Politixia
Artificial intelligence has become one of the technology industry’s biggest battlegrounds, but the debate is no longer just about which model performs best. Increasingly, it is about who controls the future of AI itself.
Open-source advocates argue that freely available models will democratize AI, lower costs, and prevent a handful of companies from dominating the market. Yet the dollars flowing through enterprise AI tell a different story. As businesses ramp up spending, the biggest winners continue to be the companies selling proprietary models.
For investors, that’s the trend worth watching because spending, not downloads, ultimately determines who captures the profits.
Enterprise Spending Tells A Different Story
On the All-In podcast, David Sacks challenged the popular narrative that open-source AI is winning. His argument was simple: ignore GitHub stars, Hugging Face downloads, and social media buzz. Follow where enterprises are writing checks.
According to the latest a16z CIO survey of 100 verified Global 2000 technology executives, open-source AI accounted for 19% of enterprise AI spending last year. This year, that figure has fallen to 11%. Closed models moved in the opposite direction, climbing from 81% to 89% of enterprise spending.
The same survey found enterprise preferences steadily shifting toward proprietary platforms. In January 2026:
| Metric | January 2026 |
| Prefer closed-source models | 36% |
| Prefer open-source models | 30% |
| Average annual LLM spending | $7 million |
| Average spending two years ago | $4.5 million |
| Expected spending increase in 2026 | 65% |
That growing budget is flowing primarily to OpenAI, Anthropic, and Alphabet‘s (NASDAQ:GOOG | GOOG Price Prediction) Google.
To put that into perspective, enterprises aren’t reducing AI investments. They’re increasing them. The question is simply where the money is going, and the answer is increasingly toward closed providers.
© 24/7 Wall St.
Cheap Tokens Don’t Necessarily Mean Valuable Work
Granted, open-source supporters raise a reasonable counterargument. Open models often cost between five and 20 times less per token than proprietary alternatives. If inference is dramatically cheaper, open models could process far more raw tokens even while generating less revenue.
That may be true for experimentation, internal utilities, or batch processing. The problem is equating cheap volume with valuable work.
Sacks argues enterprises continue relying on closed models for production systems where reliability matters most. Those applications often involve AI agents maintaining conversation history, long context windows, company-specific knowledge, and custom API integrations. Once those systems are deployed, replacing them becomes expensive and risky.
Ironically, the very features enterprises want from open source — vendor independence, portability, and data sovereignty — become harder to achieve after they’ve built workflows around proprietary models.
Those production workloads also tend to consume more tokens per task because they involve repeated interactions, larger context windows, and more sophisticated reasoning. While there is no public evidence that open-source models dominate total token usage, there is even less evidence they dominate the highest-value AI work.
The Lock-In Effect Is Becoming AI’s Moat
Perhaps the most important insight from Sacks wasn’t about market share but switching costs.
Software history shows businesses rarely migrate away from platforms deeply embedded in daily operations. AI appears to be following the same pattern. As enterprise LLM spending has risen from $4.5 million to $7 million over two years, companies are investing in agents, workflows, and integrations built around proprietary APIs. Those investments create operational inertia that favors incumbents.
That doesn’t mean open source disappears. It will likely remain the preferred choice for developers, research, experimentation, and cost-sensitive deployments. But the highest-value enterprise workloads increasingly belong to companies offering frontier performance and enterprise-grade support.
Key Takeaway
In short, popularity and profitability are becoming two different conversations. Open-source AI may generate millions of downloads and plenty of experimentation, but the a16z CIO survey suggests enterprises continue directing 89% of their AI budgets toward closed models. Cheaper tokens can drive volume, but they don’t automatically translate into the most valuable workloads.
Ultimately, investors should watch where enterprise dollars are accumulating because history shows the companies capturing spending, not attention, usually create the most lasting shareholder value.
Contact [email protected] for any questions or corrections.
GIPHY App Key not set. Please check settings