When DoorDash said Kimi K2.6 could deliver better low-level AI work at lower cost, Congress did not read it as a procurement anecdote. It read it as a national-security signal.
DoorDash's use of Moonshot AI's Kimi K2.6 for some workloads has triggered a congressional inquiry, after company leaders publicly suggested the model could outperform some U.S. alternatives on cost-adjusted utility. What is emerging is bigger than one food-delivery company. Cheap, capable Chinese open-weight models are increasingly attractive to U.S. businesses, while lawmakers are treating the same trend as a strategic vulnerability.
DoorDash
Andy Fang and DoorDash AI researchers publicly highlighted Kimi K2.6 as a high-quality, lower-cost option for some tasks.
Congress
Two House committees asked Tony Xu to provide information about how Chinese AI models were evaluated and deployed.
Market
Fortune reported that Chinese models represented 57% of tokens used by U.S. firms on OpenRouter during one week in July.
Pressure
OpenAI and Anthropic face growing price pressure as Chinese models gain adoption.
1. DoorDash surfaced a new standard: not just cheap, but usable
CNBC reported that DoorDash's AI research lab said on X that Kimi K2.6 and Anthropic's Fable 5 vastly outperformed other Anthropic models it had used, including at lower cost. The important detail is not merely that DoorDash experimented with a Chinese model. It is that the company framed the model as a practical choice for lower-level AI work.
That marks a shift in what matters operationally. The frontier question is no longer only whether Chinese models can keep up on benchmark prestige. It is whether they are already good enough, cheap enough, and flexible enough to win real workloads inside U.S. companies.
2. Congress is treating this as a security problem, not a sourcing footnote
According to CNBC and a House Select Committee on the CCP press release, Chairmen John Moolenaar and Andrew Garbarino asked DoorDash CEO Tony Xu for information and documents related to the company's evaluation and deployment of Chinese AI systems. The committees explicitly acknowledged the appeal of PRC-developed open-weight models: competitive capability, lower cost, greater customization, and alternatives to a narrow set of U.S. proprietary providers.
But lawmakers said those practical advantages do not remove the need for safeguards or reduce national-security concerns. Their stated worries include insufficient guardrails, malicious code injection, attacks against AI systems, and the broader strategic risk of growing dependence on models developed by entities under PRC jurisdiction.
3. The pricing war is no longer hypothetical
The congressional reaction only makes sense in the context of what is happening across the market. Fortune reported that Chinese AI models accounted for 57% of tokens used by U.S. firms on OpenRouter in one week in July. Moonshot, DeepSeek, Z.ai, Xiaomi, and Tencent all appeared near the top of developer usage rankings as Chinese models gained share on both performance and price.
At the same time, U.S. labs are being dragged into more aggressive pricing responses. OpenAI has already cut prices on parts of its GPT-5.6 family, while Anthropic has faced stronger cost comparisons from both developers and enterprise buyers. The contest is increasingly about unit economics, not just about who owns the strongest single model.
4. The core issue is what firms optimize for
Backdoor and sleeper-agent fears are part of the political debate, but the harder reality is economic. Companies making day-to-day AI decisions often optimize for a simple bundle: quality, cost, and deployment flexibility. Chinese open-weight models increasingly offer all three, especially when businesses want to self-host, fine-tune, or avoid paying frontier-model premiums.
That is why the deeper policy question is not merely whether Chinese models are risky. It is whether the U.S. can supply open-weight alternatives that are secure, affordable, and competitive enough that domestic firms do not feel pushed toward PRC-developed options by ordinary cost discipline.
5. The next question is how far this has spread
DoorDash is not the only company under scrutiny. CNBC previously reported that lawmakers were also probing Airbnb and Anysphere, the maker of Cursor, over exposure to or use of Chinese AI models. That suggests Congress is less focused on one company than on mapping how deeply Chinese models have entered U.S. commercial AI infrastructure.
Seen in that light, DoorDash's Kimi episode is not just about one procurement choice. It is an early, visible case of a larger trend: Chinese AI entering the U.S. market through the ordinary logic of enterprise cost optimization, even as Washington tries to constrain China at the strategic level. The next phase of the debate will hinge on whether policymakers decide the real threat is usage, dependence, or both.
Key points
- DoorDash used Kimi K2.6 for some workloads and publicly praised its price-quality balance
- Two House committees requested information about evaluation, deployment, and security safeguards
- Chinese models are gaining meaningful usage share among U.S. firms on OpenRouter
- The central conflict is shifting from model quality alone to price structure and dependence risk