The investment landscape is undergoing a noticeable thematic shift. Capital is increasingly moving away from the themes that dominated the previous cycle and toward industries tied to AI, infrastructure, defense, nuclear energy and space.
At the same time, the economic effects of the AI buildout are becoming visible beyond technology companies. Data centers are creating demand for skilled labor, transportation platforms are becoming more expensive, and AI agents are beginning to change how enterprises consume computing resources.
The broader message is that the AI economy is no longer confined to software. It is reshaping physical infrastructure, labor markets, consumer services and enterprise technology simultaneously.
ETF Themes Are Moving From Bits to Atoms
ETF demand remains strong, with net inflows on pace for a record year and July setting a new monthly high.
What is particularly interesting, however, is how dramatically investor themes have changed.
In 2020, some of the most popular thematic areas included clean energy, emerging-market technology and healthcare. By 2026, the leadership has shifted toward AI, nuclear power, space, defense and infrastructure.
That transition reflects a broader change in investor priorities.
The previous generation of technology themes was largely centered on software, platforms and digital services. Today’s dominant themes increasingly involve enormous physical investments: power generation, semiconductor manufacturing, data centers, defense systems and space infrastructure.
Whether every one of these themes ultimately produces attractive returns remains uncertain. But the investment landscape has clearly become more capital-intensive.
Data Centers Are Creating a Skilled-Labor Boom
Data centers have become one of the most important sources of new construction activity in several U.S. states.
In some regions, their impact is extraordinary.
New Mexico and Wyoming, for example, have relatively small data-center construction pipelines, yet those projects represent a surprisingly large share of private non-residential construction because the states have comparatively limited overall construction activity.
Pennsylvania provides another example. A data-center pipeline of roughly 3 gigawatts represents a significant portion of the state’s non-residential construction spending. Texas is building substantially more capacity, but because its broader construction economy is much larger, data centers represent a smaller percentage of total activity.
The important point is that data-center investment is becoming a meaningful regional economic force.
Data Centers Are Boosting Local Employment
The economic effects extend beyond the facilities themselves.
Research examining counties with operating data centers has identified several favorable trends, including:
- Stronger employment growth
- Lower unemployment
- Rising housing activity
- Higher home values
Areas where facilities are still under construction show particularly strong employment effects, although housing outcomes are more mixed.
These relationships should not automatically be interpreted as proof that data centers caused every improvement. Many facilities are concentrated in already prosperous regions such as Northern Virginia, while Texas entered the recent data-center boom after years of unusually strong residential construction.
Nevertheless, the labor-market effect is difficult to ignore.
Data centers require electricians, concrete workers, equipment operators, engineers, technicians and other skilled trades. And those workers are increasingly commanding significant wage premiums.
In some cases, data-center projects are paying dramatically more than traditional construction employers for comparable skills.
That creates an important second-order effect of the AI boom: the demand for computing infrastructure is increasing the value of physical labor required to build and maintain it.
The AI economy may therefore be creating one of the largest new opportunities for skilled trades in years.
Ridesharing Is Getting More Expensive
The feeling that rideshare trips cost more than they used to appears to be supported by the data.
Since 2024, both the average and median Uber fare have increased by roughly 20%. Lyft has followed a different trajectory, with fares remaining somewhat below early-2024 levels, although prices have recently started moving higher.
A major factor appears to be platform fees.
Uber’s platform fees have increased substantially over the past year, while Lyft’s fee structure has also begun moving higher after previously declining.
The change is not necessarily negative for drivers.
Average gross earnings per trip have also increased and recently reached record levels. Higher consumer prices are therefore flowing through at least partially to the people providing the service.
The long-term question is whether higher fares eventually reduce demand. For now, however, the rideshare market appears to be supporting both higher platform revenue and stronger driver economics.
Social Commerce Is Growing Even Faster
The broader gig economy is also changing.
While more people are participating in various forms of independent work, social commerce stands out as one of the fastest-growing categories.
The model resembles a digital version of television shopping: creators use social platforms to showcase products, build audiences and drive transactions.
Several forces could explain the growth.
Social media has become an increasingly important distribution channel for commerce, while improvements in AI tools may be reducing the cost of producing content, managing operations and running small online businesses.
