$700 Billion in Capex. $50 Billion in Revenue. AI’s Math Is Broken.

Key Takeaways

• Anthropic raised $72 B, revenue $30 B, projected $14 B loss.
• OpenAI’s gross margin ~48%, costing $2 for every $1 inference revenue.
• Hyperscalers’ 2026 capex expected near $700 B, double 2025 spend.
• Open‑source models cut API costs 10‑13×, eroding frontier pricing.
• Adoption curves are hockey sticks; switching costs remain minimal.

Pulse Analysis

The current wave of AI investment is driven more by hype than by sustainable financial fundamentals. Anthropic’s $380 billion post‑money valuation and OpenAI’s $852 billion market cap rest on revenue multiples of 27‑35 times, despite both companies posting multi‑billion‑dollar operating losses and delayed cash‑flow positivity. Such valuations assume that exponential user growth will eventually translate into profitable SaaS revenue, a premise that ignores the stark reality that each new model generation resets cost structures and erodes pricing power.

Underlying this optimism is a cost dynamic that outpaces revenue growth. Hyperscalers are committing $700 billion in AI‑related capital expenditures for 2026, a figure that dwarfs the combined revenue of all pure‑play frontier labs. Even with hardware efficiencies from TPUs, Trainium and custom silicon, the cost per inference remains high—OpenAI’s internal data shows $2 of cost for every $1 of inference revenue, yielding a gross margin of just 48%, half that of mature SaaS firms. Open‑source initiatives like DeepSeek and Llama are delivering comparable performance at 10‑13 times lower API costs, compressing the price ceiling for closed‑source APIs by 30‑50% annually.

The strategic implication mirrors the Uber saga: early platform optimism falters when unit economics remain negative and competition drives prices down. AI labs must either find a pricing model that captures lasting value—perhaps through enterprise‑grade integrations, data‑centric services, or regulated AI offerings—or accept a prolonged period of subsidized growth funded by venture capital. Until a clear path to profitability emerges, investors should treat lofty AI valuations with caution, focusing on cash‑burn metrics and the durability of any emerging moat.