Raoul Pal, co-founder of Real Vision, said in a podcast conversation with Wall Street strategist Jordi Visser that AI is reshaping the core logic of business. In their view, the old framework built around labor and capital is giving way to one centered on compute and energy. They argued that chips, electricity, data centers and AI agents are setting up a new supercycle, while the main near-term constraint is not weak demand but limited infrastructure.
Compute and power replace labor as the main constraint
Visser said business expansion used to depend on borrowing, hiring and finding office space. In the AI era, the harder question is whether companies can secure enough chips and enough power. If supply cannot keep up, bottlenecks and shortages follow, slowing earnings growth even when demand remains strong. Pal said he has built a dashboard to track the growth in “intelligence output per unit of energy,” and argued that GPUs and AI have pushed that curve into double-exponential territory.
Pal also said data center construction is running at only 30% of what has been announced. Combined with US-China competition and the lack of any single frontier AI company able to dominate on its own, he said that makes a supercycle “almost inevitable.” In his telling, bottlenecks do not end the trend. They redirect capital toward power systems and other infrastructure, creating what he described as a capex cycle that could be among the largest seen in history.
Algorithms keep improving while infrastructure falls behind
Visser argued that rising AI capability is not simply a matter of building more data centers. Better algorithms, reinforcement learning from human feedback, and stronger reasoning are also lifting performance. He described the process as non-linear: not a standard industrial model where more physical buildout directly produces more output, but one where recursive learning and algorithmic improvement keep accelerating results.
Both speakers said the market still does not fully grasp what an AI agent economy means. Pal said total addressable markets were historically tied to humans, but agents change that ceiling. Visser framed it as the arrival of billions, or even tens of billions, of new “thinkers” that consume compute without the normal costs attached to human workers. That, he said, forces a different way of reading the business cycle.
Capital crowds into infrastructure before rotating elsewhere
On market leadership, Pal said capital will not stay concentrated in names like Nvidia forever. Once power, materials or production capacity become limiting factors, attention and investment will move toward the parts of the stack that remove those blocks. Visser pointed to storage and solid-state batteries as examples, adding that if peak-load issues can be solved, the US power grid is “fully sufficient through 2030.”
They also said the application layer is being read too narrowly. Visser argued that investors remain fixated on SaaS, while large pools of capital are also flowing into what he called “human software.” He used Eli Lilly and its GLP-1 drugs as an example, saying the company has data center capacity with thousands of GPUs, and that those cash flows may be funding the next stage of human bioscience research.
Personal AI stacks and edge computing enter the discussion
The conversation also moved into how AI tools are changing individual work. Pal said he is building a “GMI brain” based on his written work, transcripts and posts from the past 21 years. He also described a tool called “Lens” for analysis and a separate system called “Vault” to store personal files, photos and recordings as an AI-searchable personal operating system. Visser said he has been uploading transcripts and specialized source material into models to build his own “knowledge brain” and make retrieval more precise.
Visser added that he operates with no team beyond one assistant while using AI agents across much of his workflow, and said the business has still been growing quickly. He expects smaller and better open-source models to push edge computing higher on the agenda, with locally run models becoming an important part of future AI use.
Crypto remains in a waiting phase as AI trades absorb capital
On crypto, Pal said strong earnings expectations around AI hardware and infrastructure have made conditions harder for narrative-driven assets such as bitcoin and the broader crypto market. Capital, he said, is being pulled into more straightforward AI-linked equity trades. He suggested that a stronger return of funds into crypto may need two things: physical AI buildout hitting capacity limits, which would send capital toward digital assets that do not require the same expansion path, and tokenization unlocking liquidity in dormant assets such as real estate, private equity, venture capital and art.
Visser said he is watching Layer1 networks closely because decentralized identity and blockchains suited for AI-agent transactions could regain an edge when the market rotates. He also said both the AI infrastructure trade and crypto now need time to digest prior inflows. Pal put that reset window at 3 to 6 months, a period in which markets may need to absorb the capital already committed before the next leg develops.