Cheap Chinese AI Models Rattle Congress and US Firms

What happens when Chinese AI models become the cheaper answer to a U.S. enterprise AI bill that suddenly looks too large to ignore?

That question now sits at the center of a Washington fight over artificial intelligence, corporate costs, and national security. U.S. companies are using more Chinese-developed models as those systems narrow the performance gap and cost less to run, according to PYMNTS. Congress is treating that shift as a security warning. XOOMAR analysis: it also looks like a pricing failure by the U.S. AI market.

Why did Chinese AI models become a budget answer before Washington treated them as a security problem?

The immediate driver is not ideology. It’s cost.

PYMNTS reports that U.S. companies have been hit by “sticker shock” from token-based AI pricing, especially as major U.S. AI companies moved away from subscription models toward billing based on tokens in prompts and generated responses. Once employees are encouraged to use AI heavily, usage can scale faster than finance teams expected.

That cost pressure matters because Chinese AI models are becoming more capable while remaining cheaper to use. Enterprises don’t always need the best model available. They need one that is good enough for the task, fast enough for the workflow, and cheap enough to run repeatedly.

That is the opening Chinese developers are now exploiting.

For XOOMAR readers tracking how pilot programs turn into budget pressure, this lines up with our earlier analysis of why enterprise AI agents can become cost traps. The enterprise buyer’s problem is simple: if every extra prompt creates an extra cost, procurement teams will search for lower-cost routing options.

Congress is reacting to the security implications. But the adoption curve is being pulled by economics.

How fast did OpenRouter usage swing toward Chinese AI models?

The sharpest data point comes from OpenRouter, the one-stop API for AI models cited in the report.

As of Tuesday (July 7), the share of tokens from U.S. companies using OpenRouter that went to Chinese models reached 45%, up from 11.5% at the start of the year, PYMNTS reported, citing CNBC.

That is not a marginal change. It suggests that corporate AI usage is moving toward Chinese models quickly where buyers can choose among providers.

The performance issue is central. The 2025 AI Index Report from Stanford University’s Human-Centered Artificial Intelligence center, cited by AP, found that the U.S. still led in producing top AI models, while China was rapidly closing the performance gap and reached near parity in 2024 on several major benchmarks. AP also reported that China leads in AI publications and patents.

XOOMAR analysis: that narrows the room for U.S. vendors to rely only on technical superiority. If buyers see Chinese models as “close enough” for many workloads, price becomes the deciding variable.

Why does Congress see cheap model access as strategic leverage?

Washington’s concern is that cheap access can become dependence.

A State Department spokesperson, quoted in the PYMNTS report, framed the issue in ideological and security terms:

“The growing use of Chinese AI models by U.S. companies raises serious concerns,” the spokesperson said. Those “AI models are designed to advance Beijing’s narratives, censor dissent and reflect [Chinese Communist Party] ideology and values.”

Congress has already moved beyond general concern. In April, the House Committee on Homeland Security and the Select Committee on China said they would jointly investigate the adoption of Chinese-developed AI models. Their chairmen sent letters to Cursor and Airbnb, warning them over their “use of or exposure” to risks from AI developed in China.

Andrew Garbarino of New York, chairman of the House Committee on Homeland Security, tied the risk directly to cybersecurity capability:

“The Chinese Communist Party is no longer just nipping at our heels in artificial intelligence; it is racing to close the gap in some of the exact capabilities that will shape the future of cybersecurity,” Garbarino said. “Recent reporting that a Chinese open-weight model can match leading U.S. models in certain vulnerability discovery and cybersecurity tasks is highly alarming.”

Airbnb’s response shows how companies are trying to draw a line. The company said its “AI activity runs overwhelmingly on U.S.-origin models” and that it uses a “limited number of China-origin models, all of which are open-source and run only through approved U.S.-based service providers.”

That distinction matters. Congress is not only looking at who built the model. It is also looking at how it is accessed, hosted, and used.

