Listen to the audio version of this article (generated by AI).

The lessons from the dot-com period have become even more relevant today for savvy investors.

Choosing the next unbeatable website turned out to be a gamble, but investing in fundamental infrastructure proved to be a more pragmatic approach.

Amazon ( AMZN ) emerged successfully while others, like Pets.com, vanished. AOL enjoyed its moment in the limelight but eventually faded. Numerous tech ventures evaporated capital at alarming rates, leaving backers with empty pockets. Yet Cisco ( CSCO ) thrived amidst the chaos, securing profits by becoming the backbone of internet traffic through its routers and switches. Its stock skyrocketed about 3,400% in just five years.

Cisco's strategy hinged on providing the essential infrastructure to keep the internet functional, rather than choosing the right digital platform.

A similar transition is currently unfolding in the AI sector, albeit with a critical difference.

Investors have recognized the necessity of infrastructure for AI applications, but they often underestimate the sheer scale of resources that AI requires when transitioning from simple tasks to more complex functions.

This shift marks the onset of an unprecedented phase in AI's evolution, which I’ve termed the Invisible AI Tax.

From Chatbots to Agents: Why the Infrastructure Bill Just Got 20x Bigger

The distinction between chatbots and AI agents demonstrates a significant divergence in required infrastructure. A chatbot merely addresses user prompts, while an AI agent actively seeks and achieves goals.

A standard chatbot might only process a few hundred tokens — the building blocks of language that it interprets to generate a response. The interaction is brief; you pose a question, and the bot replies.

Conversely, when tasked with, “Create a marketing strategy,” an AI agent engages in extensive work. It investigates competitors, aggregates internal data, drafts advertising campaigns, runs tests, collaborates with other agents, refines plans, and reports on progress until the objective is met.

What starts as a few hundred tokens can swell to tens of thousands as the agent goes through planning, execution, validation, communication with external risks, and iteration. This escalation in complexity is something many investors haven't fully grasped yet.

Data indicates that AI agents can require 20 to 30 times more physical infrastructure per task than a standard chatbot interaction.

This isn't just a marginal increase; the difference is staggering.

AI demands greater computing power, more memory, expanded network capabilities, additional cooling measures, heightened energy usage, and increased data center capacity.

The interesting part? This isn’t a hypothetical situation. Over half of leading corporations already have AI agents operating in their systems, and this adoption trend is set to ramp up swiftly.

As such, the AI landscape is moving from a phase of experimentation to one characterized by ongoing infrastructure use.

(If infrastructure is a core area of your interest, let me highlight a compelling narrative that the market may not be seeing clearly — while it isn’t related to AI, the underlying investment logic remains consistent. More on that here.)

The key question becomes: where will this additional demand materialize?

The Six Tollbooths Every Agentic AI Workload Must Pay

Imagine the AI economy as a major highway.

Every model inquiry and agentic undertaking must navigate through essential physical infrastructure, crossing six crucial tollbooths: compute, memory, networking, thermal management, power, and real estate.

We've previously examined elements of this system — including the shift to custom AI silicon, the challenges of data center networking, and the physical limits concerning power and cooling. This piece focuses on the next facet of the equation: the need for infrastructure, particularly as demands grow.

Compute is the most apparent toll. AI models are contingent on advanced chips — GPUs, specialized accelerators, and inference chips crafted for handling large volumes of simultaneous processing. Nvidia remains central to this sector, but custom chip makers are becoming increasingly relevant as major tech firms develop cost-effective, tailored solutions for their specific AI requirements.

Memory represents the second toll. AI agents need context management; they must track their past actions, current tasks, and what lies ahead. As projects grow in duration and complexity, the demand for high-performance memory intensifies, necessitating larger context windows.

The networking toll might be the least recognized, yet it's vital. AI agents must exchange data with databases, tools, APIs, external services, and among themselves. This traffic needs to transfer at impressive speeds between chips, racks, servers, and data centers, making switches, interconnects, cables, and networking silicon increasingly essential as agentic AI scales.

The next toll is thermal management. As AI server clusters run denser systems, they generate substantial heat. Given that agentic tasks run longer than basic chatbot queries, thermal output only escalates. As a result, liquid cooling systems, coolant distribution networks, and precision thermal setups have become indispensable for maintaining operation.

