The Cheap Intelligence Portfolio

The AI trade is usually framed too narrowly.

Most investors ask which model wins, which chatbot wins, or which app becomes the next platform. Those questions matter, but they miss the larger point. The better question is: what happens if intelligence gets much cheaper?

That is the AI-optimistic investment thesis. Not that every AI company wins. Not that every demo becomes a product. Not that valuations stop mattering. The thesis is that useful cognitive work — writing, coding, searching, summarizing, monitoring, designing, testing, researching — becomes cheaper, faster, and more widely available.

If that is true, AI is not just a sector. It is a cost curve.

And when an important input cost falls, the gains do not spread evenly. They accrue to the companies that own bottlenecks, distribution, workflows, data, and balance sheets.

Own bottlenecks, not slogans

The first rule is simple: do not buy AI language. Buy AI leverage.

The best AI investments are not necessarily the companies with the loudest AI branding. They are the companies whose revenue, margins, retention, or strategic position improve because AI exists.

The first bottleneck is infrastructure: chips, servers, cloud, memory, networking, data centers, power, cooling, and cybersecurity. If AI demand keeps rising, this layer gets pulled forward. But this is also where hype can turn into overbuild, so the focus should be on companies with pricing power, scale, and durable demand.

The second bottleneck is distribution. A model with no distribution is a feature. A model inside a productivity suite, cloud platform, operating system, developer tool, or enterprise workflow is a business. This is why mega-cap technology companies remain structurally advantaged. They already own the customer relationship and can embed AI into products people already use.

The third bottleneck is workflow depth. In many industries, the winner will not be the most general model. It will be the tool closest to proprietary data, compliance requirements, customer behavior, and day-to-day process.

The barbell

I would barbell the technology exposure.

Long mega-cap tech, because the largest platforms have the balance sheets, infrastructure, distribution, and acquisition capacity to keep absorbing value.

Long select small-cap AI optionality, because small teams can now build faster, create new vertical products, and become valuable acquisition targets.

Be more skeptical of the middle. Many mid-cap software companies have neither the scale of mega-cap platforms nor the speed of startups. If they do not own a system of record, proprietary data layer, or mission-critical workflow, they risk becoming features inside someone else’s bundle.

The adopters may matter more than the vendors

The cleaner long-term trade may be companies that consume AI, not just companies that sell it.

Biotech and pharmaceuticals are obvious candidates. Drug discovery, molecule screening, trial design, literature review, and clinical operations are expensive, research-heavy workflows. AI does not need to solve biology outright to create value. Even modest improvements in speed or hit rates can matter.

Cybersecurity is another winner. AI means more code, more automation, more agents, more permissions, and more attack surface. The boom creates its own security tax. Own the collectors of that tax.

The basket

Long AI infrastructure, mega-cap distribution, cybersecurity, biotech/pharma productivity, and high-quality adopters that can turn cheaper intelligence into margin expansion.

Avoid companies with vague AI narratives, weak moats, inflated valuations, and no visible path from model adoption to cash flow.

Size speculative positions modestly. Rebalance when AI exposure becomes too large. Sell when the thesis stops showing up in revenue, margins, adoption, or competitive advantage.

The point is not to own everything with AI in the name.

The point is to own the scarce assets in a world where intelligence is no longer scarce.