Article on the AI economy
Generative AI and productivity
Generative AI is mostly an economic story about productivity. Here is where the gains show up, how large they could be, and why they depend on compute, with real data.
Key takeaways
- McKinsey estimates generative AI could add the equivalent of 2.6 to 4.4 trillion dollars in value annually across the use cases it studied.
- Most of that value sits in customer operations, marketing and sales, software engineering, and research and development.
- Productivity gains come from automating routine language, code, and analysis tasks, not from replacing whole jobs at once.
- Every productivity gain runs on GPUs, so wider adoption pushes compute demand higher.
Why productivity is the heart of the AI story
Productivity is simply how much useful output people can produce for a given amount of work. When a tool lets the same team do more, or do it faster and with fewer errors, productivity rises. Over the long run, productivity growth is what raises wages and living standards, which is why economists treat it as one of the most important numbers there is, more important than most short-term headlines.
Generative AI matters here because it speeds up tasks that used to require slow, manual effort, such as drafting, summarizing, coding, and analysis. The interesting question is not whether it helps at all, but how much, where, under what conditions, and how durable the gain turns out to be once the novelty wears off and teams settle into steady use.
This article focuses on that practical view: which work generative AI actually accelerates, how large the effect could be, why results differ so much between organizations, and why each gain ties back to physical compute. The numbers come from McKinsey, with clear attribution throughout so you can see exactly where each claim comes from.
Where the productivity gains concentrate
Customer operations
AI assists with support, drafting responses, and handling routine questions, which McKinsey identifies as a major source of value across many companies.
Marketing and sales
Content drafting, personalization, and campaign work scale faster with generative tools, another area McKinsey highlights as high value.
Software engineering
AI helps write, review, and debug code, speeding up a function that is itself central to building more AI and more software.
Research and development
Faster analysis and idea generation support research and development, which McKinsey lists among the highest-value areas.
How the value breaks down
$2.6 to 4.4T
Value generative AI could add annually across studied use cases, according to McKinsey.
Source: McKinsey and Company, June 2023
4 functions
Where McKinsey says much of that value concentrates: customer operations, marketing and sales, software engineering, and research and development.
Source: McKinsey and Company, June 2023
How large could these gains be
According to McKinsey, generative AI could add the equivalent of 2.6 to 4.4 trillion dollars in value annually across the use cases it analyzed. That range is wide, but the key insight is that the value is concentrated rather than spread thin. A small set of business functions accounts for most of the potential, which makes the figure easier to act on than a single blended average would be.
This concentration is good news for planning. It means organizations can find the highest-value uses rather than trying to apply AI everywhere at once, which usually wastes effort and budget. Starting where the work is rich in language, code, and analysis tends to produce the clearest, fastest results, and those early wins build the internal case for going further.
It also means the demand for compute clusters around the functions that adopt fastest. The more a team leans on AI to draft, summarize, and analyze, the more model runs it generates, and those runs all consume real hardware in a data center somewhere, whether the team thinks about that or not.
Productivity has a physical address
It is easy to think of a productivity gain as something that happens on a laptop screen. In reality, the model behind that gain is running in a data center, drawing power and producing heat that has to be removed by a cooling system someone designed and maintains.
When productivity rises across many teams at once, the cumulative effect is a steady climb in the amount of compute that has to be available and reliably operated. The convenience on the screen and the load in the data center are two ends of the same chain.
How the gains actually happen in practice
Productivity gains rarely come from handing a whole job to a model. They come from compressing the slow parts of a task. A first draft that took an hour might take ten minutes, leaving more time for review and judgment. A code change that needed careful boilerplate gets scaffolded in seconds, so the engineer spends their attention on the hard logic instead.
Because the gain is task-level, it depends heavily on how well people use the tools. Teams that learn to prompt clearly, check output, and build AI into their workflows see more benefit than teams that bolt it on as an afterthought and then complain that it did not help. The skill of using the tool is part of the result.
This is why the same technology produces very different results across organizations. The compute is similar and widely available, but the skill in applying it is not, and that gap shows up directly in measured productivity. It is also why early disappointment is common and not always a sign that the technology has failed.
The limits worth keeping in mind
Generative AI is not free of friction. Output needs checking, sensitive work needs guardrails, and not every task benefits. McKinsey frames its figures as potential value across studied use cases, not a result every company will automatically capture. The gap between potential and captured value is often where the real work lives.
There is also a learning curve at the organization level. Realizing the gains takes process changes, training, and trust, which arrive gradually and unevenly across teams. Treating the estimates as a ceiling to work toward, rather than a promise to bank on, keeps expectations honest and keeps projects from being judged too harshly too early.
Finally, the gains are not static. As models improve and costs change, the set of tasks worth automating keeps shifting, so productivity from AI is better understood as a moving target than a one-time upgrade that a company installs and then forgets about.
Productivity gains run on hardware
Behind every productivity gain is a model running on a GPU. As more teams use AI to do more work, the total amount of inference, the compute needed to run a model, climbs steadily. Productivity at the level of a whole economy translates directly into demand for compute capacity that has to be built, powered, and operated.
That is why some people choose to own the hardware that produces AI compute rather than only consume AI tools. One response is managed GPU ownership, where you hold physical NVIDIA-powered hardware and a professional team runs the hosting and operations inside American data centers, so you are not maintaining the equipment yourself.
This does not promise any result. Operational benefits are not guaranteed and depend on utilization, uptime, demand, costs, hardware performance, and market conditions.
References and data
- The economic potential of generative AI: The next productivity frontier. McKinsey and Company. June 2023.
Common questions about AI and productivity
According to McKinsey, generative AI could add the equivalent of 2.6 to 4.4 trillion dollars in value annually across the use cases it analyzed, with most of that concentrated in a few business functions rather than spread evenly.
McKinsey points to customer operations, marketing and sales, software engineering, and research and development. These functions involve large amounts of language, code, and analysis, which is the kind of work generative AI handles well.
Mostly it assists by compressing slow, routine parts of a task, such as drafting or scaffolding code, so people spend more time on judgment and review. The gains are usually task-level rather than whole-job replacement.
Because the benefit depends on how well people use the tools. Teams that learn to prompt clearly, check output, and build AI into their workflows tend to see more measured gain than teams that add it as an afterthought.
Often longer than expected. Capturing value takes process changes, training, and trust, so early results can look modest. McKinsey frames its figures as potential, which organizations realize gradually rather than the moment a tool is switched on.
Yes. Every AI task runs on a GPU, so as more teams use AI to do more work, the total amount of inference rises. Broader productivity gains tend to push demand for compute capacity higher over time.
Want a real position in AI compute?
Talk through what owning managed NVIDIA GPU hardware would look like, with no pressure and straight answers.
Operational benefits are not guaranteed and depend on utilization, uptime, demand, costs, hardware performance, and market conditions.