DoorDash, Siemens, and Airbnb Are Quietly Switching to Chinese AI Models

DoorDash, Siemens, and Airbnb are among the Western companies turning to Chinese AI models to slash ballooning AI costs, the Financial Times reports. The shift is happening quietly, through American cloud providers rather than direct contracts with Chinese firms — but the savings are loud enough that US lawmakers have launched investigations.

🔍 THE BOTTOM LINE

A coding session that costs $10 on Claude costs under 50 cents on DeepSeek. That 20x price gap is forcing a market restructure that no export control can fix. Chinese AI models are becoming infrastructure, not products — and infrastructure choices are made on cost, not patriotism.

The 20x Cost Gap Is Real

The numbers are stark. Stu Clott, an operations manager in San Diego, told Rest of World that he switched from Claude to DeepSeek for coding, personal counseling, and banking software. The same hour-long session that cost $10 on Claude ran him less than 50 cents on DeepSeek. “The output quality, to be honest, I can’t tell the difference,” he said.

Lindy, a San Francisco AI assistant startup, switched from Anthropic to DeepSeek and saved millions, according to founder Flo Crivello. “You don’t need God to write your email,” he said on tech show MTS. “If you can get those lower tiers of intelligence for a tenth of the price, it would be foolish not to do it.”

On OpenRouter, a platform that routes tasks to different AI models, DeepSeek, Tencent, Minimax, and Xiaomi are now the four most popular choices. Vercel reported DeepSeek’s share of token usage jumped from under 1% to 17% in May — though its share of revenue stayed near 1%, meaning users are consuming vast amounts of cheap inference.

Why Chinese Models Are So Cheap

Chinese companies keep prices low through lower salaries, cheaper infrastructure, and government-subsidized compute. Many have released open-weight models — DeepSeek, Xiaomi MiMo, and Minimax rank among the most cost-efficient on the Artificial Analysis index that measures performance against cost.

The open-source approach means companies can run these models on their own servers, avoiding any data transfer to China entirely. Airbnb CEO Brian Chesky clarified that the company was not sending data to Chinese model developers when lawmakers questioned its use of Qwen and Kimi.

This builds on the broader trend we’ve covered: DeepSeek building its own chips despite export controls and Kimi K2’s open-weight coding model both signal China’s strategy of open-source ubiquity over premium pricing.

The Political Problem

The House China Select Committee launched investigations into Airbnb and Anysphere (maker of Cursor) after both disclosed using Chinese open models. The concern: even if data doesn’t go to China, using these models builds dependency on Chinese AI infrastructure and funnels revenue indirectly to Chinese companies through cloud provider agreements.

Microsoft is reportedly exploring DeepSeek or another open-source model as a lower-cost alternative for Copilot Cowork, which currently runs on Anthropic and OpenAI systems, Axios reported. If Microsoft — the company that backs OpenAI — is looking at Chinese alternatives, the price pressure is systemic, not fringe.

Ruben Garcia Jr., a Dallas developer, pays $500/month for Claude and ChatGPT for complex tasks, plus $200/month for Minimax, Kimi, and Xiaomi MiMo for 90% of his workload. “I don’t mind Chinese companies looking at my data,” he told Rest of World. “If the Chinese models come out and they are frontier and cheaper, I’m going that direction.”

What Happens When OpenAI Cuts Prices

OpenAI is considering significant price cuts as it competes with Anthropic for enterprise users, the Wall Street Journal reported. That price war is a response to the same pressure: Chinese models are eating the bottom of the market, and US providers can’t ignore it.

But price cuts alone won’t close a 20x gap. Chinese companies benefit from cheaper labor, cheaper power, and state subsidies. US providers are cutting margins on models that cost billions to train. The math doesn’t favor the American side at the commodity layer — which is exactly the layer most enterprise AI work sits on.

As Poe Zhao, founder of China tech newsletter Hello China Tech, told Rest of World: “Chinese models are becoming part of the global AI infrastructure layer. But usage is only the first step. The next test is whether Chinese model developers can convert that usage into revenue, enterprise trust, and durable distribution.”

NZ Angle

New Zealand companies face the same cost calculus. A Kiwi startup paying USD for Claude or GPT-4 faces 20x the cost of running DeepSeek on local infrastructure. The open-source route — downloading Qwen or DeepSeek weights and running inference on a local GPU cluster — avoids both the cost problem and the data sovereignty concern. It also avoids the geopolitical scrutiny that US companies face, since NZ has no equivalent of the House China Select Committee.

But NZ also has no domestic frontier model. The choice is between US models at premium prices and Chinese models at commodity prices. That’s not a choice — it’s a price signal.

❓ FAQ

Can companies use Chinese AI models without sending data to China? Yes. Open-weight models like DeepSeek and Qwen can be downloaded and run on company-owned servers or through US cloud providers. No data needs to leave the originating country.

Why haven’t more large enterprises switched? Regulated industries face compliance, data security, and geopolitical risk concerns. Medium-sized businesses with less regulatory burden are the growth market — they feel the cost pressure most acutely.

Is the quality actually comparable? For most tasks — coding, text generation, voice recognition — users report similar quality. Chinese models still lag on the most complex reasoning and frontier capabilities, but the gap is narrowing and the price difference makes the trade-off obvious for routine work.

What is the US government doing about it? The House China Select Committee has launched investigations into companies using Chinese models. Lawmakers are examining national security risks, but no legislation has yet banned the use of open-source Chinese AI models.

🔍 THE BOTTOM LINE

The AI market is bifurcating. Frontier capabilities stay expensive and American. Everything else — the 90% of tasks that power most business AI usage — is becoming a commodity, and China is the low-cost producer. DoorDash, Siemens, and Airbnb switching isn’t a betrayal of US AI; it’s the market working as markets do. The real question isn’t whether companies will use Chinese models. They will. The question is whether US policy can hold the frontier layer while losing the commodity layer — and whether that’s a stable equilibrium or a temporary one.