How to Get Rich in an AI Bubble in 2026

Apps & Software, Artificial Intelligence, Business / By / / 10 minutes of reading

The AI boom looks, smells, and trades like a bubble. Trillions in projected capex, hundreds of billions already committed to infrastructure, and a stock market that now moves in lockstep with ten AI‑heavy names. As the Financial Times put it, “America is now one big bet on AI… AI better deliver for the US or its economy and markets will lose the one leg they are standing on.” That’s not nuance; that’s a warning siren.

Revolutionary AI: How Artificial Intelligence Remasters Higher Education

But here’s the twist: you don’t need AI to “deliver” to make a lot of money from it.

The way to win this cycle is to stop trying to be Sam Altman and start thinking like the guy who sold generators, air filters, and cleaning services to the data centers that Sam Altman rents. In other words: stop chasing the “app” upside and start owning the “picks and shovels” that everyone else quietly depends on.

This isn’t hypothetical. The data‑center and power buildout for AI is already so large that firms like Blackstone openly describe their strategy as investing in the “picks and shovels of AI,” pouring capital into data centers, grid upgrades, and power equipment rather than models or apps themselves. Engineering publications now talk matter‑of‑factly about a “data center gold rush” where the real winners are designers, contractors, and specialized service firms, not chatbots. Asset managers like BlackRock are telling clients to look at power grids and data‑center credit as their way into the AI frenzy.

So the question becomes: if you’re not going to build the models, what should you actually do?

Let’s walk the stack from the ground up.

The Six Tiers of the AI Economy (And Where the Real Money Hides)

Think of the AI economy as a six‑story building. Most founders are fighting over the penthouse. The smart money is quietly buying the land, the concrete, and the plumbing.

Tier 0: Energy – The New Land Grab

Before you get GPUs, you need gigawatts. AI data centers are already forcing utilities and cities to reconsider how much power they can deliver, and multiple analyses project data‑center power demand to more than double this decade as AI workloads ramp.

That means:

  • New generation capacity (gas, renewables, nuclear uprates).
  • New transmission lines and substations.
  • New power‑quality and backup systems between grid and rack.

You don’t have to own a utility. You can own the services around it. Firms like Hanley Energy built a business around critical power and energy management for hyperscale facilities—designing and maintaining the uninterruptible power, switchgear, and monitoring that keep AI racks from going dark. The market just validated that model: Jabil agreed to acquire Hanley Energy Group for roughly $725 million plus earn‑outs, explicitly to deepen its position in data‑center power infrastructure.

Actionable play: become the region’s go‑to emergency power specialist for data centers and grid assets. 24/7 field crews, niche expertise in the specific UPS gear and monitoring platforms hyperscalers use, contracts that look boring until you see the renewal rates.

Tier 1: Chip Fabs – Blue‑Collar Jobs in White‑Glove Rooms

AI’s arms race is really a silicon race. The fabs that make NVIDIA‑class chips are multi‑billion‑dollar facilities with brutal uptime requirements and insane cleanliness standards. They do not get maintained by PhDs; they get maintained by small armies of HVAC techs, filter changers, clean‑room specialists, and construction crews.

Look at Promera (formerly DataClean): they turned “keep mission‑critical environments free of particles and overheating” into a serious business by specializing in contamination control, pre‑commissioning cleaning, and airflow optimization for data centers and other critical spaces. Their work now includes helping AI‑dense facilities manage heat and particulate risk as rack densities spike.

That’s not “AI.” It’s filters, vacuums, HEPA certifications, and a pricing model where your technicians bill at $150/hour doing tasks the big contractors don’t want to burn their $200/hour engineers on.

Actionable play: start or acquire a niche industrial‑services firm and then reposition it around fabs and high‑density compute. Pick something painfully specific—clean‑room filter changes, ultra‑pure water systems, particle testing—and be world‑class at it.

Tier 2: Data Centers – Where AI Physically Lives

Data centers are the apartment buildings of the AI age, and there’s an outright construction boom. Billions in annual spend, 50%+ growth in some metrics, and multi‑year pipelines of sites tied to AI workloads.

