Animal Spirits: Is the AI Trade Over?

Wall Street loves a grand narrative almost as much as it loves pretending it discovered that narrative five minutes before everybody else. In 2023 and 2024, the artificial intelligence trade became the market’s favorite blockbuster: chips, cloud, data centers, networking, software, consultants, and probably someone trying to sell AI-enhanced staplers. If it even looked vaguely adjacent to machine learning, investors treated it like it had just been handed the keys to the future.

Now the mood is different. The slogans are still loud, but the easy confidence is gone. Big Tech is spending stunning amounts on infrastructure, software stocks have been whacked by fears of disruption, and investors are asking the least sexy question in finance: where, exactly, is the return on all this spending? That question matters because “animal spirits” can push a trade higher for a while, but eventually the market wants invoices, margins, and something more persuasive than a keynote with dramatic lighting.

So, is the AI trade over? Not really. But the version of the trade that rewarded almost every company for merely whispering “AI” into a microphone is fading fast. What remains is more complicated, more selective, and frankly more interesting.

The Short Answer: No, but the easy-money phase probably is

The AI trade is not dead. It is maturing, splintering, and becoming pickier than a venture capitalist at a free cold brew bar. That distinction matters. A dead trade means demand collapses, spending dries up, and leaders lose their strategic importance. That is not what is happening. The biggest technology companies are still pouring historic sums into servers, data centers, networking gear, and AI software. Nvidia is still producing massive data center revenue. Microsoft still says demand is exceeding supply in key areas. Alphabet, Meta, and Amazon are still in full build mode. That does not look like a funeral.

What has changed is the market’s tolerance for fantasy. Investors are no longer rewarding every AI story equally. Hardware enablers still look powerful. Cloud platforms still look essential. But plenty of software names are being forced into a brutal identity check: are they going to use AI, or get used by it? That is a much harsher game.

Why people think the AI trade is over

Capex sticker shock is real

The first reason is simple: the bills got huge. Alphabet reported full-year 2025 capital expenditures of $91.4 billion. Meta spent $72.22 billion in 2025 and guided even higher for 2026. Amazon’s AI buildout has also become enormous, while Microsoft’s quarterly capital spending has ballooned as it races to support Azure, Copilot, and large model demand. Suddenly, the market is staring at a world where hyperscalers are not just “investing in innovation.” They are spending like they are trying to terraform the internet.

That creates two worries at once. The first is profitability. If revenue monetization lags infrastructure spending, margins get squeezed. The second is duration risk. Investors can tolerate giant spending when the payoff feels immediate. They get twitchy when management says, in effect, “Trust us, this giant pile of GPUs will make sense later.” Markets are supportive right up until they are not.

Valuation gravity finally showed up

Another reason for the skepticism is that many AI winners were priced for near-perfection. Once a stock reaches a valuation that assumes heroic execution, all future earnings reports become hostage negotiations. Even a strong quarter can disappoint if it is not strong in exactly the right cinematic way. That is why some AI leaders have looked less like unstoppable juggernauts and more like Olympic gymnasts who stuck the landing but still got marked down for a tiny wobble.

In other words, the market is no longer asking, “Is AI big?” Everyone already knows it is big. The market is asking, “Which companies capture the value, how quickly, and at what cost?” That is a much more demanding exam.

Software got hit with existential dread

Early in the AI boom, software was often treated as a natural winner. That thesis made sense at first: embed generative tools into workflows, charge more, expand seats, and call it innovation. Then the mood shifted. Investors began to worry that AI might not just help software vendors. It might erode their moats, flatten pricing power, or make some products easier to replace.

That fear triggered a painful reset across parts of the software market. Suddenly, companies were sorted into new buckets: AI beneficiary, AI adapter, or AI roadkill. That may be too dramatic, but only slightly. The broader point is that the AI trade stopped being a unified bet and became a sorting machine.

