Is the technology industry ready for AI “superheroes”?
At Star Trek, Starship Entertainment had a great president, Montgomery “Scotty”, which regularly had to explain to Captain Kerk that some things were impossible to withdraw, due to practical aspects such as physics laws.
“Cannae engines take it, the captain!” It is a famous quote that the actor may not have said by the TV program. But you get the idea.
We may approach this moment in the technology industry at the present time, as it collects the direction of AI’s momentum.
The field began to shift from Chatbots relatively simple to the most capable artificial intelligence factors that can completely complete complex tasks. Is there sufficient computing power to keep this shift?
According to Barclays’s latest report, the artificial intelligence industry will have enough capacity to support 1.5 billion to 22 billion artificial intelligence agents.
This may be sufficient to make a revolution in white collar work, but an additional computing force may be needed to operate these agents while meeting the demand for consumers on chatting bankers, and Barclays analysts explained in a note for investors this week.
Everything about symbols
Artificial intelligence agents create much more symbols for each user’s inquiry than the traditional Chatbots, which makes them more expensive.
Symbols It is the language of obstetric artificial intelligence and is at the heart of pricing models emerging in this industry. Artificial intelligence models dismantle other words and inputs in numerical symbols to facilitate and understand treatment. One symbol is about ¾ of the word.
The most powerful artificial intelligence agents may rely on “Thinking” models, such as Openai’s O1, O3 and Deepseek’s R1, which breaks queries and tasks into more management parts. Each step in these thought chains creates more symbols, which must be treated with an artificial intelligence servers and potatoes.
“Products of agents work on thinking forms mostly, and generate about 25x distinctive symbols for each query compared to Chatbot products,” Barclays analysts wrote.
“Super factors”
Openai provides Chatgpt Pro, which costs $ 200 per month and clicks in the latest thinking models. Barclays analysts estimated that if this service uses the O1 model to start operating, it will generate about 9.4 million icons annually per subscriber.
There were recent media reports that Openai could provide the services of the most powerful intelligence agent that cost $ 2000 per month or even $ 20,000 per month.
Barclays analysts referred to these “super agents”, and they estimated that these services can be born from 36 million to 356 million symbols annually, for each user.
More chips, captain!
This is a quantity that surpasses the symbols that will consume a mountain of computing power.
The artificial intelligence industry is expected to have 16 million accelerators, a type of artificial intelligence chip, online this year. Nearly 20 % of this infrastructure can be dedicated to artificial inference – mainly the computing power needed to operate artificial intelligence applications in actual time.
If the agents are launched and are very useful for consumers and institutions users, we are likely to need “many inference chips,” and Barclays analysts warned.
Analysts added that the technology industry may even need to re -display some of the previously used chips to train artificial intelligence models and use these to infer, as the analysts added.
They also expected that the cheaper, smallest and more efficient models, such as those developed by Deepseek, should be used for artificial intelligence agents, instead of special models with snake.