The Felicis Forecast

When you’re trying to understand change, it’s easy to reach for comfortable patterns.

But when you study it the way we have – up close, for more than 20 years, alongside the companies (opens in new tab) leading wave after subsequent wave of change – you learn that tidy bell curves and metaphors rarely apply.

But even for us, this moment feels different. This isn’t just another tech wave. We’re in a moment of structural inversion, where everything is up for grabs. AI is rewriting the rules of economics, redefining traditional models of labor, productivity, and value creation.

There is both noise and real magic happening right now. Discerning which is which takes a curious eye and a long-sighted view, because change doesn’t happen everywhere all at once. Instead, it starts along fault lines. These are the moments of fast rises and changing winners. Nothing is linear. Progress, growth, and descents all move at an unprecedented pace.

As if that weren’t enough, AI has arrived at a moment already rife with profound socioeconomic, organizational, and cultural changes. Where technological and behavioral changes converge, seismic transformations tend to follow.

At Felicis, we’re proud generalist investors. We believe neat taxonomies are better for understanding the way things have been rather than the way things will be. In a world where the lines and laws of every industry are being redrawn, we think generalists are better poised to spot and support the true outliers. Focusing too narrowly on specific domains risks missing the next frontier entirely.

So instead of following patterns, we forecast: we look at the deeper undercurrents of change that cut across categories, that we just can’t get out of our heads, and that seem to pop up everywhere once you notice them. We find they’re usually the ideas ambitious founders spend a lot of time thinking about, too. Consider this forecast our map of the fault lines where consumer, organizational, and socio-economic changes are colliding with AI’s potential.

AI started as magic. The $3 trillion opportunity is to make it ubiquitous.

Aydin Senkut, Sundeep Pechu, Peter Deng, Tobi Coker, Eric Flaningam

The current era of AI kicked off with a magic trick: a blinking cursor, a command, and voila – whatever text or image or line of code or whole product you’d asked for appeared. But economies and enterprises don’t run on magic. They run on big, complex, boring systems and processes. To permanently transform our economy, AI will have to get ubiquitous.

At the economic level, the opportunity is in the services – law, insurance, consulting – that make up nearly 80% of the United States GDP (opens in new tab) and that SaaS barely touched.

To take these fields on, we’re seeing AI shift from single-task completion to full workflows. These systems ingest the reams of context that real decision-making depends on: decades of PDFs, emails, contracts, and legacy databases. What comes next isn’t software for professionals but the service itself. Think AI-native law firms, insurance brokerages, insurance TPAs, tax advisory firms, and wealth management platforms. Pricing will shift from hours to outcomes, and the subsequent democratization of access to these services will give nimble, adaptable small businesses a significant advantage against well-resourced incumbents.

At large enterprises, AI will move past the pilot phase. Outside of engineering, enterprise teams have struggled to figure out how to use AI to take economically meaningful work off their plate. This isn’t because AI isn’t useful; it’s because these companies need it to be reliable, secure, and able to do very specific things that probabilistic reasoning hasn’t always been great at delivering. This is where the next wave of opportunity lies: the infrastructure that reliably and securely integrates AI agents with legacy data and systems that companies actually run on.

AI’s leaving the chatbox and reshaping our physical world.

Aydin Senkut, Peter Deng, Viviana Faga, Feyza Haskaraman, Eric Flaningam

AI’s physical implications are vast, and touch everything from manufacturing to energy infrastructure to consumer electronics. No one’s quite cracked the AI device market yet, but we think we’ll soon begin to see beautiful AI-powered devices that change how we live gain mass consumer appeal.

AI’s not just arriving in our pockets; it’s building the world and showing up on our maps. We believe three converging forces could finally kick off the beginning of the long-awaited American manufacturing boom. Models can now generate CAD designs and manage end-to-end hard goods production, while AI-powered robotics (opens in new tab) can perceive, manipulate, and move around the world in economically meaningful ways. Capital availability has made it possible to spin up full-scale industrialists that combine design, production, and supply chain management end-to-end from inception. And geopolitical incentives have only added fuel to the demand.

