Google Ventures founder Bill Maris said OpenAI and Anthropic could face serious pressure if Google decides to cut token prices by 80%. His point was blunt: the issue is not whether leading AI models can keep improving, but whether subscription and API businesses can survive a long pricing war built on heavy subsidies.
The discussion comes as OpenAI and Anthropic are described in the source material as moving toward IPO preparation. That matters. Private companies can keep leaning on fundraising to support aggressive pricing, but public markets tend to focus quickly on revenue quality, margins, and how much each paid user actually costs.
Subscription pricing is being compared with actual token use
The report cites analysis from SemiAnalysis that compared what users consumed under AI subscription plans with the public API value of those same workloads. According to the article, the gap is wide, with some plans implying subsidies as high as 70 times the monthly subscription fee. It also says more expensive plans often carry even larger subsidy multiples.
That pricing structure is framed as a deliberate strategy rather than a temporary promotion. Heavy users tend to be developers and enterprise decision-makers, the kind of customers that can pull entire teams and product stacks onto one platform. Still, the article argues that AI tokens do not create the same kind of lock-in seen in ride-hailing, food delivery, or social media.
Standardized APIs make switching easier
A central claim in the piece is that token services have weak switching friction. If Claude becomes more expensive, developers can move API traffic to GPT or Gemini. For everyday users, the switch may be even simpler. Many development frameworks already support multiple models, and interfaces are becoming more standardized across providers.
That weak lock-in changes the economics of subsidies. In the article’s framing, discounts are not building a lasting moat; they are keeping usage from slipping away. Once a cheaper alternative appears, users can leave fast.
AI agents add a new layer of cost pressure
The report also points to AI agents as a major variable. It says a complex agent workflow can consume 5 to 30 times more tokens than a normal chat session. One example cited in the source describes a developer using the $100 Claude Max plan and burning through token value close to that amount in a single agent coding session.
That matters because the more advanced the use case, the harder subsidies become to sustain. As enterprises push AI into coding, document analysis, and automation, token bills can rise much faster than simple consumer chat usage.
Maris argues Google has a funding edge others do not
Maris’s view rests on where the money comes from. The article says Google generates more than $300 billion in annual advertising revenue, giving it a mature cash engine that can support aggressive AI pricing. OpenAI and Anthropic, by contrast, are described as relying on investor capital. The source adds that OpenAI has raised more than $180 billion in total and reached a valuation above $850 billion, while Anthropic has raised more than $130 billion.
Maris put the scenario plainly on the podcast: “If I were Google and decided to arbitrarily cut token prices by 80%, what happens to OpenAI and Anthropic’s business model?” When asked how likely that was, his answer was “100%”, followed by the line “Capital as a weapon, tokens as a weapon.”
The bigger question is whether tokens become infrastructure
The article does not settle on a single winner. Instead, it outlines two broad outcomes. One is a familiar tech pattern: companies subsidize heavily, market share consolidates, and prices rise later. The other is a utility-style outcome, where tokens become a standardized resource like electricity, bandwidth, or cloud storage, pushing prices closer to cost over time.
Its conclusion leans on one factor above all others: lock-in. If model services remain easy to swap and APIs keep converging, then token competition may not end in a clean victory for any one company. It may become a prolonged contest where users keep getting cheaper compute, while providers struggle to turn that scale into outsized profits.