Chinese startup DeepSeek has released its largest AI model to date, a 685-billion-parameter model that industry observers say could intensify competition with US players.

The model, called DeepSeek V3.1, was made available on the open-source platform Hugging Face this week with little publicity. Despite the quiet rollout, early benchmark results reportedly suggest the model performs on par with proprietary offerings from OpenAI and Anthropic.

The launch could broaden access to advanced AI capabilities while raising new questions about the global balance of technological power between China and the US. For enterprise IT leaders, the release raises fresh questions about whether open-source Chinese models could reshape procurement strategies that have long centered on US vendors.

The release follows OpenAI’s decision earlier this month to publish its first open-weight models since GPT-2, positioned as offering strong performance at lower cost. Chief Executive Sam Altman told CNBC that rising competition from Chinese open-source models, including those from DeepSeek, influenced the move.

Altman also cautioned that the US may be underestimating the pace and scale of China’s advances in AI, adding that export restrictions by themselves are unlikely to provide a lasting safeguard.

Global AI rivalry

DeepSeek models have long attracted attention from developers and global enterprises for their large parameter sizes and wide context windows. With V3.1, the company has moved further into the territory traditionally dominated by US players.

“DeepSeek’s 685B open-source model accelerates commoditization of raw AI capability, eroding the closed-model moat of US players,” said Oishi Mazumder, senior analyst at Everest Group. “OpenAI, Anthropic, and Google must now differentiate through trust, governance, and enterprise-grade ecosystems rather than sheer model size.”

Analysts note that the appeal of V3.1 is not only its size but also its accessibility. By removing barriers to frontier-scale AI, DeepSeek is shaping the conversation around what enterprises expect from open models.

“By granting developers unfettered access to frontier-scale AI capabilities, DeepSeek has upended prevailing competitive dynamics,” said Prabhu Ram, VP of the industry research group at Cybermedia Research. “It raises the bar on model size, performance, and cost, forcing incumbents anchored in proprietary, subscription-driven AI systems to rethink their strategies.”

DeepSeek supporters are now waiting for the company’s next-generation model, R2. The FT reported last week that its launch was delayed after the firm struggled to complete training on Huawei’s chips, forcing a shift to Nvidia hardware for training while reserving Huawei processors for inference.

Enterprise adoption and risks

Analysts say DeepSeek V3.1 is unlikely to spark immediate disruption in the US market, where enterprises still prefer domestic vendors with deeply integrated platforms and enterprise-grade support.

“Two things will need to change for models like this from DeepSeek to gain traction in the US,” said Lian Jye Su, chief analyst at Omdia. “To begin with, there has to be an easing of the geopolitical tension, then DeepSeek needs to show that it can outperform other Western open-source models from Meta, Mistral, and Nvidia in a significant way in specific enterprise use cases.”

Outside the US, however, DeepSeek’s permissive open-source licensing could appeal to CIOs looking to accelerate internal AI development, lower costs, and gain more control through customization and self-hosting, according to Ram.

“But the model’s massive scale demands significant computing resources, requiring enterprises to carefully consider infrastructure, compliance risks, and export restrictions before adoption,” Ram added.

DeepSeek may also struggle to match the support, security, and compliance standards of enterprise-grade models like Anthropic’s Claude Sonnet 4, said Neil Shah, vice president of research at Counterpoint Research.

But the broader strategic concern is not just about adoption but about the shifting balance of global tech leadership. The US has traditionally dominated in semiconductors, operating systems, and cloud infrastructure, while China is now pushing aggressively to lead in AI despite sanctions.

“DeepSeek’s success in advancing its capabilities underscores the intensifying tech rivalry, where both sides are racing to innovate regardless of geopolitical constraints,” Shah added.