Why US Companies Are Quietly Switching to Chinese AI Models That Cut Costs by Up to 90%

Silicon Valley engineers have begun replacing default large language models with Chinese open‑source alternatives, cutting AI spending dramatically.

01 Chip stocks plunge, AI bubble warning

In July, U.S. chip equities experienced a sharp decline, triggering panic in Korean and Japanese markets. Even massive contracts—Samsung, SK Group and Nvidia signing near‑trillion‑dollar deals—did not restore confidence, leading to panic‑selling across U.S. and Korean chip stocks.

"Since November 2022, AI‑related company market caps have grown by about $27 trillion. If AI profitability falls short of expectations, the market could shrink." – Goldman Sachs & IMF

Wall Street warns of “circular trading” in the AI supply chain, noting Nvidia’s financing and equity stakes in both upstream and downstream chip firms, which could distort true market value. Time Magazine highlights a fundamental mismatch between trillions invested in AI infrastructure and only billions spent on AI services.

02 Chinese large models rewrite the rules

From AlphaGo to ChatGPT, the United States once led large‑model algorithms, but by 2026 the landscape has shifted.

On July 16, the Moon Shadow team unveiled Kimi K3 at the Shanghai World AI Conference.

2.8 trillion total parameters—world’s first open‑source model of this scale.

In the third‑party Frontend Code Arena benchmark, Kimi K3 scored 1,679 points, topping the list and becoming the first open‑source model to surpass GPT, Claude and other closed‑source leaders in front‑end code ability.

Elon Musk called it “impressive.” Nvidia CEO Jensen Huang added, “The US should not ban these models. Chinese models are excellent.”

03 "90% cheaper" – the numbers behind the claim

CNBC cites experts stating Chinese open‑source models are 60‑90% cheaper than Anthropic and OpenAI offerings while delivering comparable performance.

Coinbase (NASDAQ‑listed crypto exchange) switched its internal LLM gateway to Zhipu GLM 5.2 and Moon Shadow Kimi; token usage grew exponentially while AI spend was "almost halved." CEO Brian Armstrong predicts 80% of workloads will run on models costing 99% less within 12‑18 months.

Lindy (AI executive‑assistant startup) migrated its entire stack from Anthropic Claude to DeepSeek, reducing costs to one‑tenth. CEO Flo Crivello says the cost curve dropped "off a cliff," saving millions of dollars annually.

Airbnb adopted Alibaba’s Qwen model for customer service, describing it as “fast and cheap.”

DoorDash and Siemens migrated backend batch jobs to DeepSeek and Qwen.

"More than 90% of us don’t need Anthropic’s giant model. Companies just need a model that’s good enough for business." – LilyList founder Mayn Hold

Hugging Face ML lead Yacine Jernite observes that “more companies are gravitating toward cheaper, self‑controllable AI solutions—often Chinese offerings.”

04 Data doesn’t lie: Chinese models now dominate usage

On the OpenRouter platform, which aggregates global AI model calls, Chinese model token traffic now exceeds 60% , up from under 2% a year ago.

The U.S.–China Economic and Security Review Commission estimates that about 80% of U.S. AI startups are using Chinese open‑source models.

Alibaba’s Qwen series has surpassed 10 billion downloads, generated over 200 k derived applications, and on Hugging Face now exceeds Meta’s Llama as the most‑downloaded open‑source model family.

05 Silicon Valley panics: banning Chinese models could topple hundreds of startups

U.S. government threats of sanctions have sparked fear in the startup ecosystem.

On July 24, dozens of U.S. tech firms and institutions issued a joint statement supporting open‑weight AI models, viewed as a direct response to potential restrictions.

Nearly 200 U.S. tech startups signed a letter urging the government not to cut off access to Chinese models, warning it would weaken U.S. competitiveness.

"Hundreds of companies would go bankrupt instantly. Banning Chinese AI models would force small firms to turn to Anthropic’s expensive offerings." – Particle founder Su Hai‑er Doxi

Nvidia CEO Jensen Huang also warned that limiting open‑source models could undermine U.S. AI competitiveness.

When U.S. AI costs become unaffordable for SMEs, Chinese open‑source models serve as a “lifeline”; cutting that line would first harm U.S. startups.

06 Conclusion: from “Copy‑to‑China” to “Copy‑from‑China”

MIT Technology Review notes that China no longer copies Silicon Valley; Silicon Valley is now studying China.

From DeepSeek to Kimi K3, from GLM 5.2 to Qwen, Chinese large models combine open‑source licensing, low cost, and high performance, redefining global AI economics.

Stanford AI Index reports: “China is redefining the economics of AI.”