AI Models Aren’t Valuable. They’re a Commodity, and Investors Should Be Wary.
Open-weight models, many from Chinese labs, now trail the closed frontier by a matter of months and cost a fraction as much to use. For investors, this suggests that model capability alone is unlikely to justify current valuations of frontier AI companies.
When DeepSeek released its R1 reasoning model in January 2025, with performance close to leading US systems and weights that anyone could download, Nvidia lost nearly $600 billion in market value in a single trading session. The market’s concern proved short-lived. Over the following 18 months, private valuations of the leading frontier labs rose sharply. Anthropic raised capital at a valuation of approximately $965 billion in May 2026, and OpenAI’s most recent private valuation stands at roughly $850 billion. Both companies have filed confidentially for public listings.
These valuations rest in large part on the premise that owning the most capable AI models confers a durable competitive advantage. Our review of capability benchmarks, pricing, adoption data, and recent corporate strategy suggests that premise is weakening. Model capability increasingly shows the characteristics of a commodity. Offerings are becoming interchangeable, leadership positions are short-lived, and competition is moving toward price. The frontier labs themselves appear to recognize this, as their recent expansion into enterprise services indicates.
This article examines four signals of commoditization and considers what they mean for investors evaluating AI companies.
Signal 1: The gap between open and closed models is measured in months
Epoch AI, an independent research organization that tracks model performance, found that from January through May 2026 the most capable open-weight models trailed the best closed models by an average of about four months on its composite capability index. Its earlier analysis, covering January 2023 to October 2025, put the average lag at about three months (Exhibit 1).
For most enterprise applications, a lead of this size is shorter than a typical procurement or product planning cycle. In practical terms, any capability premium a closed lab commands today can be expected to face a low-cost or free substitute within roughly two quarters.
The most competitive open-weight models now come predominantly from Chinese developers, including DeepSeek, Alibaba’s Qwen team, Moonshot AI’s Kimi, and Zhipu AI’s GLM series. Many are released under permissive licenses such as MIT and Apache 2.0, which allow companies to download, modify, and run the models on their own infrastructure without paying the developer.
Signal 2: Prices are converging toward the cost of compute
Pricing data points in the same direction. DeepSeek released its V4 model family on April 24, 2026, one day after OpenAI launched GPT-5.5. At standard list prices, DeepSeek’s flagship V4-Pro costs roughly one-seventh as much as GPT-5.5 and about one-sixth as much as Anthropic’s Claude Opus 4.7 for an equivalent volume of input and output. The smaller V4-Flash model is priced at close to one-hundredth of those premium systems (Exhibit 2).
Because open-weight models can be self-hosted, their effective price floor is the cost of the computing power needed to run them. Closed labs must price above that floor to recover multibillion-dollar training investments, and every new open release narrows the room they have to do so.
Signal 3: Customers switch readily, and usage is shifting toward open models
Durable technology franchises are typically protected by high switching costs, network effects, or proprietary data. Evidence from the model market suggests switching costs are low. Developers are commonly advised to place model calls behind an abstraction layer so they can substitute one model for another, and to revisit their choice of model every six months or so.
Enterprise market share data reflects this behavior. OpenAI, which led the market at the start of the generative AI boom, held roughly half of enterprise spending on large language models in 2023. By early 2026, its share had declined to between 27 and 29 percent (Exhibit 3).
Usage data from developer platforms shows a parallel shift toward open models. According to an analysis of data from OpenRouter, a widely used marketplace that routes requests across AI providers, Chinese open-weight models accounted for about 61 percent of tokens processed on the platform by May 2026, and four of the five most-used models were Chinese. The same analysis found that roughly 40 percent of new model derivatives created on Hugging Face were based on Qwen. Meta’s Llama, the leading open-weight family two years earlier, no longer appeared in the top rankings.
The pace of releases reinforces the pattern. In April 2026, three flagship models from three separate labs launched within eight days of one another. When the position of market leader changes this frequently, it offers a weak foundation for long-term valuation.
Signal 4: Frontier labs are moving into services
Corporate strategy provides perhaps the clearest signal. Within a single week in May 2026, both leading US frontier labs announced major moves into enterprise services (Exhibit 4).
Reuters and the Financial Times reported that private equity investors in OpenAI’s venture were promised a 17.5 percent annual return over five years, a figure OpenAI has not confirmed. A Gartner analyst characterized the moves as an acknowledgment that enterprises continue to struggle to install and use AI effectively.
The strategic logic is understandable, but the economics differ sharply from those of software. Services revenue scales with headcount, since each new client requires additional engineers and consultants. Accenture, the world’s largest consulting firm, operates at margins of roughly 15 percent. AI businesses already face structurally lower margins than traditional software: Bessemer Venture Partners estimates gross margins for AI companies at 50 to 60 percent, compared with 80 to 90 percent for mature software businesses, because every query incurs real computing costs. OpenAI has reportedly lost about $1.22 for every dollar of revenue, with cash burn projected at roughly $27 billion for 2026.
When a company must bundle its core product with labor-intensive implementation to win customers, it is typically a sign that the product alone no longer commands a premium. That pattern is characteristic of commodity markets.
Implications for investors
Commoditization of models does not imply that AI will create little economic value. It implies that value is likely to accrue elsewhere in the stack. In previous technology cycles, the largest returns went to companies that controlled scarce complements to a commoditized input, such as customer relationships, proprietary data, distribution, and deeply embedded workflows, and to companies that used the cheaper input to transform their own operations.
