Hugging Face’s CEO said this week that roughly half of the Fortune 500 now runs open source AI models instead of renting access through a provider’s API. Cost, privacy, and customization are driving the shift. In the same week, a new McKinsey survey found that 32 percent of organizations have skipped buying a software product or feature entirely because they could build it internally using agentic coding tools instead. Two different data points pointing at the same underlying trend—large companies are increasingly choosing to own their AI infrastructure rather than rent it. Founders should understand what is actually driving that shift before deciding whether it applies to them.
What “Owning” AI Actually Means
Renting AI is what almost every founder does by default—calling an API from OpenAI, Anthropic, Google, or another provider, paying per token, and never touching the underlying model. Owning AI means running an open-weight model yourself, on your own infrastructure or a cloud instance you control, where you pay for compute rather than per-request pricing, and you have full control over the model’s configuration, the data it touches, and how it is deployed.
The Fortune 500 companies making this shift are not doing it because open models are dramatically smarter than the frontier options from major labs. They are doing it because at their scale, with massive, consistent usage volumes, the economics flip. Renting makes sense when your usage is unpredictable or modest. Owning starts to make sense when your usage is so large and so constant that the fixed cost of running your own infrastructure becomes cheaper than paying per token indefinitely.
Why This Almost Never Applies at Founder Scale, Yet
This is the same pattern we saw with Thomson Reuters building its own model a few weeks ago. The economics of owning infrastructure only work in your favor once your scale is large enough to justify the fixed cost. A Fortune 500 company processing enormous, continuous AI workloads reaches that crossover point. Most founder-sized businesses, even successful ones, are nowhere near that volume, and renting through an API remains the cheaper, simpler, more sensible choice.
The real risk here is founders reading a headline like “half the Fortune 500 stopped renting AI” and concluding they should be doing the same thing, without doing the actual math. Running your own infrastructure requires ongoing technical maintenance, security responsibility that used to belong to your provider, and capital committed to compute whether you use it fully or not. For most businesses that math does not work, and chasing the trend because it sounds sophisticated is a good way to add real cost and complexity for no actual benefit.
Where the McKinsey Build-Versus-Buy Data Actually Matters More for You
The more relevant data point for a founder-sized business is the build-versus-buy shift, not the rent-versus-own shift. Thirty-two percent of organizations skipping a software purchase because they could build the functionality themselves using agentic coding tools is directly applicable at almost any scale, because the barrier to building custom internal tools has genuinely dropped for everyone, not just companies with massive AI budgets.
This is worth a real look at your own operations. Before you subscribe to another SaaS tool for a specific internal workflow, it is increasingly reasonable to ask whether an AI coding assistant could help you or your team build a lighter, purpose-built version of what you actually need, rather than paying recurring fees for a general tool built for a much broader audience than just you. This does not apply to every tool—mission-critical infrastructure and anything requiring serious ongoing security maintenance is still usually better bought than built. But for simple internal workflows, dashboards, and single-purpose tools, the calculation has shifted meaningfully in the direction of building your own.
The Actual Decision Framework
Do not ask whether you should copy what the Fortune 500 is doing with model ownership. Ask two separate, more useful questions. First, is your AI usage volume large and predictable enough that owning infrastructure would actually save money over renting—for almost all founder-sized businesses today the honest answer is no, and that is fine. Second, is there a specific internal tool or workflow currently costing you a recurring SaaS fee that an AI coding assistant could help you build a simpler, purpose-fit version of—for many founders the honest answer here is increasingly yes, and that is worth exploring.
Conflating those two questions is the mistake to avoid. Rent versus own is a scale question that will not apply to most founders for a long time. Build versus buy for specific tools is a genuinely live question worth revisiting regularly as these tools keep improving.
If you want to think through where this applies in your own stack, this connects directly: What a Custom AI Agent Actually Costs vs What It Returns
Want results like this for your brand?
We work with a small number of founders at a time. See if you qualify.
See If We’re a Fit