Investor Chamath Palihapitiya has published what he calls an "AI investing guide" on X, mapping where he expects capital to flow across the artificial intelligence market. The Social Capital founder argues that while the fastest returns sit in power and data-center real estate, the durable long-term margins will belong to "harnesses" and the applications built on top of them. The framework arrives as hyperscalers including Microsoft, Google, Amazon, and Meta collectively commit hundreds of billions of dollars to AI infrastructure, intensifying competition for the finite resources — land, electricity, and chips — that underpin the entire stack.

Land, Power, and Shell: The Fastest Path to Returns

Palihapitiya categorized AI investment opportunities into layers, beginning with what he calls LPS — short for land, power, and shell. The term refers to the physical footprint of a data center before any chips are installed.

He described this layer as "still the most obvious and fastest path to cash on cash returns," adding: "Lots of value can be assembled and traded quickly at this layer. And as data centers get more pushback, energized land can explode in value. Very bullish here."

Palihapitiya said he and his partner, Anita Verma-Lallian, have secured close to 6 gigawatts (GW) of power running through 2029 — a scale comparable to the output of several large nuclear power plants. He had previously stated that zoning-approved land and silicon access give their owners negotiating leverage over everyone downstream in the AI supply chain. Securing power commitments of this magnitude has become increasingly difficult across the industry, as grid constraints and permitting timelines in major U.S. markets can stretch data-center projects out by years.

The Harness Layer: Where Durable Margins Reside

Above the physical infrastructure, Palihapitiya's central thesis centers on what he calls the "harness." In a July post on X, he wrote: "A modern harness + open model will crush your token consumption but keep your performance."

A harness, according to a Hugging Face glossary published on May 25, is the software wrapped around an AI model that determines what the model sees, which tools it can call, and when it stops. Anthropic's Claude Code, OpenAI's Codex, and Google's Antigravity are all examples of harnesses. Claude Code is described as "the agentic harness around Claude" in its official documentation.

Palihapitiya argued that "the harness helps enterprises own their proprietary context (what Alex Karp calls their 'alpha')," which he defined as including data, workflows, and business rules. Karp, CEO of Palantir Technologies, has previously argued that proprietary data is the defining competitive asset in enterprise AI.

Applications as Long-Term Winners

Palihapitiya's investment thesis extends to applications as well, which he identified as another long-term winner. He wrote: "Every company, with the right harness, can now imbue their alpha into the software that runs their company." The framing positions existing enterprises — not just AI labs — as potential beneficiaries if they can integrate proprietary data and workflows into model-powered applications.

Industry Figures Weigh In

Several industry leaders responded to Palihapitiya's post, with many echoing his emphasis on harnesses.

Xiaoyin Qu, founder of Tycoon AI, expressed support for the harness thesis, stating that a harness "will create margin regardless of if the model gets commoditized," because the right one can unlock large, long-horizon jobs that are worth more than any single model output.

Aaron Levie, CEO of Box, responded to a separate post highlighting the performance of various AI agents, stating that the harness is "going to become the most important variable" in the AI stack, sitting alongside raw model capability.

Context: Palihapitiya's Earlier Questions on AI Spending

In mid-July, Palihapitiya posted on X questioning the current state of AI spending, asking whether it was paying off for anyone beyond the handful of firms already collecting revenue. He pointed to buyers who can now spend $0.50 per million leading-edge tokens instead of $56 for the same volume.

His harness thesis appears to be a direct response to that concern: if models become cheap and interchangeable, value migrates to whoever controls the data, the workflows, and ultimately the applications built on top.