Valuations that assume frontier labs will earn software-like margins for decades on the strength of model leadership appear difficult to reconcile with a capability lead of about four months, a price gap of up to 100 times, and a strategic shift toward services. Investors evaluating frontier AI companies, including in the public listings expected over the next 12 months, may wish to consider the following questions:
- What share of the company’s revenue would remain if a comparable open-weight model became available at one-tenth of the price?
- How much of the company’s growth plan depends on services revenue, and at what margins?
- Which assets does the company own beyond its models, such as proprietary data, distribution channels, or workflows that customers cannot easily replace?
- How quickly has the company’s capability lead eroded following its previous model releases?
The DeepSeek episode of January 2025 offered an early indication of how quickly assumptions about model value can change. As open-weight models continue to improve, investors who distinguish between the value of AI and the value of any single model are likely to be better positioned.
About the Author
Brian Curry is a Kansas City–based Machine Learning & AI Strategist, Marketing Economist, and Applied Researcher. He drives Vector1, a personal open-source research initiative advancing marketing economics, causal inference, cognitive AI architectures, and the economics of intelligent automation.
He has led advanced data science and analytics initiatives across major enterprises including Koch Industries, Tractor Supply, Hallmark, Garmin, and AT&T. His work applies economic thinking and causal inference to connect marketing investment to outcomes, operational decisions to economic value, and AI adoption to competitive advantage.
Key Research & Open-Source Projects (GitHub: @Bodhi8):
- PyCausalSim — A Python framework for causal discovery and inference through simulation. Enables testing of counterfactual scenarios to identify true structural drivers.
- Papilon — A complex system and causal discovery library for Media Mix Modeling (MMM) and adaptive systems.
- Regimecast — Applied research and tooling for time-series regime detection and forecasting.
- CarryNet — Open-source research and modeling focused on commodity carry trade analysis.
- Foodmesh-ml — Machine learning, optimization, and simulation framework for modeling distributed food systems.
- Memory-Node Encapsulation (MNE) — A cognitive AI architecture and novel data structure for artificial episodic memory.
- The Autonomy Economy — A macroeconomic framework projecting the dual impacts of massive productivity growth and extreme price deflation in an AI-automated world.
- Meaning-Flow — Research focused on semantic content modeling and automated coverage gap analysis.
Contact: brian at vector1.ai | Vector1 | GitHub | LinkedIn
This article may be shared with attribution under Creative Commons Attribution 4.0 International (CC BY 4.0).
Sources
- Epoch AI, “Open models lag state-of-the-art closed models by 4 months,” May 2026. https://epoch.ai/data-insights/open-closed-eci-gap
- Epoch AI, “Open-weight models lag state-of-the-art by around 3 months on average,” October 2025. https://epoch.ai/data-insights/open-weights-vs-closed-weights-models
- VentureBeat, “DeepSeek-V4 arrives with near state-of-the-art intelligence at 1/6th the cost of Opus 4.7, GPT-5.5,” April 2026. https://venturebeat.com/technology/deepseek-v4-arrives-with-near-state-of-the-art-intelligence-at-1-6th-the-cost-of-opus-4-7-gpt-5-5
- CloudZero, “DeepSeek pricing 2026,” 2026. https://www.cloudzero.com/blog/deepseek-pricing/
- BenchLM, “DeepSeek V4 Pro vs Claude Opus 4.7 vs GPT-5.5: The Frontier in April 2026,” 2026. https://benchlm.ai/blog/posts/deepseek-v4-vs-claude-opus-4-7-vs-gpt-5-5
- Chris Zeoli, “China’s Open-Weight Takeover,” Data Gravity, June 2026. https://www.datagravity.dev/p/chinas-open-weight-takeover
- FutureSearch, “OpenAI Revenue, Losses, and IPO Valuation,” 2026. https://futuresearch.ai/openai-financial-forecast/
- MarkTechPost, “What is a Forward Deployed Engineer,” May 2026. https://www.marktechpost.com/2026/05/20/what-is-a-forward-deployed-engineer-the-ai-role-openai-anthropic-and-google-are-hiring-in-2026/
- Context Studios, “The AI Consulting Land Grab: Anthropic vs OpenAI,” May 2026. https://www.contextstudios.ai/blog/ai-consulting-land-grab-anthropic-vs-openai
- AI Business, “OpenAI Launches AI Consulting Company, Following Anthropic,” May 2026. https://aibusiness.com/generative-ai/openai-launches-ai-consulting-company-anthropic
- Forbes, “OpenAI vs Anthropic IPO: How They Compare,” September 2026. https://www.forbes.com/sites/investor-hub/article/openai-vs-anthropic-ipo-comparison/
- Tech Insider, “OpenAI IPO: $850B Valuation,” 2026. https://tech-insider.org/openai-ipo-850-billion-valuation-2026/
- Digital Applied, “AI Unit Economics: Pricing and Margins for AI Services,” June 2026 (citing Bessemer Venture Partners). https://www.digitalapplied.com/blog/ai-unit-economics-pricing-margins-services-2026-framework
- DevsUnite, “Open vs Closed AI Models: Which Should You Build On in 2026?,” 2026. https://devsunite.com/blog/open-vs-closed-ai-models-which-should-you-build-on-in-2026
