Benedict Evans, one of the most trusted and influential independent analysts in for artificial intelligence (AI), put in to a historical perspective the evolution and current state of AI, comparing it to past technological shifts like mainframes, PCs, and smartphones, in the kickoff keynote address at B2B Marketing Exchange (B2BMX) 2026, powered by Advertising Week, today.
“I think a good place to start in talking about AI in any way, is to try to break it apart,” said Evans. “It sometimes seems to me that AI has become the new Metaverse, in that people said, and you have no idea what they mean. It could mean anything.”
In his address to a full ballroom at the Omni La Costa Resort in Carlsbad, CA, Evans highlighted the significant investment in AI, the potential for AI to automate and create new jobs as well as the uncertainty brought on by the potential disruption AI will bring, likening it to past technological shifts that transformed industries.
Investment and Market Dynamics
Evans noted the Big Four platform companies spent $400 billion last year on AI and have budgeted $650 billion for 2026.
“We have this sort of succession of announcements of $100 billion here, 100 billion dollars there, 100 billion dollars here, 100 billion dollars there,” said Evans. “Pretty soon you’re talking about real money.”
Supply Crunch
The surge in capital is quite difficult to spend all at once, noted Evans and has resulted in a supply crunch for GPUs, electricity and electrical equipment and HVAC and everything else around actually building these things,
“A bunch of places around the U.S. where people are wondering if they actually want the giant data center around the corner from the homes,” he said. “So there’s a kind of crunch in actually getting this stuff deployed, which gets you this quote from Microsoft that it’s been basically impossible to build enough capacity since ChatGPT launched.”
This current environment has allowed for the rapid improvement in AI models, keeps generating more demand for more keeps they keep getting better and getting new capabilities, which, of course, creates more demand for more capex
“You’ve got the next model that’s going to be better, and the model that you have is basically irrelevant,” said Evans. “After six to nine months, you always have to be working on the next one, so that drives even more capex.”
Impact on Business
The impact on the software industry has experts what will the industry to look like when all the dust is settled. Traditionally, the meaning of software is, there’s no hardware, there’s no capex., Now you have network effected, and you had winner takes all effects that tend to produce monopolies like Microsoft, Google, Meta, and Apple
“Whereas it may be that LLMs end up looking more like chips. With each generation, they get harder and more expensive, and so you have fewer and fewer companies that can keep up,” said Evans. So you may end up with two or three companies or four companies that can make LLMs.”
In his discussion around AI, Evans noted that every question end in one of two ways.
“The first answer is, either it will work exactly like every other platform shift,” said Evans. “The other set of is no one knows….like, how much better will these models get? Where will the scaling rules end? How expensive will they be? Will they kill us all, or treat us as pets, like we don’t know. Check back in five years time. We’ll see.”
Historical Patterns
What Evans did predict was a surge in hiring that that would be followed by a reduction in workforce, using the bell curve of automation is elevators. The first half of the 20th century was elevators which employed people to operate them as America built a lots of buildings with elevators. The second half of the 20th century, companies like Otis Elevators automated the process that eliminated the job elevator attendants.
“This is what tends to happen with new technologies,” he said. “What happens with automation is that it’s only technology if it’s new; once it’s been around for a while, we call it automation, and then we don’t even bother calling it automation anymore.”
This has already begun with AI, with it now more commonly beiong reffered to software, according to Evans.
“I certainly started hearing in the last six months, people saying, Oh, that’s not AI, that’s just image recognition. That’s not AI, that’s just natural language processing,” said Evans. “Wait a minute, 15 years ago, that was witchcraft. 10 years ago, natural language processing was definitely AI. And today, oh, that’s not AI. That’s just software.”
