaller models, synthetic data, and inference optimization—areas where market pricing lags the economics.
◆ Potential regulatory/political backlash accelerates if value accrues too heavily to a few U.S. labs, as Nadella warned; this favors distributed/open ecosystems long-term.
◆ Underpriced opportunity: Short-term hype fade creates entry points for patient capital into post-IPO OpenAI/Anthropic if they successfully pivot to enterprise efficiency platforms, but higher conviction lies in betting on the hyperscalers' full-stack leverage and the open-source tailwinds that democratize access while capping pure model pricing power.

AI Spending Crunch Hits OpenAI & Anthropic

First-Order Effects

Second-Order Effects

Alpha Layer — Opportunities

// Share Your Analysis