Key Takeaways
- DeepSeek is raising approximately $10 billion at a $45 billion valuation, with founder Liang Wenfeng prioritizing AGI research over commercialization.
- The company plans to keep its top models open-source, employing a mixture-of-experts architecture scaling to 1.6 trillion parameters.
- Real-world benchmarks from Faros AI show open-model routes deliver top-tier quality at roughly half the cost of frontier models, accelerating enterprise adoption.
The Open-Source AGI Bet That Could Redefine AI Competition
DeepSeek, the Chinese AI lab spun out of hedge fund High-Flyer, is raising approximately $10 billion at a $45 billion valuation to pursue artificial general intelligence — not a lucrative exit. Founder Liang Wenfeng told investors in May 2026 that fundamental AI research, not commercialization, is the company’s north star. DeepSeek plans to keep its most capable models open-source, a move that sacrifices short-term revenue for long-term credibility and community adoption.
Table of Contents
DeepSeek’s Technical Architecture and Open-Source Philosophy
DeepSeek has already released models like R1 and V4 variants, employing a mixture-of-experts architecture that scales to 1.6 trillion parameters. This design allows efficient inference by activating only relevant parts of the network per token, making it both powerful and cost-effective. Explore DeepSeek’s full technical breakdown right here.
By keeping these models open-source, DeepSeek creates a gravitational pull for developers and researchers. Western labs like OpenAI and Anthropic have moved in the opposite direction, locking their top models behind API subscriptions. Liang’s strategy treats open-source as a moat — the more the community builds on DeepSeek’s work, the harder it becomes to displace.
The funding round, backed by China’s state AI fund and High-Flyer, provides the capital to continue scaling these efforts without immediate pressure to monetize. DeepSeek’s valuation of $45 billion reflects confidence in this long-term approach.
Enterprise Shift: Why Open Source Is Winning the Cost and Quality Battle
The shift toward open-source AI is not just ideological — it is being driven by hard economics and real-world performance. According to a report from The Information, rising API costs from Anthropic and OpenAI have started to erode corporate profit margins, pushing enterprises to explore cheaper alternatives. Data from OpenRouter shows that the majority of tokens processed now go to open-source models, primarily DeepSeek and MiniMax, a reversal from the previous year when Anthropic and OpenAI dominated, as detailed in the following video:
A controlled experiment by Faros AI in June 2026 compared seven model-plus-harness routes on 211 real engineering tasks. The open-model routes — Claude Code paired with GLM-5.2 or Kimi K2.6 — achieved the highest quality scores, statistically close to each other, while costing roughly half as much per task as frontier routes like Claude Opus 4.8 or GPT-5.5. Specifically, the Claude Code + GLM-5.2 route averaged $0.92 per task compared to $1.76 for Opus and $2.06 for GPT-5.5.
This cost-performance advantage is accelerating enterprise adoption. The Information also noted that Canadian AI company Cohere tripled its annual recurring revenue projection after the Fable export controls drove customers to seek vendor diversification. Moreover, initial fears of cybersecurity backdoors in Chinese models have not materialized with evidence, leading Fortune 100 companies — including large financial services providers — to express interest in or actively deploy Chinese open-source models.
The structural forces supporting open-source AI are compounding. The cost gap between self-hosted open models and closed APIs is 70-500x per token, a differential that is unlikely to narrow as open labs continue to match frontier quality. DeepSeek’s commitment to open-source positions it at the center of this shift.
The New AI Landscape: Open-Source as a Moat, Not a Monetization Shortcut
DeepSeek’s strategy flips the conventional AI business model on its head. Instead of monetizing model access through APIs, it builds value through research citations, parameter records, and community lock-in. As Western labs race to lock down their most capable systems, DeepSeek is betting that open distribution creates a deeper, more durable competitive advantage. For the AI industry, this marks a fundamental realignment: the next frontier of competition may not be about who has the best API pricing, but who has the most widely adopted open-source ecosystem.
For enterprises navigating this transition, adopting open-source AI models can reduce costs and increase flexibility. Andres SEO Expert provides specialized programmatic SEO and AI automation services to integrate and scale these models. Reach out to Andres for a consultation, and discover how Andres SEO Expert can help you lead the AI-driven future.
Frequently Asked Questions
What is DeepSeek’s strategy for AGI?
DeepSeek is raising ~$10 billion at a $45 billion valuation to pursue artificial general intelligence (AGI) while keeping its most capable models open-source. Founder Liang Wenfeng stated that fundamental AI research, not commercialization, is the company’s north star. The open-source approach sacrifices short-term revenue for long-term credibility and community adoption.
How does DeepSeek’s mixture-of-experts (MoE) architecture work?
DeepSeek’s models, like R1 and V4 variants, employ a mixture-of-experts architecture that scales to 1.6 trillion parameters. This design activates only relevant parts of the network per token, enabling efficient inference while maintaining high performance and cost-effectiveness.
Why are enterprises shifting to open-source AI models?
Rising API costs from closed providers like Anthropic and OpenAI have eroded corporate profit margins. Data from OpenRouter shows most tokens now go to open-source models (DeepSeek, MiniMax). A controlled Faros AI experiment found open-model routes (e.g., Claude Code + GLM-5.2) achieved top quality at roughly half the cost per task compared to frontier closed models.
How do costs compare between open-source and closed AI APIs?
According to the article, the cost gap between self-hosted open models and closed APIs is 70-500x per token. For example, Claude Code + GLM-5.2 averaged $0.92 per task versus $1.76 for Claude Opus 4.8 and $2.06 for GPT-5.5. This differential is unlikely to narrow as open labs continue to match frontier quality.
What are the risks of using Chinese open-source AI models like DeepSeek?
Initial fears of cybersecurity backdoors in Chinese models have not materialized with evidence. Fortune 100 companies, including large financial services providers, have expressed interest in or are actively deploying Chinese open-source models, indicating growing trust despite geopolitical concerns.
How does DeepSeek’s open-source approach differ from Western AI labs?
Western labs like OpenAI and Anthropic lock their top models behind API subscriptions. DeepSeek treats open-source as a moat: by keeping models open, they create community lock-in and make it harder for competitors to displace them. This flips the conventional monetization model, prioritizing research citations and ecosystem growth over direct API revenue.
What impact could DeepSeek’s funding have on the AI landscape?
The $10 billion round backed by China’s state AI fund and High-Flyer provides capital to scale open-source efforts without immediate monetization pressure. It signals that the next frontier of AI competition may be about the most widely adopted open-source ecosystem rather than the best API pricing, potentially accelerating enterprise adoption of open models.
