Everyone's screaming about AI stocks. Mike Novogratz is buying dirt. And power lines. And a piece of the Texas grid that froze solid in 2021.
He's not alone. Galaxy Digital's founder just announced a pivot into Texas AI infrastructure—a move that on the surface reads as savvy diversification. Dig deeper, and it starts looking like a liquidity trap masked as a macro hedge.
I've been mapping capital flows since 2017, when I built a Python script to track Ethereum gas fees and ICO vesting patterns. That script told me that 80% of those projects were doomed before they even launched—not because of bad tech, but because their liquidity structures were fundamentally broken. Fast forward to 2026, and I see the same pattern here: a big bet on physical assets that looks solid on paper, but carries the exact same maturity mismatch risks that blew up Terra, Celsius, and Three Arrows Capital.
Context: The Texas Mirage
Texas is the undisputed king of cheap energy. Industrial electricity rates hover around $0.05–0.08 per kWh—half of California's. No corporate income tax. Loose zoning laws. And a grid (ERCOT) that operates like a libertarian's fever dream: independent, deregulated, and prone to catastrophic failure when the weather turns.
It's no wonder that every hyperscaler—Meta, Google, OpenAI—has either built or announced massive data center campuses in the Lone Star State. The playbook is simple: secure land, lock in a 10-year power purchase agreement (PPA), stack NVIDIA H100s or B200s, and lease the compute to AI startups or cloud providers at a fat margin. It's the digital equivalent of selling shovels during a gold rush.
Novogratz, a former macro trader and crypto enthusiast, is now buying that shovel. But instead of buying AI stocks (NVIDIA, Microsoft, etc.), he's investing directly in the underlying infrastructure. A contrarian move? Or a desperate grab for yield outside the crypto cycle?
Core: The Macro Watcher's Lens
Let me be clear: I respect Novogratz's track record. He called the crypto cycle better than most. But I spent 400 hours in 2017 analyzing ICO liquidity fragmentation, and I spent three months in 2020 reverse-engineering Curve Finance's liquidity pools. That experience taught me one thing: hype follows liquidity, not the other way around. And right now, liquidity is rotating out of digital assets into physical ones—but the rotation is far from clean.
Here's the hard truth about Texas AI infrastructure:
First, it's capital intensive. A single 100MW AI data center—enough to house 50,000 H100 GPUs—costs between $500 million and $1 billion to build. Galaxy Digital has roughly $2 billion in assets under management (post-crypto rally). One project could consume half their balance sheet. That's not diversification; that's concentration risk masquerading as scale.
Second, the operating model is unproven for a crypto-native firm. Galaxy is a crypto asset manager and investment bank. They don't run data centers. They don't manage GPU fleets. They don't negotiate with ERCOT for demand response programs. Every data point I've seen—from my own analysis of mining farms pivoting to AI, to public filings from Hut8 and Hive—shows that the operational learning curve is steep. Margins compress when you have to hire 50 new engineers just to keep the lights on.
Third, the demand side is a Schrödinger's cat. AI compute demand is growing at 50%+ per year. That's real. But data center supply is growing even faster, driven by the same Texas land rush. CoreWeave, Lambda, Crusoe Energy—they're all building. When supply overshoots demand—and it will, because infrastructure cycles always overshoot—lease rates collapse. Novogratz may be building just as the first wave of AI commoditization hits.
I remember the DeFi summer of 2020. Everyone thought liquidity mining was free money. I ran the numbers on Curve's delayed rebalancing and realized the yield was an arbitrage trap for the smart money. The same dynamic is playing out here: everyone sees AI as an endless demand curve. They forget that data centers are long-duration assets with very high fixed costs. If AI demand plateaus—or even just grows slower than expected—those 50,000 GPUs become stranded assets.
The Liquidity Trap
Let's talk about the elephant in the room: where is the money coming from?
Galaxy Digital's core business is crypto trading, lending, and investing. Their liquidity is inherently tied to the Bitcoin and Ethereum cycles. When crypto is up, they have capital to deploy. When crypto is down, they face redemption pressures. Novogratz's AI infrastructure bet is a classic counter-cyclical play: use bull market profits to buy hard assets that generate steady cash flow during bear markets.
"Liquidity doesn't care about your grand vision. It only flows where the yield is safe."
That's my signature line, and it applies perfectly here. A data center's cash yield is only "safe" if the leases are signed, the power is cheap, and the GPUs stay in demand. None of those are guaranteed. The 2022 crypto winter taught us that even centralized lenders with "conservative" practices (like Celsius) failed when the liquidity stopped flowing. Novogratz's infrastructure is more like a bond than a stock—long tail, illiquid, and carrying refinancing risk if Galaxy's own liquidity dries up.
Another rug? No, just a liquidity trap.
I've seen this movie before. In 2022, when LUNA collapsed, the narrative was that it was a "tech failure." I published a 20-page macro thesis arguing it was a liquidity crisis—a mismatch between short-term liabilities (UST redemptions) and long-term assets (Bitcoin reserves). The same structure applies here: Galaxy is taking short-cycle capital (crypto profits) and locking it into long-cycle infrastructure. If the crypto market tanks again, they'll be forced to sell those assets at a discount, or worse, halt new investment.
Contrarian: The Decoupling Thesis
But wait. There's a world where this works brilliantly.
The contrarian angle—the one I'd debate with fellow macro watchers—is that Novogratz is actually ahead of the curve. He's not betting on AI stocks (which trade at 50x earnings and rely on hype). He's betting on the physical foundation of the AI economy. That's a commodity-style play with real barriers to entry: land permits, power contracts, GPU supply chains. If AI follows the path of the internet, the infrastructure providers (Equinix, Digital Realty) have outperformed the early dot-com stocks on a risk-adjusted basis.
Moreover, Texas is uniquely positioned. The state's independent grid might be unstable, but it offers something no other state can: the ability to negotiate direct power deals with wind, solar, and natural gas plants. A well-structured PPA can lock in rates for 15 years. Galaxy could even partner with a nuclear SMR developer (Texas has several pilot projects) to generate baseload power without carbon taxes.
"Macro doesn't rotate without a cost." The cost here is time. Data centers take 2–4 years to build. Novogratz is buying an option on future AI demand. If he times it right—if AI compute demand continues to double every 12 months—his assets could triple in value. If not, he's left with a warehouse full of silicon that depreciates faster than a used car.
Takeaway: Position for the Rotation, but Watch the Liquidity
I'm not saying Novogratz is wrong. I'm saying the risk is mispriced by the market. The crypto community cheers "diversification" without understanding the illiquidity premium attached to steel and concrete. The AI community cheers new compute capacity without asking who's paying for it.
For my readers—the macro observers, the liquidity hunters—here's what to track:
- Power contracts: If Novogratz announces a fixed-rate PPA at $0.04/kWh or lower, that's a strong signal. If he's paying floating rates, run.
- GPU supply: He needs guaranteed access to H100s or B200s. If he's buying on the spot market, the margin math doesn't work.
- Lease commitments: Any pre-leasing to hyperscalers (e.g., AWS, Google) or AI lab credits is a strong validation. Without it, the project is speculative.
Is Novogratz building the next generation of digital infrastructure, or just another liquidity trap for yield-starved capital?
I'll answer in a year, when we see the first quarterly earnings from Galaxy's infrastructure arm. Until then, keep your liquidity close. And your skepticism closer.