Why scale will not deliver ROI for AI

Many people assume AI will follow the same path as Amazon, Facebook, and Google — lose money for years, achieve massive scale, and then discover profitability on the other side.

That assumption is wrong.

More than 95% of AI users pay nothing. And why would they? Users are conditioned to believe that the “internet is free,” and so far, AI is. So they won’t pay for it later. Likewise, enterprise pilots are flailing at Microsoft and Salesforce. MIT reports that 95% of enterprise pilots fail.

So AI providers are turning to advertising. In the coming weeks, ChatGPT will begin serving ads. Google’s Gemini is expected to follow.

Have a look:

But unlike ecommerce, social media, or search, large language models must create a new content environment for every user interaction, and that act of creation carries a real cost that does not decline with scale.

Each response is generated on demand, customized to a specific prompt, and computed in real time. No matter how many users an AI system serves, it must still pay to generate each environment anew.

The core economic difference

Ecommerce, social, and search platforms monetize access to content they did not need to produce.

But AI platforms must compute the environment in which ads appear — every time, for every user — with no savings per interaction as volume increases.

Social, ecommerce, and search get their content environment for free

On platforms like Facebook, YouTube, Amazon, and Google, the platform does not create the underlying content.

• In social products, users create posts, videos, comments, and images.
• In ecommerce, sellers create product listings, descriptions, images, and reviews.
• In search, publishers create webpages, articles, and media that Google merely indexes and ranks.

In all cases:

• The content arrives fully formed.
• The platform does not pay to create it.
• It is stored once and delivered many times.

The most expensive part of the experience — content creation — is outsourced to users, sellers, and publishers at no cost to the platform.

Search, ecommerce, and social differ in interface, but their economics are identical: each monetizes access to content it did not need to pay to produce.

AI platforms must create the content environment themselves

AI systems receive no free content supply.

• Systems like ChatGPT do not receive user-generated answers.
• The “feed” is the response the model creates itself.
• Every interaction requires the system to generate new content on demand.

Even if content generation were inexpensive, it would still be a cost that ecommerce, social, and search platforms do not incur.

But AI content creation is not cheap — it is extremely expensive

This is where the economics break.

• Generating language requires large-scale inference on GPUs.
• Each response involves billions to trillions of operations.
• Power, hardware depreciation, memory bandwidth, and latency all apply.

This cost is many orders of magnitude greater than simply delivering existing content.

Scale helps ecommerce, social, and search. It hurts AI.

• More users mean more free content and lower average cost.
• More AI users mean more generated content and higher total compute spend.
• There is no free supply curve to exploit.

The bottom line

AI growth amplifies cost instead of diluting it.

That is why scale delivered ROI for Amazon, Facebook, and Google — and why it will not do the same for AI.