Josh Vehovic

Building Flowstate Compute: power, sites & GPU compute for data centers and advanced manufacturing. Former CRE broker. Writing Buying Time.

[Josh Vehovic @JoshVehovic]

I spent 10 years in commercial real estate. Now I’m focused on a new kind of real estate: land with power.

Data centers and new factories need enormous amounts of power, and the grid can take years to deliver. That’s changing which sites matter and how deals get done.

Flowstate Compute is my early-stage advisory: powered sites, bridge power, and GPU access through our provider network, helping investors, tenants, landlords and lenders get projects built. Learning in public here.

[Josh Vehovic @JoshVehovic]

The bridge power toolkit, simply: big turbines (most efficient, slowest to get), gas engines and small turbines (fast, flexible), fuel cells (fastest to permit, priciest). Most projects mix them.

[Josh Vehovic @JoshVehovic]

The biggest gas turbine makers are taking orders for 2031 delivery. If your project needs power in 2027, you need a different plan.

[Josh Vehovic @JoshVehovic]

In 2017 I walked away from a decade in commercial real estate. I'd done well, but I was trading my time for it, and I wanted my time back.

That decision sits at the heart of the book I'm writing, Buying Time. Nine years later, I'm back in real estate. Just not the way I left it.

This time I'm working for myself, on a problem I find fascinating: land with power, for data centers and new factories. Same skills, different terms.

Buying your time back doesn't always mean leaving something forever. Sometimes it means coming back on your own terms.

[Josh Vehovic @JoshVehovic]

Question for anyone building 100 MW+: what's actually harder to get right now, power, land, capital or equipment?

[Josh Vehovic @JoshVehovic]

In a lot of US markets, a big new project can wait 5 to 7 years to connect to the grid. Nobody building today can wait that long. That one fact explains most of what's happening in energy right now.

[Josh Vehovic @JoshVehovic]

The AI buildout isn't waiting on chips anymore. It's waiting on power.

[Josh Vehovic @JoshVehovic]

@DavidDTawil Wtf is he talking about

[Josh Vehovic @JoshVehovic]

End of October. Two months ago most of what I post about now would have been unfamiliar. Grateful to everyone who's taken the time to explain things along the way.

[Josh Vehovic @JoshVehovic]

Heading toward year-end, the constraint in AI infrastructure is the same one it's been all year. Not chips, not money. Power. The companies that already have energized sites have the most options going into 2027.

[Josh Vehovic @JoshVehovic]

The simplest difference between a GPU and a Bitcoin mining machine is range. A mining machine does one calculation, over and over. A GPU can train AI models, run them, render video and more. That's why GPUs hold their value better.

[Josh Vehovic @JoshVehovic]

Owning the land versus leasing it is an old real estate question. Data center companies are landing on the same answer. Several have bought the land under their sites outright, because the land and the power connection are the long-term asset.

[Josh Vehovic @JoshVehovic]

Battery storage is becoming part of the data center conversation. Large batteries on site can smooth out power, cover short gaps, and help pair facilities with renewables that don't generate around the clock.

[Josh Vehovic @JoshVehovic]

Data center developers are paying attention to latitude. Cooler climates mean less energy spent on cooling, and that adds up at scale. It's one reason Canada and the northern US keep coming up in AI infrastructure news.

[Josh Vehovic @JoshVehovic]

@DukeHerndon I worked at Abercrombie in a midwestern city and same

[Josh Vehovic @JoshVehovic]

Flowstate Field Notes

Week nine.

Theme: power keeps coming up.

Lesson: whatever part of this business you start in, you end up talking about electricity.

Next week: closing out the month and looking at what's changed.

[Josh Vehovic @JoshVehovic]

Quiet realization this week: the AI boom and the energy business are becoming the same story. Nearly every question about compute eventually turns into a question about power.

[Josh Vehovic @JoshVehovic]

Bitcoin miners had one advantage AI doesn't: flexibility. Mining machines can shut off almost instantly when the grid is strained. AI customers expect their compute to stay on. That changes how a converted site has to be run.

[Josh Vehovic @JoshVehovic]

A GPU that isn't running is an expensive paperweight. That's why operators obsess over utilization, the share of time the hardware is doing paid work. It's often the difference between a profitable GPU business and a struggling one.

[Josh Vehovic @JoshVehovic]

Buildings get valued on income. Investors are starting to value data center companies on something similar: megawatts of power that are contracted and energized. One firm found companies with signed AI leases trading at much higher multiples than those without.