Get your free personalized podcast brief

We scan new podcasts and send you the top 5 insights daily.

Joshua Tan noted that open source groups like Eleuther and OLMo stopped training frontier models due to massive capital costs. Consequently, the largest remaining open-source initiatives now rely on sovereign funding, such as Switzerland's national project, positioning public infrastructure backing as open AI's only viable third way.

Related Insights

Open source AI models can't improve in the same decentralized way as software like Linux. While the community can fine-tune and optimize, the primary driver of capability—massive-scale pre-training—requires centralized compute resources that are inherently better suited to commercial funding models.

The path to a competitive open-source AI ecosystem is blocked by a massive capital moat. The cost of a single gigawatt-scale data center has exploded to $100 billion, making it virtually impossible for anyone outside of big tech or nation-states to fund the necessary compute.

While open-source fosters broad innovation, it cannot fund the massive data centers required for cutting-edge AI. The for-profit model is the only mechanism capable of aggregating the immense capital needed to build the fundamental infrastructure that powers frontier models.

The US is behind in open-source AI development because of a fundamental business model problem. American companies struggle to justify spending billions on training a frontier model only to release it for free. Chinese companies can pursue this strategy due to different corporate dynamics and state influence.

A purely non-profit or open-source approach would have stalled AI development. The extreme expense of building ever-larger GPU clusters, which is currently the primary way to make models smarter, required a massive for-profit incentive to attract the necessary capital investment.

Unlike traditional open-source software, training AI models costs millions. To ensure sustainability, model labs are adopting commercial licenses that require large users to pay. This creates an economic incentive structure, similar to the pharmaceutical industry, to fund the high-risk, high-cost R&D for future model generations.

Dean Ball, OpenAI's Head of Strategic Futures, argues that a world dominated by open-weight models could result in "AI communism." In this scenario, AI becomes a state-provided public good rather than a market product, which would eliminate the financial incentives required to fund costly frontier AI development.

An pro-open source stance can be seen as inherently "desalinationist" for the AI industry. By commoditizing models and lowering margins, it becomes harder for frontier labs to underwrite the massive capital expenditures for the next, larger training runs, thus reducing the insatiable demand for compute.

The shutdown of Fable 5 and rising 'token scarcity' created two powerful incentives—cost and sovereignty—for enterprises to diversify away from closed, frontier models. Open-weight models are now being evaluated not just for savings, but for strategic control and resilience.

China's open-source model ecosystem is structurally unstable. The billion-dollar fixed costs for training frontier models are unsustainable for Chinese tech giants who lack a clear AI revenue narrative and cannot match the compute budgets of Western labs like OpenAI or Anthropic.