Bloomberg examines why AI subscriptions are a hard sell for consumers

The AI industry has built some of the most talked-about software on the planet. Getting people to pay for it every month is proving to be a different project entirely.

A Bloomberg examination of the sector finds that AI subscription models may struggle because few consumers are willing to pay. That matters now because companies like OpenAI, Anthropic and Google have poured heavy investment into consumer-facing products, and that spending eventually needs revenue to justify it.

The numbers behind the hesitation

As of May 2026, only 2.2% of consumers were reportedly paying for AI services. Those who did pay averaged $31 per month.

In August 2026, paid personal subscriptions to major chatbots such as ChatGPT and Claude in the US rose to 4.5%, up from 2.1% a year earlier.

The spending that does exist is lopsided. The top 1% of AI spenders average about $903 a month, while the median payer spends around $25.

Subscriptions also dominate how these products make money. Approximately 85% of revenue from leading consumer AI web products comes from subscriptions, with only 13-14% coming from advertising.

Enterprises are a different story

While consumers hesitate, businesses are using more AI than ever. A Bloomberg Intelligence survey found that enterprises more than doubled their token consumption over the past year.

A quick note on terminology. In AI, a “token” is a small chunk of text that a model reads or writes, and providers often bill by how many tokens a customer runs through the system, a bit like a utility meter for language.

Enterprise demand is real, but it isn’t loyal. In the same Bloomberg survey, 84% of enterprises said they would switch models or optimize usage if token prices rose by 50%.

Why pricing is shifting

The core tension is cost. Running large AI models, a process known as inference, is expensive, and those costs are high relative to what consumers are willing to pay.

That’s why providers are reportedly trialing pay-as-you-go elements and usage-based pricing. Instead of one price for everyone, customers pay more or less depending on how much they actually use.

If the top 1% already spend about $903 a month while the median sits around $25, a single flat tier struggles to capture what different users are actually willing to pay.

What this means

Watch for a few signals in the months ahead. The first is whether the US paid subscription rate for major chatbots keeps climbing from 4.5%, or whether that jump from 2.1% proves to be an early-adopter bump that levels off.

The second is how widely usage-based pricing spreads into consumer products. If more providers adopt metered or hybrid models, it would mark a meaningful departure from the flat-fee approach that most of the industry launched with.

The third is the enterprise side. Token consumption more than doubled in a year, and if that pace holds while price sensitivity stays high, providers will face pressure to cut inference costs rather than raise prices.