The economics of AI – are we tapping the full market potential?
Subho Mukherjee,
Vice President of Sustainability, Nokia
Judging by the exuberance behind Artificial Intelligence, one could be led to believe that the economic benefit of the AI supercycle is already guaranteed. But a shallow dive into the detail reveals a different picture.
For AI to succeed, two things are fundamental: supply and demand.
Is there supply? Yes, albeit not fully stress tested.
Is there demand? Yes, consumers and industries want to embrace AI.
But demand and an addressable market are not the same thing. Is the addressable market big enough to deliver the AI supercycle? That’s what remains to be seen.
What is the AI supercycle?
Artificial intelligence is entering a transformative multi-decade phase known as the AI supercycle, where it will become integral to all industries and technologies, driving exponential innovation and economic growth through pervasive automation and data optimization.
The difference between demand and the addressable market
If demand is the pool of waiting buyers, the addressable market is those who could do so. Focusing in on the U.S. market, arguably the most advanced AI market, illustrates the challenge.
Connectivity is widely available, but when you look at advanced connectivity technologies like 5G Standalone, a pre-requisite for reliable AI access, the numbers dip. According to Ookla, 5G Standalone availability could be as low as 24%.
Without a high performing communications network infrastructure that 5G Standalone brings, it becomes difficult to deliver the low-latency and high-reliability of AI demands and for the use of Physical AI in industrial applications. A 5G Standalone cloud-native core unlocks low-latency, deterministic performance and real-time network control that AI applications depend on.
Additional research suggests performance challenges. For the 70 largest cities, the lowest average latency performance was 34.06 milliseconds, while our own research with 1,000 technologists in the U.S. suggests 72% would want sub-29 millisecond latency within 2-3 years.
Mobile network performance of selected cities in the U.S.
City |
Download (Mbps) |
Upload (Mbps) |
Latency (ms) |
Consistency |
|---|---|---|---|---|
New York |
187.81 |
21.57 |
35.77 |
91.80% |
Austin |
219.69 |
11.58 |
49.17 |
86.30% |
Los Angeles |
185.61 |
15.47 |
38.13 |
87.70% |
Houston |
204.51 |
13.67 |
43.81 |
87.80% |
Chicago |
188.03 |
21.49 |
35.07 |
91.60% |
Source: Ookla United States: Speed Test Connectivity Report
This narrows the addressable market dramatically, not to mention other considerations that would have a material impact. How accessible is compute infrastructure? Is this a region with surplus energy? Do the AI adopters have access to the right digital skills? These elements shrink the addressable market further.
This example does not consider use cases that would rely on fixed infrastructure or private networks, but it illustrates a point - there is a difference between demand and the addressable market.
The cart, the horse and the bandwagon
Going back to our starting assumption, for AI to deliver economic value we need both supply and demand. If supply is the cart, there is no shortage of momentum; the bandwagon is already full of organizations mobilizing AI at speed.
But before we worry about whether the horse is strong enough to pull that cart, we need to confront a more fundamental problem; in many parts of the world, there is no horse at all.
Entire economies remain insufficiently connected to participate meaningfully in the AI era. In these markets, the issue is not delayed adoption, it is the absence of an addressable market altogether.
Without reliable broadband connectivity, there are no customers to sell to, no workloads to serve, and no economic upside to capture. For AI providers, this places a hard ceiling on growth that enthusiasm and innovation alone cannot break.
In other words, lack of connectivity does not just exclude people from the benefits of AI, it caps the total economic potential of the AI market itself.
If regions remain unconnected, AI sellers are competing over a progressively narrower pool of buyers, regardless of how strong global demand signals appear.
Even in regions where connectivity does exist, the challenge does not end there. To pull the AI bandwagon to market, the horse must be not just present but trained for a very different kind of work. AI workloads are interactive, bursty and event‑driven, placing stringent demands on latency, reliability and real‑time control.
Many networks were never designed for this, and it shows.
This is where the world often underestimates the role of connectivity infrastructure. Because networks have become so commonplace, we assume they will simply adapt. Yet without the right type of network (cloud‑native, deterministic and AI‑ready) supply cannot reach demand at scale.
If AI is to be a genuine driver of growth, the objective must be to raise the ceiling of the addressable market as high as possible. That means connecting regions that are currently excluded, and upgrading networks where access already exists.
Otherwise, we risk celebrating a bandwagon that is racing ahead, while leaving both people and revenue behind.
The consequence of failure
U.S. network infrastructure is potentially not ready for AI, and we have only addressed mobile connectivity in this article.
You also must factor in energy supply and skills availability, not to mention accessibility of advanced compute capabilities or connectivity between data centers to enable Agentic AI to its fullest potential.
But what about everywhere else? If the U.S., one of the worlds’ most advanced connectivity markets, is potentially not AI-ready, it is a worrying sign for elsewhere.
Get the latest insights, download the AI will Break the European internet report
Get the latest insights, download the Build first, lead forever report
The International Telecommunication Union (ITU) estimates that currently 6 billion people can access connectivity, but scrutinizing these numbers one step further, the GSMA believes only 4.7 billion people access the mobile internet.
These numbers do not reflect the realities of enterprise connectivity or business adoption of AI, but they give a view on it.
Of those 4.7 billion, one should question how many live in digitally advanced nations, where AI skillsets are most likely to be, and how many live in a region of energy abundance or one that has local compute availability, or at the very least, edge-based infrastructure to enable AI inference at the edge.
When all the caveats are considered, the addressable market for AI shrinks.
This risk presents two potential consequences:
- The AI market does not have high enough economic potential. Therefore, the achieved rewards are minimal. Minimal rewards will dampen investments and ambition, which will reduce momentum for the AI supercycle.
- If access is not addressed and we continue this path, we will create an even wider digital divide than the internet era, one which is driven by access to intelligence and the economic benefits it offers.
Neither outcome is good enough. But there is still time to do something about it.
Neither outcome is good enough. But there is still time to do something about it.
Nokia’s view of the AI supercycle is simple: prosperity comes from foundations you can trust. That means connecting intelligence, not just people, with infrastructure built for AI: advanced, secure and resilient by design.
Every decade has had its “default” demand. The 1990s were voice and text. The 2000s brought the internet era, broadband, e-commerce and always-on connectivity. The 2010s were defined by smartphones, apps and streaming video.
Now the 2020s are being shaped by AI, and it’s different to what we have experienced in the past. It’s unpredictable, interactive and event driven. It demands lower latency, higher reliability, greater agility and end-to-end real-time feedback loops.
In other words, it’s not just “more network.” It’s a different kind of network.
And above all, it must be everywhere. If coverage doesn’t improve, access remains at a premium, and AI hits a glass ceiling. So does the economic potential that comes with it.
A supercycle built on patchy foundations isn’t a supercycle. It’s a stumble.
Subhagata Mukherjee (Subho) is a global leader at the confluence of technology and sustainable impact.
At Nokia, Subho is the Vice President & Global Head of Sustainability and leads the company’s global sustainability strategy, programs, and initiatives, including its overall Environmental, Social & Governance (ESG) responsibilities. Subho’s focus areas include climate, circularity, bridging digital divide, responsible supply chain, responsible use of technology, customer co-creation, innovation and ecosystem collaboration in sustainability
Subho holds chemical engineer from Jadavpur University, general management degrees from Asian Institute of Management, Nanyang Business School, and attended executive managements programs on Strategy, Innovation and Sustainability at Harvard Business School and MIT Sloan School of Management. Subho is also an advisor and member of several multilateral private-public partnerships focused on sustainable development.
