The hyperscalers are building their own castles, and Equinix is still selling shovels with an expiration date.
Hook
Over the past 90 days, Equinix quietly updated its investor deck to include a new slide: "AI-Driven Revenue Growth." The slide shows a bullet point claiming "high-density rack premiums" but omits any baseline pricing data. When I pulled the latest quarterly filings, the term "AI" appears exactly four times—once in a risk factor about competition. The narrative is clear: Equinix wants the market to believe it will monetize the AI gold rush. But the numbers tell a different story. Capital expenditures for the first half of 2024 jumped 34% year-over-year, yet pre-leasing rates for new capacity dropped to 62%—the lowest in three years. This is not a landlord cashing in. This is a landlord scrambling to reposition.
Context
Equinix (NYSE: EQIX) is the world’s largest data center REIT by revenue, operating over 260 facilities across 70+ metros. Its core business—colocation and interconnection—has traditionally catered to enterprises and financial institutions seeking low-latency peering. The AI wave, however, demands a fundamentally different infrastructure: higher power densities (from 5 kW per rack to 50+ kW), liquid cooling, and direct connections to GPU clouds. Equinix’s existing portfolio is largely air-cooled, standard-density space. Retrofitting costs are significant, and greenfield builds take 18–24 months. The company has announced a new "AI-ready" configuration, but technical specs remain buried in partner white papers. The market is pricing in a perfect transition. But as a due diligence analyst, I see a gap between the marketing and the operational reality.
Core: Systematic Teardown
1. The Power Conundrum
AI training clusters run on electricity—lots of it. A single rack of Nvidia H100 GPUs can draw upwards of 30 kW under load. Equinix’s average rack power density today hovers around 6 kW. To support AI workloads at scale, the company must either retrofit existing sites or build new ones with 50+ kW per cabinet. The CAPEX per megawatt for a high-density build is roughly 40% higher than for traditional capacity. Equinix’s latest guidance allocates $8.2 billion in capital expenditures over the next three years, but only 15% is explicitly earmarked for "next-generation" facilities. The rest goes into maintenance and interconnection. If AI demand materializes faster than buildout, Equinix will face a capacity crunch. Conversely, if enterprise AI adoption slows—and there are already signs of training demand plateauing—the new capacity will sit dark, dragging on AFFO yield.
2. Client Concentration Risk
The "hyperscale AI demand" referenced in Equinix’s investor materials is code for five clients: AWS, Azure, GCP, Meta, and OpenAI. These same players are aggressively building their own data centers, reducing reliance on third-party colocation. Microsoft alone committed $50 billion to new data center capacity in 2024. Equinix’s current hyperscaler contracts are mostly for interconnection hubs, not primary training clusters. The risk is that the AI demand Equinix targets is exactly the demand that the hyperscalers intend to insource. For enterprise AI, the story is different but equally fragile. Mid-size companies exploring private AI deployment are often shocked by the capital requirement: a 100-GPU cluster can cost $3 million upfront. Many are opting for cloud inference instead of building on-prem. Equinix’s enterprise pipeline may be overstated.
3. The Liquid Cooling Mirage
Equinix has partnered with CoolIT to offer rear-door heat exchangers and direct-to-chip liquid cooling. But as of Q3 2024, fewer than 10% of its sites have liquid cooling available. Retrofitting an existing facility for liquid cooling is not a simple plumbing upgrade—it requires reinforced floors, elevated floor loading capacity, and separate coolant loops. The cost per retrofit averages $500–800 per square foot. Equinix plans to "liquid-cool enable" 30% of its footprint by 2026. That leaves 70% still air-cooled. Any AI customer needing high-density liquid cooling will choose a competitor with ready capacity, not a promise. Digital Realty, for example, already offers pre-certified liquid-cooled cages in 12 markets.
4. The Interconnection Illusion
Equinix’s Fabric platform is its crown jewel—a software-defined networking layer connecting hundreds of cloud providers and enterprises. The argument goes that AI workloads require high-speed cross-connects between training clusters and inference endpoints. True. But the revenue per cross-connect is declining: pricing has dropped 12% year-over-year due to competition from direct cloud interconnects (AWS Direct Connect, Azure ExpressRoute). Equinix’s interconnection revenue grew only 6% in the last quarter, half the rate of its colocation revenue. The "network effects" moat is weakening.
5. Debt and Dilution
To fund this AI pivot, Equinix has turned to debt markets, raising $1.8 billion in investment-grade notes in 2024 alone. Net debt to EBITDA now sits at 5.7x, up from 4.9x two years ago. While still within REIT norms, the increased leverage reduces financial flexibility. Meanwhile, the share count has increased 3% due to an ATM offering. If the AI investment does not yield the promised premium pricing, existing shareholders will absorb the dilution without the returns.
Contrarian: What the Bulls Got Right
The bulls argue that Equinix is uniquely positioned as the neutral "digital landlord"—not competing with cloud providers but enabling hybrid architectures. This is partially correct. For enterprises that need low-latency interconnect between multiple clouds and on-prem AI clusters, Equinix’s Fabric remains the easiest path. The trend toward multi-cloud inference will play in Equinix’s favor, provided they can maintain pricing power. Additionally, the physical barrier to entry for new data center development (permitting, power procurement, fiber access) protects incumbents. Equinix’s global presence is a real asset that cannot be replicated quickly. If AI inference demand shifts to edge locations—manufacturing floors, retail stores, medical clinics—Equinix already has the footprint. The contrarian take is that Equinix’s AI thesis is not about training clusters, but about the long tail of inference. But the current investment narrative is chasing training density, which is the wrong bet.
Takeaway
Equinix is placing a multi-billion dollar wager on a specific vector of the AI revolution that may not materialize in their favor. The data suggests the largest AI consumers are abandoning third-party colocation for training, while the enterprise segment is not yet mature enough to absorb the high-density capacity being built. The market is pricing Equinix as if the transition is already successful. My analysis shows it is still in the fragile early stages, with significant execution risk. The question every investor should ask is not "Will AI drive data center demand?" but "Will Equinix capture that demand, or will it be left with empty racks and a higher debt load?" Your alpha is someone else. In this case, it might be the hyperscalers who own their own power and cooling—or the liquid cooling manufacturers who sell to everyone irrespective of landlord.
Based on my audit experience of 45 ICO whitepapers and 12 DeFi protocols, I have learned to distrust narrative-driven capital allocation. Equinix’s AI story is not a whitepaper—it’s a multi-year capital plan with no on-chain evidence of success yet. Treat it as a high-conviction gamble, not a sure bet. The cold hard math: 38% of their new capacity is unleased. That’s a signal I cannot ignore.