AI's Great Filter

The enduring advantage is being forged in the unassailable moats of capital, talent density, and control over distribution.

The AI landscape is rapidly solidifying, moving from a wide-open field to a handful of clear market leaders. As Elad Gil notes, for many of the initial generative AI markets, the ambiguity is fading.

The Compute Arms Race

The sheer scale of capital required to compete in foundation models creates a formidable moat. Companies are now planning for multi-billion dollar, gigawatt-scale AI training clusters.

Mark Zuckerberg’s pivot at Meta exemplifies this reality. Faced with falling behind, he is overhauling Meta's entire datacenter strategy—even building massive GPU clusters in tents to prioritize speed and raw compute power.

Reinforcement Learning, the engine behind recent model improvements, is incredibly inference-heavy, demanding vast and efficient compute resources to generate the necessary data and rollouts.

Google’s massive capital expenditure, now forecasted at $85 billion for 2025, underscores that even for incumbents, the risk of under-investing in AI infrastructure is far greater than the risk of over-investing.

Talent, Too, Is Now Table Stakes

Beyond capital, the war for talent has reached a fever pitch. Zuckerberg is personally poaching top-tier researchers with compensation packages reaching hundreds of millions of dollars.

Meta's acquisition of a major stake in Scale AI and the hiring of its former CEO is a direct move to shore up its data and evaluation capabilities, which were critical weaknesses in the Llama 4 development.

Cognition's acquisition of Windsurf follows a similar pattern, consolidating talent and intellectual property in the crucial AI coding space.

The Emerging Business Models & The Power of Distribution

As the technology matures, the business models are also crystallizing, moving beyond simple API access.

Cloudflare's Content Independence Day is a power play to create a new marketplace for AI training data. By defaulting to blocking AI crawlers, Cloudflare is leveraging its significant control over internet traffic (an estimated 20%) to force a shift from a free-for-all to a paid model.

This move, however, has a clear beneficiary: Google. By allowing the Googlebot crawler for Search—which also feeds its most important AI products like AI Overviews—Cloudflare reinforces the incumbent's powerful advantage. No website can afford to block Google, making its access to data a durable moat.

This highlights the ultimate truth of this new era. As Elad states:

"We are starting to see the shift from selling seats, to selling units of cognition."

In a world of "units of cognition," the winners will be those who not only possess the immense compute and talent to generate those units but also control the distribution channels—like Search or the network edge—to deploy and monetize them. The markets are clarifying, and the moats are getting deeper.