Which policy tools are actually on the table?

The concrete policy push is narrower than a blanket private-sector ban, at least based on the available reporting.

A bipartisan bill called the No Adversarial AI Act would bar U.S. executive agencies from using AI models developed in China, including DeepSeek, as well as models from Russia, Iran, and North Korea, Reuters reported via CNBC TV18. The bill would require the Federal Acquisition Security Council to create and regularly update a list of covered AI models.

Federal agencies would not be able to buy or use those systems without an exemption, such as for research, from Congress or the Office of Management and Budget. The bill also includes a path to remove technologies from the list if there is proof they are not controlled or influenced by a foreign adversary.

The other track is strategic: lawmakers are asking whether the U.S. has a sufficient open-weight AI strategy so American companies and cyber defenders are not forced to choose between expensive or restricted U.S. models and cheap, capable Chinese alternatives.

Andy Ogles of Tennessee put it bluntly:

“When the cheap, capable, easy option for an AI model is Chinese, the rest of the world will build on it.”

That sentence captures the real policy fear: market share can turn into infrastructure dependency before Congress writes the rules.

Why are companies and lawmakers solving different problems?

Executives are trying to control AI operating costs. Lawmakers are trying to reduce exposure to Chinese technology. Security teams are stuck between them, because the rules are still forming.

That split explains why this debate is harder than a normal procurement dispute. A company may see a cheaper model. Congress may see a future choke point. A security team may see an unresolved data-handling question.

XOOMAR analysis: U.S. firms buying enterprise AI in 2026 should assume model provenance will become a board-level issue, especially if the company works with government agencies or handles sensitive operations. That does not mean every Chinese-developed model will be banned from every corporate workflow. It does mean “which model did we use?” is becoming a risk question, not just a technical one.

This also connects to the broader debate over model accountability that XOOMAR covered in Model Risk Lands on AI Firms as Trump Rejects FDA for AI. The common thread is oversight. AI buyers are being pushed to understand not only performance and price, but also governance, hosting, and acceptable use.

Practical steps now look obvious:

• Map model use: Identify where prompts and outputs travel.
• Classify data: Decide what information can be sent to external models.
• Review vendors: Ask whether third-party tools route requests to Chinese-developed models.
• Document exceptions: Keep records when lower-cost models are approved for limited use.
• Track policy: Watch federal agency rules, because procurement standards often influence private-sector compliance practices.

Where does the Chinese AI models fight go from here?

Expect a messy decoupling fight, not a clean private-sector ban.

The clearest near-term path is federal restriction, investigation, and pressure on companies that touch sensitive sectors. The No Adversarial AI Act would target executive agencies. The House committee letters to Cursor and Airbnb show lawmakers are also probing private adoption. The open-weight strategy discussion shows Congress understands the economic side of the problem.

China is moving too. PYMNTS cited a Reuters report that Chinese officials held meetings over the past month with domestic tech companies about possibly restricting overseas access to China’s most advanced AI models, including some that have not been released. Officials also discussed treating leaks or theft of proprietary AI technology as an offense under China’s national security law.

So both governments see the same thing: AI models are strategic assets.

The evidence to watch is specific. If the OpenRouter token share keeps rising, the cost argument is overpowering political risk. If U.S. vendors adjust enterprise pricing or offer more competitive open-weight options, the adoption pressure may ease. If Congress expands restrictions beyond federal agencies, corporate legal and procurement teams will have to move faster.

For now, the thesis is clear: Chinese AI models are not gaining traction in U.S. firms because Washington missed a talking point. They’re gaining traction because price and performance are meeting inside real enterprise budgets. That is exactly why Congress is alarmed.

Impact Analysis

• U.S. firms are turning to cheaper Chinese AI models as enterprise AI costs rise.
• Congress sees the shift as a national security risk, not just a procurement trend.
• The story highlights a pricing challenge for U.S. AI companies as usage-based billing scales.