Does another glowing prediction about artificial intelligence actually mean anything beyond a temporary bump for chip stocks?
That’s the million-dollar question that Goldman Sachs’ latest missive attempts to answer. Their analysts are touting ‘agentic artificial intelligence,’ a fancy term for AI that can autonomously perform complex tasks, as the next big thing. The headline number? A projected 24-fold increase in global token consumption by 2030, hitting a mind-boggling 120 quadrillion tokens processed monthly. Sounds impressive, right? It’s the kind of forward-looking stuff that gets the venture capital crowd buzzing and the suits on Wall Street nodding sagely.
And then there’s the supposed ‘margin inflection’ for AI players, thanks to decreasing computing costs. Jim Schneider, a senior equity analyst over at Goldman Sachs Research, is quoted saying that concerns about hyperscalers’ compressed free cash flows will be fixed by raising gross margins, which then supposedly gives them more ‘headroom to spend.’ It’s a neat, tidy circle of logic that Wall Street loves.
The Bottlenecks No One Wants to Talk About Too Loudly
But let’s pump the brakes for a second. Every time someone whispers sweet nothings about the next AI gold rush, reality tends to elbow its way into the room. Right now, that reality comes in the form of a semiconductor shortage that Schneider himself admits could last another 12 to 18 months. Building new chip plants takes time—he figures it’ll be two years before supply truly catches up. Two years. That’s an eternity in the tech world, especially when we’re talking about supply chains that are already stretched thinner than a Silicon Valley startup’s runway.
And what about adoption? Goldman’s report acknowledges it’ll be a long, slow burn, particularly for businesses. Consumer-facing agents are already showing up in places like China, described as ‘always-on’ background helpers. But for enterprises? The integration, testing, and compliance hurdles are significant. Schneider notes that adoption rates are ‘still relatively low today,’ especially for small to medium-sized businesses.
The Long Tail of Agentic AI Adoption
Here’s the kicker: Goldman forecasts that only 12% of knowledge workers will be using agentic AI by 2030. That jumps to 37% by 2040. This isn’t exactly an overnight revolution; it’s more of a glacial creep. Contrast that with a PYMNTS survey from last August where over half of Chief Product Officers were just ‘considering’ or ‘exploring’ agentic AI. Fast forward a few months to November, and the number actively piloting or using it in production processes had shot up to nearly 1 in 4. That’s a significant shift from passive interest to active implementation, but still a far cry from the 24-fold token surge Goldman is predicting within a decade.
This whole narrative feels… familiar. It’s the classic tech hype cycle. New, powerful technology emerges, analysts project exponential growth, the market gets excited (and prices soar), and then the messy, human-scale realities of implementation, cost, and actual utility sink in. Who’s really making money here in the short term? The chip manufacturers, obviously. The cloud providers, no doubt. But the companies trying to use this stuff? That’s a much murkier picture.
So, When Do We See the Cash Flow Inflection?
Goldman’s report paints a picture of future prosperity driven by AI agents. They’re arguing that increased token consumption will eventually lead to higher revenues and, crucially, better margins for tech giants. The math is appealing: more activity equals more money. The problem, as I see it, is that the timeline for this ‘margin inflection’ is incredibly fuzzy. While consumer-facing applications might move faster, the enterprise adoption Schneider himself highlights is a marathon, not a sprint. Companies aren’t going to flip a switch and suddenly be awash in AI-driven profits. They’ve got integration headaches, data privacy nightmares, and the ever-present need to prove ROI. I’ve seen this movie before, and it rarely ends with immediate, across-the-board cash flow miracles.
The sheer volume of data processing predicted – 120 quadrillion tokens per month – is staggering. It implies a fundamental shift in how digital tasks are performed. Think of every email drafted, every piece of code written, every customer service query handled, but done by AI, autonomously. This requires an immense amount of computation, hence the projected surge in token consumption. If computing costs do continue to fall, and if the applications built on these agents are genuinely valuable and widely adopted, then yes, the economics could work out. But that’s a lot of ‘ifs’ in a world still grappling with basic digital infrastructure in many sectors.
The core of the Goldman thesis rests on the idea that increased efficiency and new capabilities unlocked by agentic AI will create new revenue streams and optimize existing ones, ultimately boosting profit margins. It’s a logical progression, but one that relies heavily on the quality and impact of the AI agents themselves, not just their existence. We’re talking about AI that doesn’t just answer a question, but actively goes out and does things on your behalf, potentially across multiple platforms and systems. That’s a huge leap.
