On July 16, China’s Moonshot AI launched Kimi K3, a 2.8-trillion-parameter open-weight Mixture of Experts (MoE) model that could give developers and poorer countries cheaper autonomous AI agents, simultaneously expanding Beijing’s influence, and sending a harsh warning to Silicon Valley’s AI companies and chip investors that size alone no longer guarantees control.
Kimi K3 could widened access to advanced AI, but it also gives China a stronger position in the tools, standards, and business systems that developers worldwide may depend on. Kimi K3 is seeking to be true to its word, promising to take open-source AI into the geopolitical power play between the US and China.
The model arrives as Chinese developers compete with leading US companies through openness, lower prices, and practical performance rather than closed systems alone. For many developers, those autonomous agents market advantages can matter more than a small technical lead held by a closed model.
Giant Model with Smaller Price
Moonshot describes their LLM powered autonomous agents Kimi K3 as the world’s first open model in the three-trillion-parameter class, and the largest open-source AI model released so far.
Scheduled for full open-weight free release on July 27, Kimi K3 delivers frontier level reasoning, extended coding capacity, and a 1-million-token window that competes directly with Silicon Valley’s Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol.
Beijing is trying to make open-source AI one of its key geopolitical assets, with Kimi’s offering near-frontier intelligence at just $3 per million input tokens – that’s just $0.30 for cached queries.
Moonshot reported “frontier-level performance across our evaluation suite, consistently outperforming other tested models,” although its own assessment acknowledged that it “still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol.”
Powered by a hyper architecture featuring 896 expert modules that activate just 16 experts per inference step. Kimi K3 also achieved top honors on Arena.ai’s Frontend Code Arena with a score of 1,679, successfully outperforming US closed systems in human-inference blind testing.
It’s worth noting that the model demonstrated its autonomous capabilities by operating for 48 hours straight to design, optimize, and verify an AI accelerator chip using open-source electronic design automation (EDA) tools.
Kimi K3 was designed for long and complex workflows and to be one of the best autonomous AI agents. It combines native vision capabilities with a one-million-token context window, supporting lengthy coding sessions, large software repositories, and tasks that mix text, images, and other information.
The Kimi-parent said the autonomous AI agents model completed in about two hours a task that would normally take an experienced researcher between one and two weeks.
Moonshot’s model will allow universities, startups, public institutions, and independent developers to use advanced capabilities without paying US models’ high prices to limit depending entirely on American platforms.
Budget, in this case, matters for users in third world countries, where computing budgets and access to expensive models remain limited. A free downloadable model could help local developers build education tools, health services, translation systems, scientific platforms, and government applications around local needs.
Open-Source AI Becomes China’s Influence Strategy
Seen as the second “DeepSeek shock,” Moonshot’s AI release will have an impact on autonomous agents market, showing how quickly confidence can shift from one player to another.
China’s model is built exclusively to go against the US’ multi-billion-dollar closed clouds, creating a crack in the American AI infrastructure foundation, with Beijing seeking architectural efficiency and cost-friendly accessibility to outpace the American cloud clouds.
After the announcement, US technology shares weakened with the Philadelphia Semiconductor Index recorded its sharpest weekly fall in more than a year. Nvidia, AMD, and Broadcom also declined as investors questioned whether American companies could protect their advantage.
The market reaction suggests that agentic AI autonomous systems investors are no longer looking only at model size or chip demand.
So, will costly US systems be able to defend their prices when Chinese rivals offer equally strong – and in some cases stronger – performance for less than one dollar per million tokens, compared with tens of dollars for some Silicon Valley’s service from autonomous AI agents?
Building autonomous AI agents and then having it being released openly and cheaply can reduce the value of exclusivity, weaken pricing power, and make advanced AI easier to copy, adapt, and distribute across markets that Western companies do not fully serve.
The pressure of being the best autonomous AI agents is spreading inside China as well.
Shares of Zhipu and MiniMax fell in Hong Kong after Kimi K3’s announcement, showing that Moonshot is not only competing with US leaders. It is also forcing Chinese developers to defend their own prices, products, and investor expectations.
China’s wider enterprise AI push strengthens that autonomous AI agents strategy.
At the World Artificial Intelligence Conference in Shanghai, Ant Group, Tencent, Alibaba, and Baidu promoted agents built for business operations. Their autonomous agentic AI products aim to place Chinese AI inside daily corporate work, from desktop control and cloud tasks to resource management and customized applications.
Ant Digital Technologies presented Agentar 2.0 with 200 prepared digital expert templates and hundreds of agent tools. Tencent expanded WorkBuddy across mobile platforms and smart glasses, while Alibaba introduced Meoo Team. Baidu said daily queries to its DuMate agent had risen twentyfold since March.
Ant Group chief executive Cyril Han said, “This shifts the focus from individual efficiency to a wholesale upgrade of the organisation’s collective productivity. We believe this is where the ultimate value lies.”
Both autonomous AI agents demonstrated China’s capacity to build more than large models, as it creates an open, lower cost ecosystem that connects models, agents, devices, and enterprise software. Each new user gives Chinese platforms more reach and places Beijing closer to the rules governing how AI is used.
