ESPOO, Finland — Nokia CEO Justin Hotard says the artificial-intelligence infrastructure boom is nowhere near finished, arguing that companies would build data centers roughly twice as fast if they could secure enough memory chips and electricity.
That is an unusually bullish message at a moment when investors are openly debating whether the world is spending too much money on AI.
Hotard told CNBC that he does not believe the industry is currently overbuilding because customers remain willing to deploy substantially more capacity than supply chains and power grids can support.
“If we could build 2x faster, our customers could build 2x faster, they probably would,” Hotard said, arguing that AI infrastructure remains in its early stages.
The statement does not mean Nokia expects global data-center construction to officially double.
It is a hypothetical description of pent-up demand.
But the underlying point is becoming increasingly important:
The biggest obstacle to the next phase of the AI boom may no longer be demand for GPUs. It may be whether the world can provide enough memory, networking, electricity and physical infrastructure to operate them.
Nokia has become an unlikely AI infrastructure winner
For many consumers, Nokia is still associated with mobile phones.
That business has not defined the company for years.
Today, Nokia is primarily a telecommunications and networking-equipment company selling infrastructure used by carriers, cloud operators, governments and increasingly AI data centers.
Under Hotard, who became CEO in April 2025 after running Intel’s Data Center and AI Group, Nokia has aggressively repositioned itself around what it calls the “AI supercycle.”
That shift is already showing up in the financial results.
Nokia reported that sales to AI and cloud customers jumped 105% year over year in the second quarter of 2026, reaching roughly €446 million.
Its Network Infrastructure sales rose 12% in constant currency, with Optical Networks growing 20% and IP Networks climbing 16%.
Comparable operating profit increased 18% to €434 million, beating the roughly €382 million analysts surveyed by LSEG had expected.
Those numbers suggest Nokia’s AI strategy is moving beyond corporate presentations and into actual revenue.
Why AI needs much more networking than traditional computing
A conventional corporate data center can often tolerate small amounts of network congestion.
AI training clusters cannot.
Large AI models are trained across thousands—or potentially tens of thousands—of expensive processors working together.
Those chips constantly exchange enormous quantities of data.
If the network becomes congested, a multimillion-dollar cluster of GPUs can sit partially idle while processors wait for information from one another.
Nokia argues that AI therefore changes the role of networking from background infrastructure into a critical performance component.
Its AI networking architecture focuses on:
high-capacity Ethernet switching;
ultra-low latency;
lossless data transfer;
automation;
optical interconnects;
and reliable communication between enormous clusters of accelerators.
That creates a potentially enormous opportunity.
Every additional AI processor requires not only computing power but also networking equipment capable of keeping that processor productive.
Nokia is betting Ethernet can challenge proprietary AI networks
One of the biggest technology battles inside AI infrastructure is emerging around how those processors communicate.
Nvidia built much of its AI ecosystem around high-performance proprietary networking technologies.
But major cloud and hardware companies are also backing Ultra Ethernet, an industry effort to adapt standard Ethernet for the extreme performance requirements of AI workloads.
Nokia is a member of the Ultra Ethernet Consortium and is building products around the technology.
The strategic appeal is obvious.
Ethernet is already widely used across global data centers.
If it can provide the performance needed for large-scale AI training, customers gain a more open ecosystem with equipment from multiple vendors.
For Nokia, that creates an opportunity to compete for networking dollars flowing into infrastructure that might otherwise remain dominated by a smaller group of specialized suppliers.
Nvidia itself invested $1 billion in Nokia
Perhaps the clearest validation of Nokia’s AI ambitions came from Nvidia.
In October 2025, Nvidia agreed to invest $1 billion in Nokia, purchasing about 166.4 million new shares and taking approximately a 2.9% stake in the Finnish company.
The companies also formed a strategic partnership covering both AI-RAN and data-center networking.
Nokia said the money would help accelerate development of its 5G and 6G software on Nvidia architectures and expand its AI and cloud networking business.
The two companies also agreed to explore integrating Nokia switching and optical technologies into future Nvidia AI infrastructure.
