Tomasz Tunguz (Theory Ventures) · 2026-03-02 · 94d

The AI Commoditization Moment: How Generic Models Are Disrupting Frontier AI Pricing

Chinese AI models have achieved pricing parity with US frontier models while maintaining comparable performance, creating a 90% price discount similar to generic pharmaceuticals. The compressed timeline from innovation to commoditization—measured in weeks rather than decades—poses a fundamental challenge to the business model of high-cost AI development. Three forces drive this commoditization: model distillation, hyperscaler subsidies, and DeepSeek's ultra-low training cost precedent.

15 metrics· Cited 0× in the knowledge base ·Open source ↗

Metrics in this report

Alibaba Cloud LLM Pricing Reduction

97%

maximum

Hyperscaler subsidy

Chinese AI API Price Collapse

90%

decline

2024 pricing compression

Chinese AI Lab 2025 Revenue

1.8$B

total

All Chinese labs combined

Chinese Model Average Input Token Price

0.48$/1M tokens

average

Zhipu GLM-5, Minimax M2.5, DeepSeek V3

Chinese New Year 2026 AI Subsidy Spending

1.1$B

total

Baidu, ByteDance, Tencent combined

DeepSeek V3 Annual Recurring Revenue

220$M

ARR

With 122 employees

DeepSeek V3 Input Token Price

0.14$/1M tokens

lowest

DeepSeek official API pricing

DeepSeek V3 Training Cost

6$M

total

Model development

DeepSeek V3 vs GPT-5.2 Price Differential

90%

discount

Equivalent capability comparison

GPT-4 Training Cost

100$M

minimum estimate

Model development

GPT-5.2 Input Token Price

1.75$/1M tokens

median US

OpenAI pricing

OpenAI 2025 Revenue

22$B

total

US frontier models

OpenAI Daily Token Volume

8.6trillion tokens

daily

API usage

US Frontier Model Average Input Token Price

3.38$/1M tokens

average

Anthropic Claude Opus, OpenAI GPT-5.2

US to Chinese AI Lab Revenue Ratio

12:1ratio

2025

$22B vs $1.8B