And then there’s the physical infrastructure it takes to power our models. In 2025, the total value of construction on data centers was nearly equal to construction on both warehouses and offices, driven by the world’s largest companies racing to double or even triple their data center capacity in the next four years. Monthly data center construction starts hit a record $25.2 billion in January 2026 (opens in new tab), and the major hyperscalers are collectively expected to spend $600–700 billion (opens in new tab) this year.

We think that the demand this will place on everything from construction permitting (opens in new tab) to energy production will accelerate innovation in industries that impact our ability to build – an area where we’ve been hungry for breakthroughs for decades now.

Knowledge is now a commodity, and that makes expertise invaluable.

Sundeep Peechu, James Detweiler, Peter Deng

Early foundational models scraped all of their data for free from the internet. All of that knowledge – the stuff of articles and blog posts and how-to-guides – is now a commodity, accessible and usable by anyone. AI, as it currently exists, has raised the floor on knowledge. Common knowledge has diminished in economic value, while expertise’s value has skyrocketed.

The evidence is all around us. We used to distinguish between researchers and builders (opens in new tab); now, we’re seeing the rise of the AI researcher-founder (opens in new tab). Teams are becoming more senior (opens in new tab). The followings of microauthorities have boomed on Substack, X, and LinkedIn. And education will need to be fundamentally reimagined to put mentorship at its core.

The opportunity now is for AI tools that accelerate human expertise by creating shortcuts to the depths of highly specialized knowledge. To move beyond the shallows of the free internet, AI will need access to living expertise on how decisions are made, not just what they look like. And this doesn’t have to just benefit the most senior or learned among us. We each develop unique expertise honed by the lives we lead; we just need to match that expertise to where it’s most economically useful.

We’re seeing this disposition towards expertise also change how we tackle our hardest technical problems beyond AI. It’s no longer enough to just dream it. It’s the founders who’ve spent enough time in deep study of the scientific minutiae of what it takes to put new industries in space or transform manufacturing lines with robotics who will push frontier industries forward.

Autonomous systems have redrawn the map of conflict and control.

Aydin Senkut, Jake Storm, Viviana Faga

In the past four years, the entire technological paradigm of defense has changed. Over 70% of the casualties and damages of war are committed by unmanned aerials (opens in new tab). Drones that cost thousands of dollars and take a day to build are destroying arsenals worth billions of dollars and accumulated over decades. The rise of autonomous systems have redefined warfare, surveillance, and national security. The urgency to build, manage, and defend against them is real: our borders, safety, and power hang in the balance.

Adapting to and winning in this new paradigm requires a full reimagining and rebuilding of the defense tech stack, from software to hardware, and detection to defense. And the consequences and opportunities reverberate far off the battlefield, as drone deliveries arrive in our cities: New York City recently piloted a cargo drone (opens in new tab)program to move goods between South Brooklyn warehouses and lower Manhattan.

The speculative economy is consuming the real one.

Peter Deng, James Detweiler, Eric Flaningam, Tobi Coker

For most people, “the economy” used to be an abstract thing. Money moved slowly and opaquely, and your investment choices were straightforward: stocks, bonds, and real estate. That worked okay, as long as most people felt like they were also benefiting from it. But creating wealth in the traditional economy has gotten more difficult for many Americans as middle and lower wages have stalled –despite rising productivity (opens in new tab)– and the entry point to homeownership has become unattainable.

Enter “the speculative economy”: everything from booms in cryptocurrency and retail investing to prediction markets and the explosion in creators and the tools they use to monetize. If the real economy is hands-off for all but a few gatekeepers, access and participation are what matter in this economy. Everyday people are creating entirely new wealth that’s more connected to the world they actually live in. (Plus, it’s fun.)

We don’t see this hunger for economic participation slowing down: as more people engage, the more interested they are in new forms of ownership. There’s an opportunity for startups that tap further into collective intelligence and open up new forms of engagement.

In every wave of change, the bottleneck is also the greatest opportunity.

