Some AI companies get paid before the customer runs a single task. Others get paid only when the work is done.
Miami has a company for each of the four main models. Here is how AI companies make money, and what each model takes.
Key Takeaways
- AI companies make money four main ways: renting compute, renting space, selling software and doing the work.
- Neocloud customers sign contracts of two to five years and pay whether they use the capacity or not.
- Miami has an example of each model, from Optidata and Metrobloks to Comp AI and Maestro.
- Each model has a price of entry: capital, power and permits, trust, or licenses.
- The risks are GPU debt coming due by 2028, customers who become competitors, and local opposition.
How Do AI Companies Make Money?
AI companies make money by renting compute, renting space, selling software or doing the work for customers.
- Rent compute. Customers use GPU capacity by the hour or on a multi-year contract.
- Rent space. Tenants pay for floor space, power and cooling for their own equipment.
- Sell software. Customers buy an AI product that does a job, such as compliance work.
- Do the work. The company delivers finished work, such as a processed mortgage file.
A software company sells a tool. An AI company often sells the result.
Each model has a price of entry. Compute takes capital, space takes power and permits, software takes trust, and the work takes licenses.
Many AI companies mix models. Optidata, for example, says it delivers full-stack CPU and GPU infrastructure on one platform.
Miami has an example of each, which makes it a useful place to compare them.
The models also scale differently. One software product can serve many customers at once, while compute and space are sold in fixed units.
How Miami AI Companies Make Money
Neoclouds rent GPU compute, and their customers sign long contracts. A customer commits to fixed capacity for two to five years and pays whether it uses it or not.
Optidata, a Miami provider, lists Miami (US1) as the first region for its Neocloud platform. It offers dedicated NVIDIA GPU instances.
MAIN’s neocloud explainer covers the model in full.
Colocation is the rent-space model. A tenant brings its own equipment and pays for space, power and cooling.
Equinix runs a downtown Miami site, and Flexential markets AI and GPU colocation in Fort Lauderdale.
Metrobloks is building MIA-A1 near Sweetwater, with 16.8 megawatts of stated critical IT capacity. Iron Mountain is building MIA-1 in Westview, phasing up to 16 megawatts.
MAIN’s AI infrastructure explainer covers both builds.
Comp AI sells software that does the work. Its agents handle onboarding, policy generation, evidence collection and vendor assessments.
More than 1,000 businesses use the platform from its Aventura headquarters.
They span healthcare, financial services, higher education, professional services and software. MAIN’s compliance explainer tells the full story.
Maestro does the work itself. The South Florida mortgage startup built seven agents, one for each stage of a loan.
A mortgage file spends about 38 days getting to closing, and only about 11 hours is hands-on work. Maestro’s agents go after the waiting.
Founder David Rogove explained the model in a MAIN Exclusive conversation. MAIN’s agent explainer shows how those agents work.
The Data: Which AI Business Models Are Growing?
Neoclouds show the most measurable growth. MAIN’s neocloud explainer reports estimates of more than $25 billion in 2025, with forecasts approaching $180 billion by 2030.
Hyperscalers are among the buyers. Microsoft alone has committed more than $60 billion to neocloud partnerships, because it cannot build fast enough on its own.
The rent-space model is growing in Florida too. As of August 24, MAIN’s tracker counted more than 150 projects in the pipeline and about 110 operating data centers.
About 27 of the operating sites are in Miami-Dade. For the Miami-area funding picture, read MAIN’s Q3 2026 report.
Why AI Business Models Are Hard
Compute takes capital, and much of it is borrowed. Analysts point to tens of billions in GPU-backed loans coming due between 2026 and 2028.
Customers can turn into competitors. When Bloomberg reported in July that Meta was exploring its own cloud business, Nebius and CoreWeave shares briefly dropped around 15%.
Space takes power and permits. Metrobloks sited MIA-A1 about 200 feet from a 230-kilovolt substation, with dual-line redundancy.
Metrobloks targets a Q4 2026 opening, so MIA-A1 has yet to prove the space model in Miami.
Permits can shape the story too. Iron Mountain’s Westview site was classed as a telecommunications hub, so it was approved by right with no public hearing.
Residents nearby have raised concerns about noise, water use and whether the promised jobs will reach the community.
Opposition is the other barrier. Palm Beach County rejected Project Tango, a hyperscale proposal in Loxahatchee, 5-1 on July 15.
As of August 24, 19 Florida jurisdictions had paused or banned new projects. Heat raises the bar too, because some GPUs draw over 1,000 watts and racks can pass 100 kilowatts. Metrobloks designed MIA-A1 for rack densities of up to 150 kilowatts.
If Comp AI’s plan works, it grows from compliance vendor into a full security platform. That puts it against rivals built for the older, periodic model.
Software and the work take trust. Maestro logs every agent action with a timestamp, and a person can take back control at any point.
Regulated work also takes licenses. Maestro holds them in roughly 20 states for third-party mortgage processing, which lets its agents process files directly.
Renting, selling and doing the work all pay. Debt, power and trust decide who keeps the money.
Frequently Asked Questions
How do AI companies make money?
AI companies make money by renting compute, renting space, selling software or doing the work for customers. Miami examples are Optidata for compute, Metrobloks for space, Comp AI for software and Maestro for the work.
Which AI business model works best?
No single model wins. Each has a price of entry. Compute takes capital, space takes power and permits, software takes trust, and the work takes licenses. The best fit is the one your resources match.
What is the best business to start with AI?
Software and doing the work need the least physical infrastructure, while compute and space need capital and power. Miami founders are building in AI agents, compliance and infrastructure.
What are some good AI startup ideas?
Look at where Miami AI companies are building: AI agents for regulated work, compliance software, GPU cloud and data center space. MAIN’s company directory lists them, and founders raising can browse the investor directory.
What is the fastest-growing AI business?
Neoclouds. MAIN’s neocloud explainer reports estimates of more than $25 billion in 2025, with forecasts approaching $180 billion by 2030. The risks are debt due by 2028 and customers building their own capacity.
Which AI company is the most profitable?
Few AI companies publish profits, and MAIN does not quote figures it cannot verify. Long contracts, such as the multi-year neocloud deals above, are the clearest revenue signal.
Sources:
- What Is a Neocloud? Inside AI’s Fastest-Growing Infrastructure, MAIN
- Every Florida AI Data Center We’re Tracking, MAIN, August 24, 2026
- Metrobloks MIA-A1 Is Quietly Becoming Miami’s Next AI Data Center, MAIN, August 16, 2026
- AI Power Demands Force Data Center Cooling Overhaul, MAIN, August 7, 2026
- MAIN Exclusive: David Rogove’s Mortgage AI Agents Replace the Middlemen, MAIN, September 3, 2026
- Miami’s Comp AI Raises $34M to Expand Beyond Compliance, MAIN, September 18, 2026
- What Is an AI Agent? Miami Firms Show How, MAIN
- MAIN Data: Miami Startups Q3 2026, MAIN, September 24, 2026
- Optidata, company website