Consumer AI Has the Users but Not the Payers
Only 2.2% of consumers were paying for AI services as of May 2026, spending an average of $31 per month. That pairing of low conversion and modest ARPU, average revenue per user, describes the consumer business, according to data cited by TechCrunch.
Netflix counted 325 million subscribers in the same reporting. Consumer AI has wide use, cultural presence and daily habits. It does not have a similar base of people willing to pay directly for access.
Enterprise sales grew quickly. Anthropic revenue rose 12-fold in 2025 to nearly $4.6 billion, as reported by Reuters. Earlier in that ramp, the company had hit $3 billion in annualized revenue, a figure reported in May 2025 by Reuters. OpenAI had projected at that time that it would end 2025 with more than $12 billion in total revenue.
Costs grew faster. Anthropic spent $7.33 billion on compute and infrastructure last year. Its operating loss widened to $8.06 billion, according to Reuters. Distribution added a second cost. Anthropic pays over $350 million in distribution fees to cloud platforms, as reported by Reuters. For consumer AI delivered through app stores, cloud marketplaces and bundled assistants, paid acquisition and revenue share leave even less of the already small $31 per month.
Pricing power weakened at the same time. Cheap open models, models whose weights are freely available, have caught up on capability with frontier models, the most advanced proprietary systems, according to Ars Technica. Paying for frontier models buys a 4-month head start at 5x the cost, in the same reporting.
The wider economic return is still small. AI-related investment has lifted U.S. GDP by less than 0.4% since 2022, according to Reuters. Capital spending arrived first. Broad productivity effects have not followed.
The broader context here is a mismatch between cost structure and revenue model. Training and serving large models require continuous spending for accelerators, power, networking and staff. Consumer subscription revenue arrives in small monthly amounts and only from a narrow slice of users. Enterprise contracts can bridge the gap, and the Anthropic revenue path shows they are doing much of that work, but consumer remains a high-volume, low-yield channel where costs are counted in billions and revenue is collected in tens of dollars.
In my view, the PC era and the early commercial internet offer a useful parallel, not because the technology is similar but because the payment sequence is similar. General-purpose platforms often lose money on broad access while paid uses concentrate in workplaces, where time saved has a direct price. Consumers benefited greatly, just not always through a direct subscription. Free, ad-supported, bundled and employer-paid access carried earlier waves long before household spending caught up.
Looking at what this means for builders, the pressure points are inference efficiency, differentiation that survives a four-month window, and distribution without a platform tax. Inference is the cost of running a model to answer queries. If open weights deliver near-frontier quality for routine summarization, coding assistance, search and drafting, then paid consumer tiers need durable strengths in memory, agency, integration or trust, not raw benchmark leads. A brief quality edge rarely holds a 5x premium once good-enough options exist at scale. Otherwise $31 per month becomes a ceiling paid only by power users, while most users stay on free tiers subsidized by someone else.
Worth flagging is how this could still resolve optimistically. Low willingness to pay now does not fix willingness to pay later. Once models are embedded in workflows where they reliably complete multi-step tasks, payment can shift from novelty subscription to utility billing. Infrastructure built at a loss today lowers latency and cost per token for that future. The near-term economics are hard. The long arc still points toward cheaper inference making capable assistance widely available, even if the path to getting paid runs through enterprises first.