That partnership matters because Nvidia is simultaneously a potential competitor, supplier and ecosystem gatekeeper.
If Nokia’s equipment becomes embedded deeper inside Nvidia-powered data centers, the commercial opportunity could expand dramatically.
The partnership is already moving into telecom networks
The collaboration is not limited to conventional data centers.
Nokia is also developing AI-RAN, where cellular networks increasingly use the same accelerated computing architectures deployed for artificial intelligence.
In September, Nokia said operators including A1 Group, Chunghwa Telecom, du, e&, Mobily, stc, TPG Telecom and Zain Saudi were testing or deploying AI-RAN technologies using Nvidia platforms.
Nokia says early deployments have delivered more than 20% improvements in spectral efficiency, with additional software improvements planned in 2027 and 2028.
If that technology scales, Nokia could potentially benefit twice from the AI boom:
inside data centers;
and inside telecom networks carrying AI-generated traffic.
Nscale shows what Nokia’s data-center strategy looks like in practice
Nokia is already supplying networking systems to AI infrastructure operator Nscale.
At Nscale’s renewable-energy-powered facility in Stavanger, Norway, Nokia provides the data-center fabric used to connect large GPU clusters.
Nscale subsequently selected Nokia as a preferred networking partner for broader global AI infrastructure expansion.
The deployment includes:
data-center switches;
IP routing;
optical networking;
and programmable network software.
The significance is not the Norwegian facility alone.
It demonstrates Nokia’s attempt to become the networking layer connecting everything from GPUs inside a rack to data centers located in different regions.
That market could become enormous if AI compute continues expanding globally.
Hotard says memory is one of the biggest constraints
The irony is that AI companies are now consuming so much semiconductor capacity that the boom is creating shortages elsewhere.
Memory is especially important.
AI accelerators rely heavily on high-bandwidth memory and other advanced memory products.
Explosive demand has given manufacturers such as Micron, SK Hynix and Samsung enormous pricing power and encouraged suppliers to prioritize higher-value AI products.
The resulting tightness has spilled into the wider electronics market.
Nokia itself has acknowledged pressure from rising memory costs.
Reuters reported that the same shortage helping Nokia sell more AI networking equipment can simultaneously raise the cost of components used in its telecom products.
That produces one of the strangest dynamics of the AI boom:
AI demand can increase Nokia’s revenue while also making Nokia’s own products more expensive to manufacture.
Electricity may be an even bigger obstacle than chips
The second constraint Hotard highlighted is power.
This may ultimately become the more serious problem.
An AI data center containing hundreds of thousands of accelerators can consume enormous quantities of electricity.
Goldman Sachs recently increased its estimate of U.S. data-center capacity to around 64 gigawatts by the end of 2026 and projects approximately 90 GW by the end of 2027.
Yet in many regions, utilities cannot connect new facilities quickly enough.
Transformers can take years to procure.
Transmission projects face permitting delays.
New generating capacity takes time to build.
Some grid operators are already warning that large data-center requests exceed the capacity immediately available.
Morgan Stanley says these power constraints are now rippling into the semiconductor supply chain because delays in data-center energization can delay deployment of memory, optical components and other hardware.
That means owning GPUs is no longer enough.
Companies increasingly need to secure electricity years in advance.
Japan’s $15 billion AI project shows how important power has become
The emerging model can already be seen in Japan.
Power producer JERA recently partnered with Dell and AI infrastructure developer RHAELM on a $15 billion data-center project near Tokyo.
The planned facility is expected to reach 400 megawatts.
JERA is contributing something that may now be just as valuable as the computing equipment:
access to land and long-term electricity supply.
The project will be built near JERA’s Chiba thermal-power site, making the relationship between AI infrastructure and energy infrastructure unusually explicit.
This could increasingly become the template globally.
AI companies will not simply choose locations based on fiber connectivity and tax incentives.
They will choose locations where hundreds of megawatts of dependable electricity can actually be delivered.
Nokia believes current AI models alone can keep the boom running
Another important part of Hotard’s argument addresses concerns about whether AI development should slow.
Some AI executives and researchers have called for greater caution in developing increasingly powerful frontier models.
