Feature
The Gist
- Bubble concerns. Goldman Sachs questions the financial sustainability of generative AI investments.
- Limited returns. Generative AI has yet to deliver significant financial returns despite high spending.
- Efficiency gains. While AI improves some tasks, transformative applications remain elusive.
Late last month, Goldman Sachs published a remarkable, 32-page report wondering aloud whether all the investment in AI is worth it.
It asked, “GenAI: too much spend, too little benefit?” and placed a skeptical interview with MIT economist Daron Acemoglu right at the top. In the conversation, Acemoglu questioned the scaling laws boosted by generative AI research labs and said he doubted any transformative applications of the technology would take hold within the next decade.
Goldman Sachs Questions Generative AI Bubble
Goldman knew exactly what it was doing. As money’s flooded into generative AI projects — funding chips, model training, startups and big tech companies — it’s been somewhat heretical to ask whether (or when) the investment would net a return. But now, with untold billions going in and the timeline to results merely a guess, Goldman and others are starting to question where this all leads.
“I do think the bubble is being questioned at least,” Acemoglu told me via email. “It may not be bursting, but there is a healthier discussion. One can hope.”
Related Article: AI Bubble About to Burst? What CX Leaders Need to Know
Generative AI's Financial Return Questioned
Generative AI has wowed to date, but the financial return has been limited. The technology’s helped some businesses make some tasks more efficient, its automated work in areas like customer service and coding, and it’s been useful for millions of people accessing broadly available bots like ChatGPT, Claude and Gemini. (It’s also useful for image generation.) But the bigger promise of the technology aiding in scientific breakthroughs, working as an agent and even reasoning through requests has yet to come to fruition. “Will this large spend ever pay off?” Goldman asked in its report. As Acemoglu said, it may take a while.
Generative AI Faces $1 Trillion Cost
Investors are typically patient with potentially transformative technology, but the problem is generative AI is still incredibly expensive to train and run. Goldman said the tech industry is poised to spend $1 trillion on generative AI in “the coming years,” and Sequoia recently said $600 billion in revenue will be needed to make a return on the investment in generative AI spent so far. Models and chips are becoming more efficient, but training generative AI models still costs hundreds of millions of dollars, if not more than $1 billion.
Related Article: Generative AI Might Be Slamming Right Into a Resource Wall
Generative AI's Honeymoon Stage Ending
The spending will likely continue apace for now, but the honeymoon stage may be over. Big tech companies, spending the bulk of the money on generative AI, are unlikely to slow down given competitive pressures. But they’re already hearing from customers who are wondering where the benefit is. Chevron’s chief information officer Bill Braun, for instance, told The Information that he’s still looking for a generative AI application that can provide value to the company. Successful consumer applications of the technology, meanwhile, are hard to find. Millions of people have tried the bots and not returned.
“If this is the amazing magical thing that will change everything, why do most people say, in effect, ‘very clever, but not for me’ and wander off, with a shrug?” asked analyst Benedict Evans. “And why hasn’t there been much growth in the active users (as opposed to the vaguely curious) in the last 9-12 months.”
The next set of models — GPT5, Claude4, Gemini2 — will have to deliver major improvements or the questioning will build. Market caps may suffer, sales calls may be harder to book and the momentum will slow. But anyone familiar with AI research knows this is not a new phenomenon. The technology is defined by spurts of incredible progress followed by moments of disillusionment, only to repeat the process over and over again.
