Goldman Sachs says AI adoption surges across healthcare, retail and manufacturing

Goldman's newest AI readout draws a sharp line between experimentation and actual operating change. In London, a two-day European AI and Semis symposium brought together 500 participants, and Alexander Duval, Goldman Sachs Research's head of Europe Tech Hardware & Semiconductors, said the mood was upbeat because generative AI adoption is showing up in hard usage data, including one hyperscaler reporting a fivefold jump in token usage over the past 12 months.

Where the proof point is strongest

The biggest shift in Goldman's framing is that AI is no longer being treated as a single, abstract technology story. It is now an industry-by-industry adoption story, with the clearest evidence appearing where workflows are repetitive, data-heavy, and expensive to run by hand. Goldman says some commonly used large language models have become multiple times cheaper than earlier versions for both training and inference, which matters because cheaper models widen the set of use cases that can justify deployment.

That is why the report lands better when it talks about real tasks rather than future promises. Goldman points to surgeons using AI trained on 25 years of video data, industrial companies deploying thousands of autonomous robot agents that run 24/7, and retailers using generative AI for stock-taking and predictive analysis. The through line is simple: AI adoption looks most convincing when it shortens a cycle, lowers labor intensity, or improves a decision that can be measured in dollars.

Healthcare: data-rich work is moving first

Healthcare is one of the clearest examples because the work is both high-stakes and highly structured. Goldman highlights surgeons using AI trained on 25 years of video data to solve complex problems more efficiently, which is a good example of where the technology can augment expertise rather than replace it. In a sector where precision matters and errors are costly, AI becomes valuable when it helps clinicians see patterns faster, prepare better, or reduce time spent on repetitive review.

For Goldman employees covering healthcare clients, the implication is that AI conversations are getting more practical. The relevant questions are no longer whether a hospital system is “interested in AI,” but which clinical or administrative workflows can actually be improved, what the time savings look like, and how the risk is managed. That is the difference between a slide deck and an implementation plan, and it is the difference clients will increasingly expect bankers to understand.

Retail: the payback is fast enough to matter

Retail is where Goldman’s numbers get especially concrete. The report says generative AI is being used for stock-taking and predictive analysis, with some deployments generating three- to four-times returns on investment within three to four months. That kind of payback window is exactly why AI is moving from pilot projects to budgeted work, because retail operators live and die by margin pressure, inventory accuracy, and forecast quality.

This is also the kind of proof point that changes internal conversations at Goldman. If a retailer can point to ROI in a single quarter, then the AI discussion is no longer about novelty, it is about throughput, shrink, demand planning, and labor allocation. That matters for analysts and associates pitching clients because the most credible AI story is one tied to operating leverage, not to generic digital transformation language that sounds impressive and tells a CFO very little.

Manufacturing: AI is becoming an always-on worker

Manufacturing may be the most visible sign that AI is moving beyond office software. Goldman says industrial companies are deploying thousands of autonomous robot agents that run 24/7, a reminder that the technology is being used not just to assist human work, but to take on repetitive tasks at scale. In factory settings, that can mean better uptime, steadier output, and less dependence on manual monitoring.

The report also notes that one company said it has automated one week per month of human workload with AI. That is the sort of metric that cuts through hype quickly, because it translates into time returned to workers and cost removed from the process. For Goldman clients in industrials, that kind of result will shape where capital goes next, which vendors get renewed, and which internal teams become responsible for proving that AI systems are delivering something more than attractive demos.

Why cheaper models are not killing spending

A useful twist in Goldman's analysis is that lower model costs are not automatically reducing capital spending. The firm says absolute capex remains high because companies are using the extra compute to train and deploy more sophisticated multimodal systems across video, speech, images, and other formats. In other words, falling costs are expanding ambition, not ending it.

Goldman also points to recent milestones such as large language models earning a gold medal in a Math Olympiad, which underscores how quickly model capability is still advancing. That matters because the market often treats cost and capability as opposites. Goldman’s framing suggests they are moving together: models are getting cheaper to use and more powerful to deploy, which is exactly why adoption is spreading into more functions and more sectors.

The macro case is still about diffusion, not invention

Goldman’s broader research has been saying for some time that AI could be a large macro force if adoption spreads widely. The bank estimated in 2023 that generative AI could raise global GDP by about 7% and lift productivity growth by 1.5 percentage points over a 10-year period. It later said AI could start having a measurable impact on U.S. GDP in 2027 if diffusion broadens enough, and recent tracking still shows adoption among U.S. businesses below 19%.

That gap between capability and adoption is the real bottleneck. Goldman’s economists have also argued that jobs are being lost in occupations where AI can substitute for human labor, while employment rises where AI augments workers. In a separate workforce estimate, the bank said AI could displace 6% to 7% of the U.S. workforce if widely adopted, though the effect would likely be transitory as new job opportunities are created. The practical takeaway is that the labor market story is less about a single shock than about a long, uneven reallocation of work.

What Goldman people should take from this

For Goldman analysts, associates, VPs, and managing directors, the useful lens is not whether AI sounds powerful. It is where it is already producing measurable productivity gains, because that is what clients will care about in diligence, strategic planning, and capital allocation. The sectors moving fastest are the ones where a better forecast, a faster workflow, or a reduced manual burden shows up quickly enough to influence a quarter, a budget, or a bonus discussion.

The firms and bankers who will be most valuable are the ones who can separate durable adoption from slide-deck theater. In this market, the strongest AI narratives are not the loudest ones. They are the ones that can point to shorter payback periods, lower operating cost, and a clear answer to the question every client will keep asking: what is AI actually doing for the business today?