.jpg?w=1024)
One of the world’s largest open-weight AI models is moving closer to the enterprise data center.
Moonshot AI’s Kimi K3 is now available through Dell Enterprise Hub and DigitalOcean’s Inference Engine, giving organizations new options for deploying the model on-premises or accessing it through managed cloud infrastructure. Moonshot AI says Kimi K3 contains 2.8 trillion parameters, uses a mixture-of-experts architecture, and supports a context window of up to one million tokens.
The two integrations reflect a broader enterprise AI shift: Businesses increasingly want access to powerful models without being locked into either a proprietary API or a single deployment environment.
How Dell and DigitalOcean are offering Kimi K3
Dell Enterprise Hub allows customers to deploy Kimi K3 on supported Dell infrastructure using containerized software obtained through Hugging Face. Dell said the containers are optimized for selected deployment systems and accompanied by SHA-384 image hashes and malware-scan results.
DigitalOcean’s Inference Engine takes a cloud-based approach. Developers can access Kimi K3 through its serverless inference service without provisioning the underlying infrastructure, or deploy the model using a customized cloud configuration.
Together, the integrations give enterprises a choice between keeping the model within infrastructure they control and using a managed service for faster API-based development.
This flexibility comes as Chinese AI models gain greater traction among U.S. businesses on OpenRouter, challenging the assumption that enterprise AI workloads will remain concentrated among U.S. model providers.
Why local deployment may appeal to enterprises
Local deployment may appeal to organizations that handle financial information, intellectual property, proprietary source code, or other sensitive data. Running Kimi K3 on infrastructure controlled by the organization could reduce the need to send that information to an external model provider.
Open weights can also give development teams more visibility and customization options than a closed model accessed exclusively through an API. However, data sovereignty still depends on how the model, supporting software, telemetry, and network access are configured.
These considerations are becoming more important as Chinese open-weight models move from developer experimentation into enterprise procurement.
Kimi K3’s size still comes with infrastructure demands
Despite its 2.8 trillion total parameters, Kimi K3 activates only a portion of the model for each request. Moonshot AI says the model uses approximately 104 billion active parameters and engages 16 of its 896 experts through its mixture-of-experts architecture.
That design can reduce inference requirements compared with activating every parameter, but it does not make the model lightweight. Enterprises would still need substantial storage, memory, and accelerator capacity to operate it locally.
Moonshot AI also offers quantized weights that reportedly reduce the model’s storage footprint from approximately 5.6 TB to 1.56 TB, a reduction of roughly 72%. Quantization can make large models easier to store and run, although its effects on accuracy and performance depend on the implementation.
Where open-weight models fit alongside proprietary AI
Kimi K3’s arrival on Dell and DigitalOcean gives enterprises another alternative to proprietary AI services, but it does not make open and closed models interchangeable. Performance can vary by benchmark and workload, while deployment cost, governance requirements, customization, and data controls may matter as much as raw model scores.
Ihab Tarazi, chief technology officer and senior vice president at Dell Technologies, argued in a company blog post that enterprises should move beyond a simple “open-versus-proprietary” debate. Instead, he described the two approaches as serving different workload, governance, performance, and deployment requirements.
Dell is not alone in framing open-weight models as part of a mixed AI ecosystem. Nvidia and Microsoft recently joined other technology companies in supporting continued access to open-weight AI.
Kimi K3’s expanding availability suggests that enterprise AI competition will increasingly be decided by more than benchmark results. Where a model can run, how much control it gives customers, and what infrastructure it requires may prove just as important as its raw capabilities.