Vercel's latest platform data exposes a structural paradox that most analysts have missed. Over the past two months, open-source models have surged from 28.4% to 62% of total token consumption on the platform. Yet these models account for just 8.6% of total expenditure. The divergence is not a rounding error. It is the most important signal in AI infrastructure right now.
When I first audited tokenomics during the 2017 ICO wave, I learned that usage volume and economic value rarely travel together. The blockchain remembers every step; do you? The same principle applies to the AI model economy unfolding on platforms like Vercel. What the data reveals is a market splitting into two distinct layers, with open source models scaling at the bottom while closed models monetize at the top.
The Data Landscape
Vercel is not a neutral observer. As a platform that hosts millions of web applications, its AI usage data reflects the behavior of developers shipping production code. Between January and March, total token consumption across the platform jumped 59% quarter-over-quarter. That surge is not uniform. Open source models drove most of the growth, with their token share climbing from roughly 28% to 62% in roughly sixty days.
The more revealing number sits on the expenditure side. Closed source models, led by Anthropic, claim 65.1% of total spending. Anthropic alone accounts for 30% of token volume but the price per token is roughly double the market average. Meanwhile, open source models, including the emerging DeepSeek family, handle 62% of tokens at a unit cost roughly one-fifteenth of Anthropic's. The gap is not a statistical artifact. It is the result of price strategy and model capability meeting in an open market.
DeepSeek: The Second Place Signal
Perhaps the most disruptive data point involves DeepSeek. The open source model family has surpassed Google's Gemini suite in token consumption, making it the second largest model provider on Vercel. This is not a benchmark score or a lab result. This is developer behavior in production environments, and it tells a story that excludes any interpretation.
Open source models have crossed what I call the usability threshold. That threshold separates models that work in theory from models that are reliable enough to run in production. Based on my audit experience, once a model crosses this threshold, price becomes the dominant variable in adoption decisions. DeepSeek's price strategy has fundamentally reset the pricing anchor for the entire market.
The full implication is not simply that DeepSeek is better than Google at every task. It is that for a significant subset of production workloads, the performance gap is small enough that the price differential outweighs it. Developers are voting with their wallets and their workloads, and the votes are split by cost.
The Value Density Problem
Here is where the analysis gets uncomfortable. The 62% token share in the open source ecosystem does not translate into 62% of the market's economic value. It generates only 8.6% of spending. The 15x gap between token share and dollar share creates a fundamental question: what are these tokens actually doing?
The data suggests an answer. Most open source token consumption is probably concentrated in high-frequency, low-complexity tasks: code completion, text classification, information extraction, and routine content generation. These are workloads where marginal cost matters more than marginal quality. Meanwhile, complex reasoning, nuanced creative work, and high-stakes enterprise applications are still being routed to closed models where quality justifies a premium.
Patterns emerge only when chaos is organized. The Vercel data is organizing a pattern that the market narrative has not fully absorbed: the AI model economy is forming a two-layer structure. The bottom layer is high volume, low margin, and largely open source. The top layer is lower volume, high margin, and dominated by a few closed-source providers.
The market is not monolithic, and treating it as such leads to poor investment decisions. This is the same structural distinction I had to make in 2020 during DeFi Summer, when I manually verified the liquidity lock mechanisms on Uniswap v2 pools. The logic of token supply applies to model supply as well. An increase in volume does not mean an increase in value, and liquidity does not always equal safety.
The next year will determine whether this two-layer structure is stable or whether it collapses into a different configuration. The two main variables are the open source models' ability to move up the value chain, and the closed source models' ability to defend their premium pricing.
I have seen this pattern before. In the 2022 bear market, I watched the liquidation of leveraged positions across Celsius and Three Arrows Capital and realized that liquidity is a forward indicator, not a lagging one. The same is true here. Token volume is the forward indicator of adoption, but expenditure is the forward indicator of value. When the two diverge, one of them is lying.
The question is whether the token volume expansion in the open source sector is a precursor to value capture or a permanent feature of a two-tier market. The next twelve months of data will answer that question. The blockchain remembers every step, and so does the Vercel dashboard.
Code is law, but intent is the evidence. The intent behind the adoption of open source models is visible in the data. The question now is whether the intent will translate into revenue or remain stuck in a volume trap. Due diligence is the armor against narrative hype. The numbers are telling us where the market is heading. We should be listening.