Claim
AI model prices fell 41% while token usage rose 50% between July and October 2026. Frontier models' share of total token consumption slipped from 53% to the mid-40s in two months. In commodity markets, benchmark leads do not create durable advantages. Distribution, interface ownership, and routing data do.
Mechanism
Commodity dynamics follow a known pattern: price and capability converge across providers, which shifts competition from product differentiation to volume and distribution. Tunguz names three moves for labs and builders in this environment. First, partner and resell models rather than competing solely on proprietary capability. Second, control the interface or harness the user operates in, because the interface controls attention, and attention is the scarce resource when underlying models are interchangeable. Third, capture routing data. An intermediary that routes queries across multiple models at scale generates training signal that improves future model quality. TypeSafe's Jev handles routing at $0.04 per million input tokens, roughly 75 times cheaper than routing through standard LLMs, which means whoever routes the most queries can do so profitably while accumulating compounding data advantages.
Conditions
Holds when: model capabilities are close enough that users cannot reliably distinguish output quality across providers; price is falling fast enough that switching costs are lower than switching savings.
Fails when: a lab produces a model with a capability gap large enough to justify a significant price premium (e.g., reasoning on frontier scientific tasks); a single application locks users in at the workflow level regardless of model quality.
Evidence
Tunguz cites the following data points from his portfolio and market observation in the essay:
- Token prices fell 41% since July 2026 while token usage rose 50%.
- OpenAI cut prices on its Luna model by more than 80%.
- Frontier models' share of total tokens fell from 53% in August to the mid-40s by early October.
- Spending among the top 1% of adopters fell 9.7% in August, to $7,205 per employee per month.
- TypeSafe's Jev routes at $0.04 per million input tokens, roughly 75x cheaper than standard LLMs.
"Distribution has become the moat."
"the spoils do not accrue to the lab with a marginal benchmark lead"
"Winning share becomes the only game that matters"
Signals
- Your model costs are falling faster than your output quality is improving
- Customers ask about price per query before asking about capability
- Routing and intermediary infrastructure in your stack is cheaper than your primary model calls
Counter-evidence
The commodity argument assumes models remain substitutable. If one lab produces a step-change in reasoning capability (e.g., reliable scientific research autonomy), that lab's model commands pricing power and the distribution-moat thesis weakens. Tunguz acknowledges OpenAI's run rate is approaching $70B, close to Anthropic's, which suggests both frontier labs still have meaningful pricing power on their best models even as commodity pressure builds at the lower tiers.
Cross-references
- AI competitive advantage has moved from model access to mastering three harness disciplines: Tunguz's earlier framework on winning through the harness layer. This card extends that argument with the data showing the commodity inflection has now arrived.
- Run up the stack, use commoditized capabilities as inputs to create higher-order human value: the broader pattern of moving to higher-value layers when the layer below commoditizes. Distribution and data are higher-value layers when model quality converges.