Power is the fifth toll. AI agents operate continuously, working across numerous enterprises, executing tasks around the clock. This ongoing operation demands grid enhancements, onsite energy solutions, long-term power agreements, and a steady energy source.

Finally, we have real estate. Every piece of infrastructure, from servers to cooling systems, must reside within a specific space. This need for specialized data centers with access to land, power, cooling, and fiber optics becomes increasingly prominent.

A chatbot taps all these tollbooths, but an agent utilizes them aggressively.

This is the essence of the Invisible AI Tax; as the agentic economy broadens, every operation will contribute to this toll.

The Numbers Are Already Showing Up In Earnings

The tollbooths stand poised to reap substantial rewards.

At Google Cloud Next, CEO Sundar Pichai revealed that their AI models currently process over 16 billion tokens every minute, representing a significant 60% rise since the last quarter. Additionally, hundreds of Google clients have each consumed more than one trillion tokens in the last year alone.

One trillion tokens each.

Nvidia’s CEO, Jensen Huang, has reported that the required inference computing power is now 100 times greater than previously anticipated — and we're still in the early stages.

Expenditures on AI infrastructure by major tech firms are skyrocketing. Demand for AI-related memory is surging. Networking goals are continuously increasing. Cooling demands are experiencing backlogs. Energy corporations are securing long-term contracts with cloud giants. Data center owners are leasing out space as quickly as they can expand.

The companies who operate these vital infrastructure components aren't waiting for demand to materialize; they are back-to-back reporting rising earnings.

And we’re only beginning to see the effects of the agentic demand multiplier.

What This Means for Agentic AI Stocks

Winners and losers will emerge from the AI model competition.

OpenAI. Google. Anthropic. Meta. xAI. Competitors from China. Various open-source efforts. Proprietary systems. Some will come out on top; others will falter.

Identifying the ultimate victor is challenging, and even the most astute technology investors have made errors in this arena.

Yet, irrespective of which model prevails, the demand for infrastructure remains unchanged.

Every model will need computing power. Every agent will require memory. Each workflow will need networking resources. Every server will demand cooling solutions. Every data center will necessitate reliable power. Every server will require proper housing.

This makes the Invisible AI Tax so significant.

The infrastructure firms best positioned to benefit will gain revenue as the demand for AI rises.

As agents multiply, so too does the necessity for infrastructure.

The initial phase in this transformation revolved around validating AI's capabilities. The subsequent phase will focus on financing its extensive use.

That’s precisely where the tollbooth companies find their place.

The Real Risks (This Isn’t a Free Lunch)

However, this scenario does not imply that these stocks are free from risk.

Many stocks are currently trading at elevated valuations. Interruptions in the capital expenditures of major tech firms would negatively affect the sector overall. Some companies face significant customer concentration. Others, particularly in next-gen power, cooling, and optical networking, are still navigating considerable operational risks.

These challenges are primarily about timing and sizing; they don't undermine the fundamental argument.

The transition from chatbots to agents amplifies infrastructure consumption on a per-task basis. An exclusive group of companies stands to benefit financially as this consumption escalates.

The Infrastructure Always Gets Paid

Many investors remain fixated on the AI race, attempting to identify the leading service. Unfortunately, that's not the right approach.

The race's winning entity still has to navigate the infrastructure road. And that road comes with a toll.

Whichever company collects this toll will reap the benefits, regardless of which AI model secures first place.

I've dedicated years to uncovering companies positioned at the nexus of undeniable, inescapable trends — akin to the Ciscos of past times, rather than the Pets.com failures.

At this moment, I'm particularly intrigued by one investment opportunity that dominates my watchlist.

This opportunity isn't directly connected to AI infrastructure, but the principle is the same: locate the tollbooth, not the traffic.

The global money market, valued at a staggering $480 trillion, is undergoing a thorough transformation as its foundational systems are rebuilt. I believe Elon Musk is spearheading this change, and the opportunity to position yourself early is swiftly closing.

I've prepared a detailed briefing on the current developments, outlining which stocks I believe will thrive in this environment, and why this could represent the most significant wealth-building narrative of the decade.

Here’s everything I know.