DeepSeek, the Chinese AI startup behind the influential R1 reasoning model, is deliberately choosing long-term artificial general intelligence (AGI) research over near-term commercial growth. According to internal investor meeting minutes, founder Liang Wenfeng stated that the company is not competing for consumer traffic, enterprise revenue, or trying to become the next ByteDance or Tencent. Instead, products and commercial services are secondary outcomes of its core AGI research mission.

What Is DeepSeek's Long-Term Technical Roadmap?

DeepSeek's path toward AGI follows a specific sequence of technological milestones. The company believes development will progress through chain-of-thought reasoning, AI agents, continuous learning, and eventually embodied AI that can interact with the physical world. Notably, the company is deprioritizing areas like video generation and world models, which Liang views as useful applications but not central to developing true intelligence.

Continuous learning represents the next major technical challenge after AI agents. Current AI systems require users to provide relevant context for each task, but DeepSeek believes future models should learn continuously from experience without constant human input. Coding agents are a priority because they could improve both the company's products and its own research capabilities.

How Does DeepSeek's Strategy Explain Its Business Model?

This research-first approach directly shapes how DeepSeek operates commercially:

• Open-Source Philosophy: DeepSeek plans to keep its strongest models open source and use the same models internally and externally, rather than release weaker versions to the public. This contrasts with competitors who often reserve their best capabilities for paying customers.
• Pricing Strategy: The company's API pricing is designed to recover hardware costs in approximately 10 months, rather than maximize profit margins. This low-cost approach has made DeepSeek's models accessible globally and contributed to its reputation for affordability.
• Product as Research Tool: Consumer products and enterprise services are intended to support the company's longer-term technical goals, not drive revenue. The company views these offerings as ways to fund and validate its AGI research.

What Are DeepSeek's Biggest Constraints?

Despite its ambitious roadmap, DeepSeek faces significant resource limitations. Founder Liang stated that "the biggest gap between us and the US lies in resources," emphasizing that differences in talent, model capabilities, and applications all stem from disparities in computing power. The company currently has computing resources equivalent to about 20,000 Nvidia H-series GPUs, with most hardware delivered over the past two months.

Founder Liang

The scale gap is substantial. Today's largest AI models have roughly 800 billion active parameters, compared with tens of billions for China's most advanced models, a roughly 10-fold difference. Even with its recent $7.4 billion funding round, DeepSeek could not afford to train a model at that scale if it spent all new funding on computing power.

"The biggest gap between us and the US lies in resources. All the differences we see in talent, model capabilities, and applications can be attributed to differences in computing power resources," said Liang Wenfeng, founder and CEO of DeepSeek.

Liang Wenfeng, Founder and CEO of DeepSeek

US export controls restrict Chinese access to Nvidia's most advanced AI chips, and China invests less capital in AI overall compared to the United States. To address these constraints, DeepSeek is working closely with Huawei Technologies to optimize its models for Huawei's Ascend computing ecosystem and developing its own high-level programming language called Tile Language to reduce reliance on Nvidia's infrastructure.

How Does This Strategy Position DeepSeek in the AI Industry?

DeepSeek's research-focused approach distinguishes it from competitors building standalone products and enterprise businesses around their models. The company believes it trails leading US AI companies by 12 to 18 months but has achieved comparable results using only around one-20th of their computing power. This efficiency has made DeepSeek's open-source releases, particularly the R1 reasoning model released in January 2025, globally influential.

The strategy also reflects a broader shift in the AI industry. As the field moves from releasing individual models toward developing agents and systems that can assist with AI research itself, DeepSeek is positioning itself as a research organization first. Its products and API services exist to support that mission, not to compete directly with consumer-focused AI platforms.

This approach helps explain why DeepSeek has gained such attention despite operating under significant hardware constraints. By focusing on research efficiency and open-source distribution rather than proprietary moats, the company has demonstrated that advanced AI capabilities don't require unlimited computing resources. For the broader AI industry, DeepSeek's strategy suggests that the path to AGI may not require the largest budgets, but rather the most efficient research priorities.

" }