Strategy for AI Development: Demise of the Intermediate Layers

In the dynamic world of Artificial Intelligence (AI), the market is undergoing a significant shift, splitting into two distinct extremes: commodity and frontier airlines. This transformation, often referred to as the AI barbell strategy, is reshaping the landscape of AI companies, investments, and procurement.

On the left side of the barbell, we find companies and projects that focus on commodity AI. Characterized by open-source models, zero margins, and infinite competition, this side of the market is home to cost-conscious startups, academic institutions, open-source contributors, and hobbyists. Successful examples include Hugging Face, Together AI, Replicate, and local LLM communities, which offer free or cheap options.

The right side of the barbell, on the other hand, is where competitive advantage, quality, and innovation reign supreme. Companies like OpenAI, Anthropic, and DeepMind offer high compensation packages and unlimited resources for cutting-edge research. These firms require massive capital investments, with entry costs of $1B+ and potential requirements of $100B+ to win. They dominate the frontier airlines, offering exponential returns and the potential for existential risk/reward.

The barbell strategy in AI isn't optional; it's mandatory. Companies, investors, and individuals must choose: embrace the extreme safety of commoditization or the extreme risk of frontier competition. There is no profitable middle. The middle ground that feels safe is actually the killing field.

This bifurcation is also evident in the financial model, distribution strategy, and technology stack of the AI market. The financial model is breaking in the middle, with the middle model of SaaS subscriptions not being viable. The distribution strategy is concentrating at the extremes, with self-service, viral, free distribution on the left and direct sales, partnerships, and platforms on the right. Similarly, the technology stack is focusing on open-source, commodity hardware on the left and proprietary models and custom silicon on the right.

Regulatory compliance costs are crushing the middle of the regulatory barbell, making it impossible to compete. On the left side, companies are too small to regulate, have open-source immunity, and distributed responsibility. On the right side, there is direct regulatory engagement, compliance resources, lobbying power, and regulatory capture.

The AI economy is witnessing a power law return, with everything going to winners, nothing to the middle, subsistence to commodity, and no normal distribution. This death of gradualism means no stepping stones to success, no building to scale, no progressive risk-taking, and only extreme positions being viable.

The future barbell structure promises a completely commoditized left side, with quality approaching the frontier airlines, community-driven innovation occurring, and margins approaching zero. The right side will have 2-3 total winners, trillion-dollar valuations, platform monopolies, and attempts at Artificial General Intelligence (AGI).

Notable AI startups that tried to occupy the middle ground between commodity and frontier airlines are either dead or dying, such as Jasper AI, Stability AI, Inflection AI, and Character.AI. Even tech giants like Google are transitioning their AI platforms, with Google Assistant transitioning to a more advanced AI platform called Gemini starting in 2025.

In the AI markets, companies must either be on the level of OpenAI or commodity providers, with no profitable middle ground. Enterprises are adopting a barbell approach to procurement, using free or cheap options for commodity needs and paying for the best for strategic needs, with middle vendors being eliminated.

VCs face their own barbell, with safe bets not investing in AI, risky bets backing potential OpenAI competitors, and middle bets guaranteeing losses. Regional AI hubs are failing, and talent is migrating to the extremes, with San Francisco being an AI hub for frontier airlines, while the open internet is a hub for commodity AI.

In conclusion, the AI market is undergoing a significant transformation, splitting into two extremes: commodity and frontier airlines. The barbell strategy, which involves holding extremely safe assets and extremely risky bets, avoiding the middle entirely, is becoming the norm. Companies, investors, and individuals must make a choice: embrace the extreme safety of commoditization or the extreme risk of frontier competition. There is no profitable middle.