Investors worry that a regulatory or voluntary slowdown could undermine the hundreds of billions of dollars being spent on data centers.
Hotard argues the infrastructure boom could continue even without another dramatic breakthrough in frontier AI.
His reasoning is that companies have barely begun deploying existing models across the broader economy.
Businesses still need compute capacity for:
AI agents;
coding assistants;
customer service;
drug discovery;
financial analysis;
industrial automation;
robotics;
search;
enterprise software;
and inference.
If that view is correct, data-center demand does not depend solely on OpenAI, Anthropic or Google unveiling ever-larger models every year.
The commercial rollout of technology already invented could sustain spending for years.
The amount of money being committed is extraordinary
That bullish argument is supported by enormous spending plans.
Reuters reported that Anthropic alone has commitments totaling at least $518 billion over roughly a decade for cloud and AI infrastructure, much of it through agreements that cannot easily be canceled.
Meanwhile, major U.S. hyperscalers are expected to pour hundreds of billions more into AI infrastructure.
Goldman Sachs estimates U.S. hyperscaler investment could approach $1.1 trillion by 2027.
And HPE recently announced a $1.2 billion AI infrastructure order from cloud provider Vultr, while raising the long-term growth outlook for its networking business.
Nokia is therefore not chasing an imaginary market.
The spending already exists.
The unresolved question is whether the eventual revenue generated by AI will justify it.
That is where the AI-bubble argument begins
Hotard’s optimism directly challenges one of Wall Street’s biggest concerns.
Companies are spending extraordinary sums on AI.
But investors still do not know whether end users will generate enough profit to support those investments.
Oxford Economics estimates AI-related investment could total around $3.8 trillion between 2024 and 2028, requiring hundreds of billions of dollars in annual revenue simply to justify the capital deployed.
Credit investors are beginning to ask tougher questions.
Some debt associated with huge AI projects has weakened as lenders worry about:
GPU depreciation;
electricity availability;
customer concentration;
and whether future AI revenues will actually materialize.
That does not prove an AI bubble exists.
It shows that investors are becoming more demanding.
Nokia benefits even if individual AI models lose
This is one reason Nokia’s position is interesting.
The company does not necessarily need one particular AI laboratory to dominate.
Whether customers use OpenAI, Anthropic, Google, Meta or another model, those systems still require connectivity.
In investment terms, Nokia is trying to sell the picks and shovels of the AI infrastructure boom.
Networking equipment must connect:
GPUs to GPUs;
racks to racks;
buildings to buildings;
and data centers to users.
That makes Nokia somewhat less dependent on which AI application ultimately wins.
But it remains highly dependent on the overall level of infrastructure spending.
If the AI capital-expenditure boom slows materially, Nokia will feel it.
Europe could become Nokia’s biggest missed opportunity
Hotard has repeatedly warned that Europe risks falling behind the United States and China in AI infrastructure.
He said earlier this year that limited power availability, slower infrastructure investment and regulatory complexity were restricting European data-center development.
Data centers already account for roughly 3% of European Union electricity consumption, and AI is expected to push that figure higher.
For Nokia, this creates a strategic contradiction.
It is one of Europe’s most important networking companies.
Yet much of the most aggressive AI infrastructure investment is occurring elsewhere.
If Europe cannot expand its data-center footprint quickly enough, much of Nokia’s future growth may depend on U.S., Middle Eastern and Asian customers.
Nokia is also pushing into sovereign AI
Governments increasingly want critical AI infrastructure located inside their own borders.
That includes:
computing;
data storage;
networks;
cloud systems;
and eventually AI models.
Nokia has positioned its networking products around this sovereign AI trend, emphasizing open architectures and infrastructure that can be deployed independently of a single U.S. cloud provider.
That theme could become particularly important in Europe, the Middle East and Asia, where governments increasingly view AI computing capacity as strategic infrastructure similar to energy or telecommunications.
The same thinking is extending into space.
Nokia and Finland’s ICEYE recently announced plans to build sovereign low-Earth-orbit communications systems for governments and defense customers, with initial satellites expected from 2028.
