The crypto and AI communities are cheering TPG's $3 billion deal for Netrality Data Centers as a sign of institutional confidence. But as someone who spent 2017 debunking Zilliqa's scalability claims by tracing transaction finality edge-cases, I've learned that big money often masks structural flaws. This acquisition is not a bet on AI innovation—it's a high-leverage real estate play in a sector where the underlying assets may not deliver what the narrative promises.
Context: The AI Infrastructure Land Grab Netrality Data Centers operates over 10 interconnection hubs in secondary U.S. markets like St. Louis, Kansas City, and Philadelphia—cities far from the hyperscale clusters of Northern Virginia or Silicon Valley. These are carrier hotels, not purpose-built AI factories. TPG, a private equity giant with a history of digital infrastructure investments (DataBank, Cirion), is acquiring them to ride the AI compute demand wave. The $3 billion valuation implies an enterprise value per MW of roughly $800-$1,000 million, in line with recent deals like KKR's CyrusOne purchase. But this comparison ignores a critical variable: AI workloads require radically different infrastructure than traditional colocation.
Core: A Technical Teardown of the Value Proposition Let's start with power density. Modern AI training racks—NVIDIA DGX H100 clusters, for instance—draw 30-40 kW per rack, often exceeding 60 kW with liquid cooling. Traditional colocation data centers, especially those built a decade ago, typically max out at 10-15 kW per rack. Netrality's facilities, designed for network peering and low-density server hosting, would require extensive retrofitting: new electrical substations, liquid cooling loops, higher-capacity switchgear, and potentially even grid capacity upgrades. Based on my dissection of similar infrastructure deals during the MakerDAO collateral audit in 2020, I know that 'compliance-first' often means 'maintenance-first'—the hidden CAPEX can devour projected returns.
Examine the physical reality: Netrality's total IT load capacity is estimated at 300-400 MW based on the purchase price. To support even 200 MW of AI-grade compute (say 50,000 H100 GPUs at 700W each), you'd need to double the cooling infrastructure and triple the power distribution. The cost: $10-$15 million per typical data center hall for a full AI retrofit. Assume 10 halls across Netrality's portfolio—that's $100-$150 million in unexpected CAPEX, not accounted for in the $3 billion price tag. Valuation without CAPEX inclusion is a clock waiting to stop.
Then there's latency and connectivity. Hyperscalers like AWS and Azure have spent years building out fiber rings and peering points in Tier 1 markets. Netrality's secondary locations offer cheaper power (15-20% lower than Ashburn), but that advantage is offset by higher latency to AI model endpoints—critical for inference but tolerable for training. However, most AI training today happens on large clusters co-located with cloud providers or specialized GPU clouds like CoreWeave. Netrality would need to either build direct interconnects to these clouds or attract a new breed of AI tenants willing to accept higher latency for lower cost. The bull case hinges on a tenant mix that doesn't exist yet.
Complexity hides risk. The deal's structure compounds these technical problems. TPG is likely using significant leverage—60% LTV is common for such infrastructure deals—meaning equity at risk is only $1.2 billion. But with floating-rate financing, rising interest rates can turn a 15% IRR into a 5% return. During the Terra/Luna collapse forensics in 2022, I modeled how debt-laden structures amplify systemic fragility: a 2% downturn in rental income (from power cost spikes or tenant churn) can trigger margin calls. The same dynamics apply here, just in real estate instead of stablecoins.
Contrarian: What the Bulls Got Right Admittedly, the strategy is not entirely without merit. Secondary markets offer genuine power cost advantages—up to 30% lower per kWh compared to Loudoun County. For AI companies running large-scale pre-training runs that are cost-sensitive, this matters. Moreover, Netrality's carrier hotel model, with multiple fiber providers already on-site, reduces the cost of last-mile connectivity for tenants. If TPG can secure anchor tenants with long-term leases (5-10 years), the cash flow becomes predictable. The bulls argue that AI demand will grow sufficiently to fill these facilities, and that the valuation multiple (15-18x EBITDA) is reasonable given the sector tailwinds.
They have a point. But I've seen this movie before—during the DeFi Summer of 2020, when MakerDAO's KNC collateral threshold looked attractive until a Chainlink oracle exploit exposed the fragility. The bull case relies on perfect execution: perfect retrofitting, perfect tenant acquisition, perfect power pricing. Execution risk is the highest form of risk in infrastructure, and it's rarely priced in.
Takeaway: Accountability or Anecdote? This deal will be a litmus test for whether AI infrastructure is a durable asset class or a speculative real estate bubble propped up by zero-interest rate hangover. If TPG succeeds, it will validate the thesis that secondary markets can host cost-effective AI compute, accelerating the decentralization of training capacity. If it fails, the $3 billion will become a cautionary tale of how hype can distort asset prices faster than any smart contract exploit.
Trust no one, verify everything. I'll be watching the post-deal footprint: utility contract renewals, tenant announcements, and any hints of debt restructuring. Until then, I'm treating this acquisition as a high-risk bet on a technical transition that has yet to occur. The code—or in this case, the balance sheet—does not lie. But people do.