Aydin Senkut, Jake Storm, Feyza Haskaraman

Behind every great wave of change, there’s the unglamorous backbone that actually makes it work. Globalized digital payments needed Adyen. E-commerce needed Shopify. Cloud needed CrowdStrike. Creators and influencers needed Canva. Remote teams needed Slack and Notion. The pattern’s the same: the gold rush gets headlines, but it’s the picks and shovels that determine who wins and how fast.

To find the enabling backbone of each wave of change, look for where things begin to bottleneck. For AI, that’s clearly security.

CEOs and CPOs love their new AI stacks. CISOs? Not so much (yet). Companies must deploy agents or risk falling behind, but no playbook exists for securing them. Agent identity security is fundamentally different from human identity. Every new agent and model compounds the problem. Security teams are fighting on two fronts: governing the AI tools their own people are adopting, while defending against a wave of faster, more sophisticated AI-powered attacks from outside.

AI security will require a trust, governance, and intelligence layer that sits across data, identity, and AI workflows. And the opportunity for the company that gets it right is even bigger than the one created by the shift to cloud.

Most people don’t want to DIY.

Aydin Senkut, Viviana Faga

There's a version of the AI story that goes like this: give everyone the tools, and they'll build what they need. The whole world vibecodes now, right?

Decades of purchasing data tell us this story is incorrect. Geoffrey Moore's decades-old framework (opens in new tab) still holds: The majority of the market looks nothing like the early adopters or the pragmatists. Most of the mainstream purchasers want a proven product with as little risk and friction as possible. They don’t want to experiment, and they don’t want to even interact with new software unless they absolutely have to.

That means for every Claude Code project you see on X, there’s a much larger market that will never open a code editor. Your dentist probably doesn’t want to Claude Code their way to inbox zero. Your plumber probably doesn’t want to vibecode a custom scheduling and invoice manager. And it’ll take more than a chat interface to get your accountant out of Excel.

The turnkey market is about to be reshaped by AI-native applications built by founders who truly understand these industries. AI creates an opening to build applications that don’t feel like software at all, but instead deliver seamless, end-to-end services. Founders who've lived in these industries can now build AI-applications that automate everything except the work that requires human hands. They can take a contractor's $1 million operation and turn it into a $10 million business, and to the buyer, it won’t look like software. It’ll look like the service they always wish they had.

The online world was built for humans. It will soon be rebuilt for agents.

Peter Deng, Tobi Coker, Feyza Haskaraman

There was a time when roads were built for horse drawn carriages. The ruts worn into stone paths determined the width of wagon wheels for centuries. When cars arrived, it was not enough to put an engine on a cart. The roads had to be repaved. The entire infrastructure of movement had to be rebuilt for a new kind of traveler.

Software is no different. It is full of human assumptions. Interfaces are meant to be seen and clicked. Workflows assume attention and memory. Search returns links because humans evaluate options. Every layer expects a person in the loop. Agents break that assumption. They do not browse or interpret. They execute. They require structured data, persistent state, and direct pathways to action. Search becomes retrieval, apps become execution environments, identity expands from a single user to fleets of agents operating with scoped permissions and durable state.

Software will stop being something you use and start being something that gets used on your behalf. The interface layer gives way to an action layer. This is one of the most significant infrastructure opportunities since the shift to cloud. Cloud required companies to rebuild how data was stored and served. The agent era requires rebuilding how software works at every layer, and this is where new generational companies will emerge.

We may actually figure out how the world works.

Sundeep Peechu, James Detweiler

Humans are a symbolic species: our understanding of the world has always been limited to and filtered through what we can translate into language. But so much of the world and the way it exists is beyond the bounds of language.

AI can sort through vast troves of data faster and more precisely than any single mind or organized group of people, but it can also do something bigger than that: it can act as an interpreter to parts of our existence we long ago accepted as mysteries. Its inputs don’t need to be in languages we understand: a token is a token is a token. It can create semantics and language where none currently exists, whether that’s the sound of a dog barking or EEG data.

We’re obsessed with how this single theme will change everything from daily communication (like silent speech) to longevity. Imagine, for example, AI ingesting troves of raw materials, mining them for pharmaceutical properties, and subsequently creating entirely new medicines. We now have a tool that finds the signal in noise once believed irreducible.