OpenAI: Will AI Soon Be Billed Like Electricity?

The idea that artificial intelligence (AI) could become a paid resource, akin to water or electricity, is taking shape. Sam Altman, the CEO of OpenAI, recently shared his vision at the BlackRock Infrastructure Summit in Washington, as reported by Business Insider. He envisions a future where AI would be consumed and billed like a public utility, based on actual usage.

Today, AI is already ubiquitous in our daily lives, simplifying various tasks, from information searches to homework assistance. However, the question of the future cost of this technology is becoming increasingly pressing. Altman mentioned a future where paid AI would function like a public service. This means that everyone would consume this innovation and pay according to actual usage, much like electricity today.

An Economic Model Based on Tokens

The model proposed by OpenAI relies on the use of "tokens," units of computation that are used to process and bill each piece of data. Sam Altman explains that AI will become a paid resource, purchased according to needs. Specifically, users would no longer pay for a global tool, but for each unit of computation used. Thus, each request made by a user would be processed by the AI and then billed as energy consumption.

Altman emphasized that his team envisions a future where AI would be a public service, similar to electricity or water, responding to growing demand. The user would submit a request, the AI would process it, and then the consumption would be billed, just like our current heating or electricity bills.

Challenges Related to Computing Power

To function, AI requires considerable computing power. This computational capacity is essential for training models and processing user requests. If the infrastructure fails to keep up with demand, two major problems could arise: the cessation of sales of certain capacities by companies and a rapid increase in prices.

In such a scenario, access to paid AI could become limited, reserving this resource for the wealthiest users. Altman also mentioned the possibility that governments might intervene during energy crises to prioritize the allocation of computing power to essential sectors such as health, defense, and education.

The Importance of AI Infrastructures

In response to growing demand, tech companies are heavily investing in data centers, which house thousands of specialized processors to run AI models. Lisa Su, CEO of AMD, estimated that the planet may need more than ten yottaflops of computing power in the next five years, a capacity ten thousand times greater than that of 2022.

However, this expansion faces a major obstacle: electricity. Data centers consume energy comparable to that of small cities, which is already straining electrical grids. According to Gartner, 40% of AI centers could face energy constraints by 2027 if grids do not adapt quickly.

The Implications of Paid AI

The main advantage of paid AI lies in its simplicity: users only pay for what they consume. A simple request costs little, while a complex task requires more computation and increases the bill. This model encourages more efficient use of computing resources.

However, it raises questions of access. If AI becomes a billed resource, some users may be excluded for financial reasons, while others would have access to more powerful tools. Additionally, price volatility could pose a problem, as unlike electricity, which is often regulated, the cost of AI would be subject to market fluctuations.

In short, we are moving from an era of experimentation to one of commodification. In my opinion, if AI becomes a public service, it will likely need to be regulated as such. It is impossible to leave a resource as critical as augmented human intelligence without guarantees of universal access and price stability.