Why the AI trade is not over

Demand for compute is still enormous

If you want the simplest reason the AI trade is still alive, look at the demand for compute. Nvidia’s fiscal 2026 results showed just how strong AI infrastructure spending remains, with full-year revenue reaching new highs and data center revenue doing the heavy lifting. Microsoft has said customer demand continues to exceed supply. Amazon is still talking about a much larger future for AWS driven by AI workloads. Meta keeps signing large infrastructure deals. This is not the behavior of an industry stepping away from the table. It is the behavior of an industry still ordering dessert after a five-course meal.

And there is an important nuance here: inference is now joining training as a serious demand engine. The market’s first AI chapter focused on building gigantic models. The next chapter is about using them continuously, inside enterprise workflows, search, advertising, developer tools, customer support, and agentic systems. Inference creates recurring infrastructure demand, and recurring demand is the sort of thing investors usually enjoy once they stop hyperventilating.

Picks and shovels still have receipts

The “picks and shovels” part of the AI trade remains the cleanest expression of the theme. Chips, memory, interconnects, cooling, optical components, and networking continue to benefit because they sit upstream from the monetization debate. They do not need every enterprise AI app to become a home run tomorrow. They mostly need the hyperscalers and model builders to keep building, and so far they are.

That is why companies tied to AI infrastructure have generally held up better than many application-layer names. Broadcom’s AI semiconductor momentum remains strong. AMD is scaling its data center AI franchise. Networking and server suppliers continue to matter because AI is not just a model problem; it is a system problem. The market may argue about who wins the app layer, but it still needs the pipes, the silicon, and the electricity.

Monetization is slower than hype, but it is not imaginary

One common mistake in bear cases is assuming that because AI monetization has been uneven, it must therefore be fake. That is too simplistic. Monetization is showing up, just not in one neat line item with fireworks around it. Microsoft has real Copilot and Azure AI revenue traction. Alphabet is seeing AI demand across Cloud and its broader product ecosystem. Meta is using AI to improve ad targeting, engagement, and content ranking, which matters because boringly better ads are still very profitable ads. Amazon is embedding AI across AWS and commerce operations. None of that means every dollar of capex will earn a glorious return. It does mean the revenue side of the equation exists.

The more realistic debate is about timing. The market fell in love with AI quickly, but large platform monetization usually arrives in layers. Infrastructure gets built first. Products get adopted next. Pricing power stabilizes later. Margin expansion shows up after that. Investors who expected the entire cycle to complete before lunch were always going to be disappointed.

What changed: the AI trade became several different trades

At the start of the boom, people talked about “the AI trade” as if it were one giant basket. That framing is now too crude. There are really several overlapping trades.

The first is AI infrastructure: semiconductors, networking, servers, memory, and the companies that supply the physical backbone. This bucket still looks strong because the demand is visible and the bottlenecks are tangible.

The second is cloud and platform monetization: Microsoft, Amazon, Alphabet, and selected model providers. This bucket depends on turning usage into durable revenue while managing giant capital outlays. It is still attractive, but investors are scrutinizing every margin point like detectives in a procedural drama.

The third is enterprise software. This is where the stress is highest. Some companies will use AI to deepen their moat, automate workflows, and justify premium pricing. Others will discover that their once-luxurious software bundle is really just an expensive wrapper around tasks AI can now perform more cheaply.

The fourth is second-order beneficiaries: power generation, grid equipment, cooling systems, data center real estate, and cybersecurity. These areas may not get all the glamour, but glamour is overrated. Cash flow pays the rent.

What could actually kill the AI trade?

If the AI trade were truly going to break, the trigger would probably come from one of four places.

The first would be a clear collapse in return on capital. If hyperscaler spending keeps rising while customer monetization stalls, the market will stop granting the benefit of the doubt. At some point, “investment phase” starts sounding less like strategy and more like a polite excuse.

The second would be supply-side constraints that become economically destructive rather than temporarily inconvenient. Power shortages, data center bottlenecks, export restrictions, and component delays can all slow deployment and distort margins. Growth stories hate friction, and AI still has plenty of it.