Western institutions may still choose US AI autonomous agents and models because of data security, regulation, or trust. The Kimi K3 lesson here is that technical leadership alone does not guarantee commercial control.
Developers will follow autonomous AI agents access and price, poorer countries will follow opportunity, and markets will punish any company that mistakes scale for permanent power.
Europe’s fear of losing access to US and Chinese AI is pushing Brussels toward building its own digital autonomy, but security is only the surface. The real contest is over contracts, market power, and who controls the continent’s next wave of technology spending.
Washington has already tested restrictions on access to advanced American systems, while Beijing is reportedly discussing limits on Chinese model exports and the China tech self reliance.
The backup plan of the EU digital sovereignty strategic autonomy was never truly European, just another form of dependence, cheaper than US technology, but still controlled elsewhere.
Europe’s Digital Autonomy Fight about Economic Power
Due to their budget-friendly nature, Chinese models became more attractive, easier to download, and often available as open-source systems. For European startups, that means reduction in operating costs that will allow smaller developers to compete without paying American prices.
Yet, the China tech self reliance US rivalry means Europe’s cheaper AI options are still tied to a geopolitical contest that’s completely out of its control, and reach.
This reduced the European digital self determination that could have taken place.
“The Opus task cost me €24 and the DeepSeek task €1.30, something in that area,” said Head of the European AI Forum, Daniel Abbou, describing the difference of Chinese technology after using Anthropic’s Opus and a DeepSeek model.
Between May and June, DeepSeek doubled its market share on OpenRouter, but along the way, exposed Europe’s weakness, as its prospect for a digitally enhanced directed autonomy remains dependent on foreign technology.
A European startup may save money today, but its business can still collapse tomorrow if Beijing closes access to its latest models.
For Brussels, AI independence has become an economic project, with governments that finance local platforms can direct spending toward European developers instead of sending more money to Amazon, Microsoft, Anthropic, or Chinese providers.
Local developers gain public contracts, infrastructure support, and a protected customer base. Smaller countries gain bargaining power because they can join shared European systems instead of negotiating alone with US technology giants.
France is already moving in the direction of technological independence by planning to replace Zoom, and other American videoconference tools, with a French alternativeVisio.
Germany is also building a domestic AI platform. Companies in both countries are also working on AI chips designed to compete with US and Chinese technology. Together, these projects aim to keep more investment, technical knowledge, and control inside Europe instead of sending them to foreign technology providers.
These projects remain modest, but the political stance is clear. Europe cannot fully separate from foreign technology, so it must choose where dependence is most dangerous and where local investment can create the biggest economic return.
“One hundred percent autonomy in digital services is not at this stage something that is feasible,” said France’s Minister for Artificial Intelligence and Digital Affairs, Anne Le Hénanff, admitting that complete digital autonomy is unrealistic.
“We just need to decide what we don’t want to be dependent on.”
That decision will determine who receives Europe’s next wave of technology spending.
Demand surges for Chinese AI models amid push by corporations to cut costs
Companies from Silicon Valley to Europe are turning to Chinese artificial intelligence models to cut costs and reduce reliance on US technology. DoorDash, Siemens, and Airbnb are among corporations… pic.twitter.com/L5eUe7iqWG— Mani (@cmani) July 15, 2026
Slower, Divided Market for US Tech Giants
The security argument gives European governments political cover to build local systems, but the commercial impact reaches further. Every French platform, German AI system, European chip project, and public procurement rule creates a smaller space for foreign providers.
US technology giants will not disappear from Europe, even if Europe has technological independence. Their cloud services, data centers, models, and software remain embedded in European business. But growth could be slower, more managed and more split between national projects.
The European Commission is also looking to simplify its legal system with a new rulebook, better consultation tools, and an IT platform to track EU laws, implement rules and regulatory overlaps. In theory, smarter regulation could help local developers spend less time navigating complex procedures and more time building products.
The problem of complete digital autonomy is execution.
Vague timelines, broad urgency exceptions, and understaffed national administrations could leave the system as fragmented as before. As China tech self reliance heightens, Europe cannot protect local developers with speeches while burying them under rules that larger US companies can afford to manage.
The bloc also has little leverage when access to advanced models is controlled abroad. Commission President Ursula von der Leyen recently took a friendly line with US President Donald Trump after Washington restricted international access to some Anthropic models. Europe offered market access and cooperation on AI safety, but the exchange showed how few cards Brussels holds.
Philip Fox of the Kira Center said Europe now needs a long-term answer.
“That’s where we simply need a long-term answer,” Fox said. “How can Europe somehow secure its access to the best models for the next five years?”
Europe does not have that answer yet to complete technological independence. But the search for one could redirect public money, create local suppliers, and give smaller European states more weight in technology negotiations.
The digital autonomy winners would be governments seeking control, local developers seeking contracts, and countries seeking leverage. The losers would be foreign technology giants facing a Europe that is harder to enter, slower to scale, and less willing to accept dependence as the price of innovation.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.