Here, the upside is almost aggressively unsexy:

  • Specialized cleaning (for floors, racks, cable trays).
  • Thermal imaging inspections for fire and overheating risk.
  • Roof, plumbing, and insulation services with data‑center‑grade SLAs.
  • Cable tracing, labeling, and documentation to reduce downtime and human error.

Promera is again a template: they show how you evolve from “we clean” to “we manage contamination risk and thermal efficiency,” letting you move from janitorial pricing to resilience pricing. Blackstone’s own AI‑data‑center thesis is basically: own or finance the shells, the power, and the cooling; the tenants (AI models) will change but the rent checks won’t.

Actionable play: start as a service vendor, then stack software and sensors on top—thermal cameras, particulate monitors, predictive maintenance. Sell outcomes: “We reduce your risk of catastrophic downtime by X%,” not “We sweep your floors twice a week.”

Tier 3: Foundation Models – The Oil Rigs

This is where OpenAI, Anthropic, Google, xAI live. The economics here are brutal for outsiders:

  • Enormous fixed costs in GPUs and training runs.
  • Exponential scaling: new frontier models can cost billions to train.
  • Depressing unit economics: every new user can increase your marginal cost.

For a normal entrepreneur, this tier is almost purely customer or supplier territory. You sell fuel, food, or tools to the rig. You don’t build your own rig.

Actionable play: if you’re not sitting on billions and a GPU cluster, don’t try to compete here. Negotiate volume discounts, become a preferred integrator, resell capacity, or build payment rails or compliance wrappers around these models.

Tier 4: Plumbing – Orchestration, Frameworks, and Circular Money

This is the software infrastructure layer: APIs, observability, orchestration, frameworks. Think “Stripe/DataDog/MongoDB, but for AI.” The opportunity is real, but so is the weirdness: much of the early revenue is circular. Microsoft invests in OpenAI and is OpenAI’s biggest customer. Amazon invests in Anthropic and sells Anthropic its cloud. Money goes from balance sheet A to revenue line B to capex C and back to A as per Blackstone.

If you’re building in this tier, you’re in a knife fight with highly funded infra teams. That’s venture‑scale risk, not “durable cash‑flowing business” risk.

Actionable play: only go here if you have real infra chops and access to capital. For most people, the smarter move is to use these tools to modernize old‑school businesses (HVAC, utilities, industrial services) rather than try to compete with them.

Tier 5: AI‑Native Apps – The Noisiest Tier, the Weakest Moat

This is the visible layer: chatbots, AI CRMs, AI content tools, co‑pilots for every job. The barrier to entry is so low that anyone with a weekend and Replit can ship a prototype. That’s the problem.

The actual challenges:

  • Business models: customers don’t want another tool; they want a clearly quantified cost center to vanish.
  • Scaling: the distance from hacky prototype to reliable, secure, integrated product is huge.
  • Talent risk: your best engineer is your moat, and Meta or another hyperscaler can make them a $2M offer.

There are exceptions. Co‑Counsel, part of Thomson Reuters’ legal AI push, works because it’s inserted directly into existing legal workflows, research, document review, drafting inside a corpus lawyers already trust (Westlaw). It’s not “Ask my AI anything,” it’s “Cut this seven‑hour review to three, inside the tools you already pay for”.

Actionable play: if you insist on building at this layer, pick one expensive workflow in one vertical, measure it, then halve it. And build governance around your technical co‑founder like your life depends on it, because it does.

This Is a Bubble. That’s the Opportunity.

Look at the macro picture:

  • Hundreds of billions already spent or committed to AI data‑center capex.
  • A huge share of recent S&P 500 gains concentrated in a handful of AI‑exposed mega‑caps.
  • Energy and power utilities seeing a wave of M&A and new project finance explicitly tied to AI loads.

That’s bubble‑ish. But unlike tulips, AI is already wired into real productivity in customer support, code, design, logistics, and more. For example, Australian retailer Temple & Webster used AI to rewrite and maintain product descriptions across more than 200,000 pages and now uses GPT‑based agents to handle a chunk of pre‑sale live chats, cutting both time and cost.

So yes, valuations can and probably will compress. But the power plants, transformers, racks, filters, and technicians do not vanish when multiples do. That’s what you want to own.