Nokia still has to prove AI can become a large enough business
Despite the excitement, investors should keep Nokia’s current scale in perspective.
AI and cloud sales are growing rapidly.
But they remain a minority of total company revenue.
Nokia generated about €4.82 billion in Q2 net sales, while AI and cloud customer sales were roughly €446 million.
That means traditional telecom and network infrastructure still matter enormously.
The company cannot simply become an “AI stock” overnight.
It has to continue competing in:
mobile infrastructure;
IP routing;
optical networking;
fixed networks;
and carrier equipment.
Ericsson, Cisco, Arista, HPE, Broadcom and numerous specialist companies are all fighting for parts of the same infrastructure spending pool.
Competition in AI networking will become brutal
AI networking is valuable precisely because bottlenecks are expensive.
That will attract competitors.
Nvidia already offers its own networking technology.
Broadcom is deeply embedded in Ethernet switching and custom AI infrastructure.
HPE has strengthened its networking position and is winning large AI deals.
Arista Networks has become a major data-center networking provider.
Cisco is pushing deeper into AI infrastructure as well.
Broadcom recently raised its AI revenue forecasts, expecting roughly $115 billion of AI-chip revenue in fiscal 2027 and potentially $230 billion in 2028.
Nokia therefore has a major opportunity—but not an uncontested one.
Nokia’s advantage may be connecting everything, not owning the AI chip
Nokia does not need to beat Nvidia in GPUs.
Its strategy is to own more of the connective tissue surrounding them.
That includes:
inside-the-data-center Ethernet;
IP routing;
optical links;
connections between facilities;
telecom networks;
and eventually AI-enabled radio systems.
The more distributed AI becomes, the more important those connections become.
Training may happen in one giant cluster.
Inference may happen in regional data centers.
AI agents may operate across networks.
Telecom infrastructure itself may run AI workloads.
That creates a world in which connectivity becomes inseparable from computing.
Nokia is betting that this structural change can give it a second growth engine beyond traditional telecom infrastructure.
The “2x faster” claim should still be treated carefully
Hotard is the CEO of a company that directly benefits from faster AI infrastructure spending.
His confidence is therefore not neutral economic analysis.
It represents Nokia’s view of its customers and markets.
There is no independent proof that worldwide data-center construction would literally double if every memory and power constraint disappeared.
Some customers might accelerate spending.
Others could eventually pause because of financing costs, weak AI revenues or regulatory pressure.
The statement is best understood as evidence that Nokia currently sees more customer appetite than available supply can satisfy.
That distinction matters.
But independent data support the bottleneck argument
Even with that caveat, other industry evidence points in the same direction.
Goldman sees U.S. data-center capacity continuing to expand strongly through 2027 despite growing political opposition.
Morgan Stanley says electricity constraints are already affecting AI hardware deployment.
HPE is winning billion-dollar AI infrastructure orders.
Anthropic has committed hundreds of billions to computing capacity.
And Nokia’s own AI and cloud sales more than doubled in its latest reported quarter.
Taken together, those figures suggest the AI infrastructure boom remains real.
What remains uncertain is its ultimate return on investment.
The next phase of AI may be less about smarter models and more about physical infrastructure
For the past several years, the AI race was dominated by questions such as:
Which model is smartest?
Who has the most GPUs?
Which chatbot has the most users?
The next phase may be much more physical.
Who has enough power?
Who has enough memory?
Who can secure transformers?
Who can build transmission lines?
Who can connect 100,000 processors without bottlenecks?
And who can move enormous quantities of data between those processors reliably?
That shift plays directly into Nokia’s strategy.
The Finnish company is betting that once AI leaves the research lab and becomes part of everyday business infrastructure, networking becomes just as essential as computing.
Hotard’s claim that data centers could be built twice as fast should not be treated as a literal industry forecast.
But it does capture the central tension in today’s AI boom:
Companies still want more computing capacity than the physical world can currently provide.
And if that remains true, Nokia’s biggest opportunity may no longer be connecting people to mobile networks.
It may be connecting the machines powering the AI economy itself.
