Goldman’s Air-Conditioning Play: The Physical Bottleneck Crypto Can’t Solve

Goldman Sachs upgraded a company that installs air conditioning. Comfort Systems USA. Target price: $2,159. Reason: the AI infrastructure boom. This is not a crypto play. It is a signal. The front-runner didn’t anticipate the bottleneck in construction. The market fixated on GPU supply, then power transformers, then cooling systems. Now it’s the entire physical plant. A building services firm with a $10 billion market cap is suddenly a direct beneficiary of AI. Crypto faces the same constraint but lacks the institutional trust to scale. Context matters. Comfort Systems USA provides mechanical, electrical, and plumbing services for large commercial buildings. Its core competency is designing and installing HVAC and power distribution for high-density data centers. In 2025, a single hyperscaler data center can consume 500 megawatts. That requires dozens of specialized contractors. Comfort Systems is one of the few with the capacity to handle these projects. Goldman’s upgrade reflects a recognition that AI’s growth is now limited by physical construction capacity, not just silicon. Compare this to crypto. The industry has spent years building layer-2 solutions, sharding, and rollups to solve scaling. But the real scaling problem is physical: where will the energy come from? Who will build the facilities? The same labor and materials are now being consumed by AI. Crypto miners already face power rationing in some regions. The narrative of "decentralized physical infrastructure networks" (DePIN) attempts to solve this by distributing nodes, but coordination costs remain high. A single AI data center can be built in 18 months with $5 billion. A DePIN network of thousands of nodes requires years of incentive design and unmatched hardware logistics. Based on my 2021 audit of Axie Infinity, I saw how revenue models depended on perpetual user inflows. That was a Ponzi. The AI infrastructure boom is not a Ponzi—it has a real end customer: corporate AI workloads. But it creates a parallel fragility. Every construction worker, every transformer, every ton of copper that goes into an AI data center is diverted from other uses, including crypto mining. The market is pricing Comfort Systems USA as a scarce resource. Crypto should see the same signal. Now, the core teardown. First, incentive structure: AI infrastructure spending is driven by cloud providers who can monetize compute immediately. Crypto mining requires token price appreciation to remain profitable. When AI companies can pay $10 per megawatt-hour more for power, miners get shut out. Second, systemic fragility: the AI data center boom is concentrated in a few hands—Amazon, Google, Microsoft. If their AI revenue disappoints, capital expenditure will drop, and Comfort Systems’ stock will fall. Crypto’s mining demand is also concentrated, but in publicly traded miners who have weaker balance sheets. Third, regulatory alignment: the SEC is comfortable with income-producing assets like data centers. Crypto mining is still treated as an environmental risk. Goldman’s report can be read as a regulatory signal: physical assets are safer than digital tokens. A bug is just a feature that hasn’t been exploited yet. The bug in the AI construction boom is that it centralizes compute in locations with cheap power and lax zoning. The feature for crypto is that decentralized compute networks become the only viable alternative for censorship-resistant applications. But the market is not pricing that. It is pricing the opposite: centralization works better for speed. Contrarian angle: what did the bulls get right? They saw that AI needs real infrastructure. They bought the ticker that builds it. They did not fall for vaporware. The bull case for Comfort Systems is that its backlog is visible and its contracts are long-term. The same logic applies to crypto mining infrastructure companies like Bit Digital or Hut 8, but those have additional token price risk. The bulls might argue that crypto can ride the same wave by colocating mining with AI compute. But that requires technical compromises. AI chips (GPUs) and mining chips (ASICs) have different power and cooling requirements. The co-location thesis is weak. Takeaway: The next time you hear a crypto project touting a new layer-2 or a DePIN token, ask who will pour the concrete. AI is consuming the physical resources crypto needs. Until the industry acknowledges that the real bottleneck is not code but construction, investors should look at firms like Comfort Systems USA instead. They are the picks and shovels of the next decade. Crypto’s picks and shovels are still in the garage. Based on my 2017 EOS audit, I learned that community trust can hide code flaws. Today, AI trust hides physical fragility. The front-runner didn’t see the bottleneck. Now it’s visible. Code is law, but physics is higher.