OpenAI Isn't A SaaS Company. It's a Utility

OpenAI has vast capex needs, which SaaS companies don't

Most investors, and most commentators, have misunderstood OpenAI. They see a cutting-edge AI company with SaaS-like recurring revenue, exponential user growth, and a sticky developer platform. In short, they think it’s the next Snowflake or Salesforce. That’s a mistake.

In reality, OpenAI increasingly resembles a capital-intensive infrastructure utility. The product may be tokens instead of kilowatt-hours, but the economics rhyme. And the more it scales, the more obvious this becomes.

SaaS on the Surface

At a glance, OpenAI looks like a textbook software play:

• It sells API access to developers, metered by usage.

• It offers a direct-to-consumer product (ChatGPT) with a monthly subscription.

• It plugs neatly into Microsoft’s enterprise sales machine via Azure.

So far, so SaaS.

The private markets are treating it accordingly. With a reported $90 billion valuation, and a revenue run-rate in the low billions, OpenAI is being priced like a high-growth, high-margin software company. Investors are anchoring to metrics and multiples borrowed from Salesforce, Datadog, and Stripe.

But dig one layer deeper, and the comparison falls apart.

Beneath the Surface: The Infrastructure Beast

Training and serving large AI models is nothing like serving web apps or SaaS dashboards. The economics are brutal:

• Training CapEx is exploding: Training GPT-4 required tens of millions in compute. GPT-5 and beyod will require hundreds of millions, possibly billions. Altman’s own chip-and-compute fundraising pitch reportedly aimed for $7 trillion in global infrastructure. (Though this has since been revised to a more modest $500 billion compute cluster dubbed Stargate.)

• Inference isn’t cheap either. Even with optimizations, like mixture of experts (MoE), serving GPT-4 costs real GPU time and electricity. Every token generated pulls on a fragile, expensive stack of compute, bandwidth, and cooling. There is no SaaS-like “zero marginal cost” story here.

• Physical bottlenecks dominate. OpenAI’s growth is gated not by customer demand but by physical infrastructure: data center capacity, GPU supply chains, energy draw, thermal management, and sovereign compute agreements.

This looks much less like Salesforce, and much more like NextEra Energy, Intel, or AT&T.

Case Study: OpenAI vs NextEra Energy

Let’s compare OpenAI to a real utility: NextEra Energy (NEE), the largest electric utility in the US:

Ironically, this table was compiled with ChatGPT’s help. OpenAI might look like a utility but it is far more useful than most utilities.

NextEra owns generation, transmission, and distribution assets. It operates under regulatory oversight. Its growth is steady, its margins are stable, and its costs are amortized over decades.

OpenAI owns none of its compute infrastructure (yet). It relies on Microsoft subsidies and is exploring sovereing compute alliances and chip design because it’s running into physical, not virtual, constraints. Yet it’s being valued 9x richer per dollar of revenue than the nation’s largest utility.

That is not just overvaluation. That’s a category error.

Faith-Based Valuation

OpenAI’s current valuation implies one of two things:

1. That it will invent AGI and become the most important company in human history, extracting productivity rents from every sector.

2. Or that it will successfully execute a vertical-integration play: chip design, sovereign compute, data pipelines, agent ecosystems, and high-value apps layered atop commoditized models.

Either way, investors are betting on something far removed from SaaS economics. They’re betting on faith.

And if those bets don’t pay off? If margins compress? If inference gets commoditized? If regulatory regimes kick in? Then OpenAI will look less like a hypergrowth platform, and more like a mispriced infrastructure company with unsustainable burn.

The Bottom Line

OpenAI is not a software company. It sells software-like products, but its cost structure, capital needs, and scaling constraints make it far closer to a utility or an infrastructure play. Its investors are making a SaaS bet on a company that is structurally CapEx-bound and geopolitically entangled.

If OpenAI were valued like a utility, its price would be closer to $10-15B. But that would require seeing it for what it actually is, not what investors hope it might one day become.

Until then, the mismatch between valuation fantasy and infrastructure reality is one of the most underappreciated distortions in tech finance today.