Better targeting and recommendation systems could also make it easier for creators to reach potential customers.
The result is a new form of entrepreneurship in which audience, commerce and technology increasingly overlap.
AI Usage Is Becoming Increasingly Uneven
AI adoption is growing across businesses, but usage is far from evenly distributed.
The gap between typical enterprises and the most advanced AI users is expanding rapidly.
Enterprise AI output has increased substantially, but the highest-consuming companies are pulling away from the rest. Among leading organizations, token generation has risen dramatically over the past year.
This suggests that AI adoption is moving through a familiar pattern: experimentation begins broadly, but a smaller group of companies eventually discovers applications capable of generating much deeper productivity gains.
The most advanced organizations are also moving beyond simple chatbot interactions.
They are increasingly using:
- Plugins and specialized tools
- AI skills
- Coding agents
- Automated workflows
- Autonomous or semi-autonomous agents
In other words, the leading edge of enterprise AI is shifting from asking AI questions to giving AI systems work to perform.
Legal Teams Are Becoming Major AI Power Users
Technology companies are not the only organizations increasing their use of advanced AI tools.
Knowledge workers across industries are adopting coding and agentic systems, with legal professionals showing particularly rapid growth in recent months.
The trend demonstrates how quickly AI capabilities can move from technical departments into traditionally non-technical professions.
As these tools become easier to deploy, the definition of an AI user is also changing. A lawyer, analyst or financial professional no longer needs to be an AI engineer to make use of increasingly sophisticated systems.
AI Agents Are Changing Compute Economics
Perhaps the most important development is happening beneath the surface.
Only a relatively small share of AI users currently operate fully deployed agents, but those agents already account for a disproportionately large amount of AI compute consumption.
Agentic systems can generate several times more token usage than conventional human interactions, and their consumption has grown rapidly this year.
The reason is structural.
A human typically interacts with an AI system through a relatively simple sequence:
Prompt → Response
An agent works differently:
Goal → Plan → Execute → Review → Iterate → Execute again
The system repeatedly accesses context, performs actions and updates its working memory until it reaches an objective.
That produces much higher token consumption.
Cached Context Is Becoming Critical
A large share of agentic AI consumption comes from cached context rather than entirely new prompts.
This is economically important because cached tokens are substantially cheaper than repeatedly processing the same context from scratch.
But there is a trade-off.
Agents need to retain and access increasingly large amounts of information as they work through complex tasks. That makes memory infrastructure particularly valuable.
This helps explain why high-bandwidth memory and other forms of advanced memory capacity have become critical components of the AI hardware supply chain.
The AI infrastructure story is therefore expanding beyond GPUs.
As agents become more capable, memory, networking, storage and data-center capacity all become increasingly important.
Agents Could Disrupt Legacy Automation Software
There is another potential consequence of agent adoption.
Traditional workflow-automation platforms were designed around predefined processes. Users manually configure triggers, actions and integrations.
AI agents can approach the same problem differently.
Instead of defining every step in advance, users can increasingly describe the desired outcome and allow an agent to determine how to accomplish it.
Early traffic data already suggests that some traditional automation platforms are facing pressure.
Established services such as Zapier, Make and n8n have experienced declines in web traffic in recent periods, while newer AI-native automation platforms have gained traction.
It would be premature to conclude that traditional automation software is disappearing. Established providers have enormous customer bases and are incorporating AI into their own products.
But the competitive model is changing.
The question is no longer simply:
“Can AI automate this workflow?”
It is becoming:
“Does the user still need to build the workflow at all?”
That distinction could become increasingly important as agents become more capable.
The Bigger Picture
Several seemingly unrelated trends are beginning to converge.
Capital is moving toward physical AI infrastructure.
Data centers are reshaping regional labor markets.
Skilled workers are commanding higher wages.
Consumer platforms are changing their pricing models.
Social commerce is expanding the gig economy.
Enterprise AI usage is becoming increasingly concentrated among power users.
Agents are dramatically increasing compute and memory requirements.
Traditional automation software is facing a new competitive model.
The AI transition is therefore becoming much broader than the development of increasingly powerful models.
It is changing where capital goes, what workers get paid, how businesses operate, how consumers buy services and what infrastructure the economy requires.
The most interesting phase may be just beginning.
