The S-1 drops in three days. The whisper is $1.35 billion. And every institutional desk in New York is asking the same question: is this the top tick of AI capex, or the first real signal that retail colocation is the next REIT wave?
Csquare, a mid-tier data center operator targeting AI workloads, is about to test the market's appetite for physical infrastructure in a high-rate, high-hype environment. The number is big enough to move sectors. The details are thin enough to trigger a cold sweat.
Let's cut through the noise. I've watched this cycle before—2017's ICO land grab, 2020's DeFi yield sprint, 2022's Terra collapse. Each time, the market falls in love with a narrative before checking the balance sheet. This time, the narrative is 'AI infrastructure scarcity.' The balance sheet? We haven't seen it yet.
Context: Why This IPO Matters Now
Retail colocation—renting physical space, power, and cooling for customer-owned servers—isn't sexy. It's the plumbing of the digital age. But AI changed the flow. A single rack of NVIDIA H100 GPUs can pull 30-50 kilowatts, compared to 5-10 kW for traditional servers. That power density requires specialized cooling, upgraded electrical feeds, and long-term power purchase agreements.
Csquare claims to specialize precisely in this 'AI-ready' colocation. Their IPO, rumored at $1.35 billion with a potential valuation north of $2.4 billion, is meant to fund expansion of high-density data centers in key markets like Northern Virginia, Chicago, and Silicon Valley.
The timing is brutal. Ten-year Treasury yields are hovering near 5%. Capital-intensive real estate plays are getting hammered. Yet AI demand forecasts—70%+ CAGR in data center power consumption through 2027—are pulling operators into the market. Csquare is the canary. If it flies, expect a wave of copycat filings from CyrusOne, Vantage, and others. If it crashes, the AI infrastructure narrative takes a direct hit.
Core: What We Actually Know (and What We Don't)
The only hard data point is the $1.35 billion target. Everything else is inference. Based on my work monitoring on-chain capital flows and institutional custody movements during the 2024 ETF front-run, I learned that market movers rarely telegraph their hand. But the ledger—in this case, the forthcoming S-1—will tell the truth.
Let's stress-test the implied numbers.
Capital allocation: Assume 50% of proceeds go to construction and 50% to GPU procurement or pre-leasing. With H100s at ~$25,000 per unit, that's roughly 27,000 GPUs worth of support capacity. In practice, Csquare won't buy chips—they lease the facility. But the power infrastructure to run 27,000 H100s requires approximately 200-300 megawatts of critical load. That's a $2-3 billion total investment, meaning the IPO covers only a fraction. The rest must come from debt or future cash flows.
Revenue model: Retail colocation typically charges per kilowatt per month, with pass-through for power costs. In a high-rate environment, customers push for long-term fixed rates. Operators hedge with PPAs. If Csquare locked in cheap power contracts two years ago, their margins look great. If they're buying spot power in 2025, the electricity line item will eat EBITDA alive.
My test: In 2020, during the DeFi yield farming sprint, I personally executed trades on Uniswap and Curve, logging every gas fee and slippage error. That discipline—empirical stress-testing before committing capital—applies here. I would need to see three things in the S-1: occupancy rate (preferably >80%), average contract duration (3-5 years minimum), and power cost pass-through terms. Without those, the $1.35B is a blind bet.
Competitive landscape: Equinix (market cap ~$70B) and Digital Realty (~$40B) dominate retail colocation. Csquare cannot compete on scale. Their edge must be niche: higher power density per rack (e.g., 50kW+ vs. Equinix's standard 30kW), faster deployment timelines, or locations with scarce grid capacity. The IPO's success hinges on whether institutional investors believe AI native workloads will demand a new breed of data center, not just repurposed legacy facilities.
Contrarian Angle: The Blind Spots Everyone Is Missing
1. The 'Retail Colocation' label is misleading. Most of the AI demand is going to hyperscale cloud providers (AWS, Azure, GCP) building their own custom data centers. Retail colocation captures the tail—medium-sized AI startups, hedge funds running proprietary models, and enterprises with data sovereignty requirements. That tail is real but lumpy. If one major tenant goes bankrupt (and in a bear market, many AI startups will), occupancy can drop 20% overnight.
2. Power availability is the real bottleneck, not capital. In Northern Virginia, the largest data center market, utilities are warning of multi-year delays for new grid connections. Csquare may raise $1.35B but cannot build for 18-24 months. The yield is sweet, but the exit is sharper if customers cancel during the delay.
3. The 'AI infrastructure' narrative glosses over obsolescence risk. Chips evolve fast. A facility optimized for H100 liquid cooling may need retrofitting for Blackwell, Rubin, or custom ASICs. CapEx for upgrades eats into the returns investors are underwriting today. Listen to the whispers, but trust the ledger—the S-1 will show maintenance capex assumptions. If they're low, run.
4. The hype is pricing in perfection. Comparable publicly traded data center REITs trade at 25-35x P/AFFO. If Csquare comes at 40x+ with a promise of 30% AFFO growth, it's pricing in a flawless execution curve. Any miss—rate hikes, chip delays, tenant churn—will compress that multiple viciously. Speed is the only currency that doesn't lie. The first quarterly report post-IPO will reveal the truth.
Takeaway: Three Signals to Watch
- The S-1 filing. Look for EBITDA margin, occupancy rate, and top 5 customer concentration. Anything above 40% revenue from one client is a red flag.
- IPO pricing momentum. If the book is 10x oversubscribed, the market is screaming 'all-in' on AI infra. If it struggles to fill, the narrative is already fraying.
- Post-IPO deployment speed. A company that closes a land purchase and breaks ground within 6 months signals operational credibility. One that sits on cash signals a narrative in search of reality.
Chaos is just data waiting for a pattern. Csquare's IPO will provide the data point. I'll be running my own on-chain and fundamental checks. You should too.