Kimi K3: 2.8 trillion Chinese AI model of parameters suspends new signatures after explosive demand

The Chinese artificial intelligence model Kimi K3, from Moonshot AI startup, interrupted new subscriptions a few days after its release — a direct response to the volume of orders that exceeded the available infrastructure capacity. With 2.8 trillion parameters, the model reached the market causing agitation in the U.S. technology industry and overthrowing giant actions such as NVIDIA and Microsoft. The decision, according to the company, is temporary: new signatures must be resumed as soon as the computational capacity is expanded.

Kimi K3: Why did the demand overload the infrastructure?

The Kimi K3 is a large-scale language model (LLM) with impressive 2.8 trillion parameters — volume requiring high-performance GPU clusters and robust interconnection networks. Moonshot AI, a startup founded by former researchers from large Chinese laboratories, did not publicly reveal the exact size of its infrastructure, but the suspension indicates that capacity planning fell short of the explosion of interest. Chinese technology companies, which previously depended on foreign semiconductors, now run to expand data centers with national chips, but the offer is still limited.

Kimi K3 and the effect on the global technology market

The release of Kimi K3 led to an offer shock in the sector: actions by American companies such as NVIDIA, AMD and Microsoft fell in the week of the announcement, reflecting the fear that China might reduce its dependence on Western hardware and software. Meanwhile, the LLMs race of trillions of parameters remains heated. For Brazilian developers, the episode signals that the Sino-American competition for computational power can imprison and delay access to cutting-edge models in the short term.

Kimi K3: Moonshot AI plans to resume subscriptions

The startup stated that it is working to expand its computational capacity "as soon as possible", but did not give concrete deadlines. Sector analysis indicates that the bottleneck is in the scarcity of advanced GPUs — US sanctions limit the export of chips like NVIDIA H100 to China, forcing companies to migrate to domestic alternatives, such as Huawei accelerators. It is the same kind of challenge that led Anthropic to launch Claude Sonnet 5 with a focus on reasoning efficiency, while Moonshot seeks to balance power and infrastructure.

Frequently Asked Questions about Kimi K3

When will Kimi K3 accept new signatures again?

Moonshot AI did not confirm a date, but indicated that the resumption will take place as soon as the computational capacity is expanded — it is estimated that in a few weeks.

Is the model available to Brazilian companies?

Currently, Kimi K3 is only accessible via direct signature on the Moonshot AI platform, without explicit geographical restrictions, but temporary suspension prevents new accessions from any country.

How does Kimi K3 compare to models like GPT-4?

With 2.8 trillion parameters, Kimi K3 surpasses GPT-4 in scale (estimated at 1.7 trillion), but performance metrics in public benchmarks have not yet been widely disclosed by startup.

---

Read the original news in AP News.