Five Practical Ways to Profit Without Building a Model

1. Use AI as a Margin Weapon in Existing Businesses

Start with the most boring question in capitalism: “Where is payroll doing the same thing over and over?”

In a furniture retailer, that was customer service emails and product copy. In your business, it might be:

  • Quoting and proposal generation.
  • Dispatch and scheduling for field crews.
  • First‑pass QA checks on reports or inspections.
  • SOP generation and update.

You don’t need to be an AI company. You need to be the HVAC contractor, cleaning firm, or fab‑maintenance shop with 20–30% higher margins because AI eats half your admin.

2. “Buy Boring, Serve the Boom”

This is the core thesis. Instead of founding yet another AI SaaS, you:

  • Buy a small but healthy local contractor—electrical, HVAC, industrial cleaning, testing, or inspection.
  • Layer in AI to streamline back‑office and routing.
  • Retarget the company’s pitch and certifications toward data centers, fabs, or grid assets.

Blackstone’s institutional version of this is buying up data‑center campuses and power assets. Your version is buying the five‑truck company that already cleans “critical environments” and teaching it to speak AI data‑center language.

3. Be a Workflow Fixer, Not an Idea Guy

The worst sentence in tech right now is: “I have an idea for an AI app that could be used by anyone.”

The best sentence is: “I can cut your [very specific line item] in half within 90 days.”

Co‑Counsel works because it said to lawyers: “This tool plugs into how you already draft, review, and research—and then doubles your output,” and it’s being built by someone with distribution into law firms.

You don’t have to be in law:

  • For data centers: slash the time to complete safety audits.
  • For utilities: automate the triage of maintenance tickets and outage reports.
  • For contractors: auto‑generate compliant documentation from field photos and voice notes.

You don’t sell AI. You sell a fixed, measurable outcome.

4. Move Now, While Everyone Else Is Still “Exploring AI”

McKinsey and others keep finding the same thing: a majority of companies are still in early or experimental stages with AI, and adoption is uneven across sectors. Academic work from Harvard and Stanford finds that naive use of AI can actually slow workers down if they’re spending more time fixing outputs than doing the work themselves.

That’s a feature, not a bug, if you move first. It means:

  • You can bake AI literacy into your hiring and training long before it becomes table stakes.
  • You can build playbooks for your vertical—“How we do dispatch,” “How we do documentation,” “How we do compliance”—that are hard for slower incumbents to copy.

The bottleneck is not tokens; it’s human creativity and operational discipline.

5. Think in Agents, Not Prompts

The next wave is “agentic” AI: systems that don’t just answer questions but pursue goals—scheduling jobs, pulling data from multiple systems, initiating follow‑ups. You will see a lot of pseudo‑autonomous “agents” burn money doing glorified task‑runner demos.

The way to use this tech is ruthlessly practical:

  • An agent that continuously monitors data‑center thermal and particulate data and opens a work order when thresholds are breached.
  • An agent that processes legal intake forms, drafts a first‑pass summary, and routes them to the right humans—Co‑Counsel‑style—but for your niche.
  • An agent that digests your field photos and sensor logs and proposes maintenance schedules, which your human leads approve or adjust.

You are not handing over judgment. You are compressing the distance between raw data and the decision in front of a human.

The Real Flex: Own What No One Finds Sexy

In every tech mania, the same pattern repeats:

  • The headlines follow the platforms and apps.
  • The quiet fortunes are made in the pipes, land, logistics, and tools.

Right now, AI is the story. But behind the story are very physical systems: transformers, chilled water loops, HEPA filters, concrete pads, diesel tanks, and grid interconnects. Behind those systems are service businesses. And behind those businesses are owners.

You don’t have to guess which model wins, or whether this S‑curve is over‑discounted or under‑discounted. You just have to make sure that, whatever happens to AI valuations, someone still needs you to keep the lights on, the racks cool, the rooms clean, and the workflows moving.

That’s the bet: not that AI wins, but that the world has already committed so much concrete, copper, and capital to AI that the “picks and shovels” will pay out for years—boom, bust, or both.

Own the boring stuff. That’s where the leverage is.