The third would be macro conditions. Higher rates, credit stress, or a broader equity drawdown can compress valuations across the board, even for strong structural themes. Sometimes a great story gets punched in the face by the bond market.

The fourth would be commoditization. If models become cheaper and more interchangeable faster than expected, some AI companies may discover that being technologically impressive is not the same thing as having durable economics. The market has started to price that risk more aggressively in software, and it may keep doing so.

So, is the AI trade over?

No. But the lazy version of it probably is.

The market is moving from wonder to accounting. From “AI changes everything” to “show me who gets paid.” From a broad momentum wave to a narrower competition over economics, scale, distribution, and staying power. That shift can feel bearish if you got used to every AI headline lifting every related stock. In reality, it is what a real long-term theme looks like after the honeymoon ends.

That is why the right answer is not that the AI trade is over. It is that the trade has graduated. It is older, crankier, more expensive, and less likely to applaud nonsense. Honestly, good for it.

In 2026, the question is no longer whether AI matters. It clearly does. The question is which businesses turn that importance into durable earnings without setting their free cash flow on fire. Investors who can separate infrastructure necessity from software wish-casting will probably do fine. Investors who still buy anything with an “AI strategy” slide deck may be funding someone else’s learning experience.

What the last two years felt like for investors

For investors, the experience of living through the AI trade has been equal parts exhilaration, confusion, greed, fear, and the occasional urge to throw a valuation model out the window. At first, it felt easy. You did not need a deep theory. You only needed to recognize that generative AI had broken into the mainstream and that compute was suddenly the most glamorous commodity on Earth. If you owned the obvious winners, you looked brilliant. If you missed them, you spent a lot of time pretending you were “waiting for a better entry point,” which is investor dialect for “I am annoyed.”

Then came the second phase, and this is where the emotional texture changed. The headlines got bigger, but the certainty got smaller. Every earnings season became a referendum not just on revenue, but on destiny. A company could beat expectations and still get punished because its capex was too high, its AI revenue disclosure was too vague, or management sounded a little too excited in that suspicious way executives do when they are asking shareholders for patience and another truckload of money.

There was also the strange psychological whiplash of watching the market love AI in one corner and fear it in another. Investors cheered the chipmakers selling the tools, then panicked about the software companies that might be replaced by the tools, then rediscovered that some software companies could use the tools to become stronger. It was less like a clean trend and more like a rotating cast of heroes and suspects. One week, infrastructure was king. The next week, software was washed. Then software bounced, semis wobbled, and everyone rediscovered the word “selectivity” as if it had been hidden in a cave.

Another part of the experience has been learning that AI is not just a technology story. It is also a power story, a balance-sheet story, a labor story, and a patience story. Investors had to think about electricity demand, data center leases, export controls, debt loads, depreciation, and whether a chatbot feature actually creates revenue or merely makes a demo more entertaining. That is a lot of homework for a trade that originally looked like a simple momentum rocket.

And yet, despite the volatility, many investors came away with the same conclusion: this theme is too important to ignore, but too complicated to buy blindly. That may sound obvious now, but markets specialize in making obvious truths feel revolutionary about six months late. The real experience of the AI trade has been growing up alongside it. What began as an adrenaline rush became a discipline test. What looked like a one-click bet turned into a messy evaluation of business models, moats, capital intensity, and managerial credibility.

That is probably the healthiest outcome. Durable trends should survive scrutiny. The AI trade has been forced to endure exactly that, and while plenty of stocks have been bruised in the process, the broader story remains intact. Investors are not dealing with the end of AI enthusiasm. They are dealing with its adulthood. And adulthood, as always, involves fewer fantasies, more bills, and a much sharper eye for who is actually carrying the weight.

Conclusion

The AI trade is not over. It is simply no longer a costume party where everyone wearing a silicon badge gets a trophy. The new phase belongs to companies that can prove demand, absorb spending, protect margins, and turn AI from an exciting capability into a durable business. In that sense, animal spirits are still alive. They are just being supervised now.