Wall Street Dumped Chip Stocks and Piled Into Software, and Agentic AI Is the Reason

The Great AI Handover

Something remarkable happened on Wall Street this week. For the first time since the AI trade began in early 2023, investors are aggressively pulling money out of the semiconductor stocks that built the AI boom and pouring it into the software companies that are finally proving they can turn AI into actual revenue. The rotation is being called "The Great AI Handover," and the numbers are striking: the SOXX semiconductor ETF has given back significant gains while the software-heavy IGV has begun clawing back from a brutal 25-30% decline from its 2025 highs.

The trigger isn't a single event. It's the convergence of three forces: Nvidia's "sell-the-news" earnings reaction (down 5.5% on a blowout quarter), Salesforce demonstrating 50% quarter-over-quarter growth in agentic AI deals, and a broader realization that the AI value chain is shifting from "who makes the chips" to "who makes the chips useful."

From Chips to Code

The first phase of the AI investment cycle was simple: companies needed GPUs to train and run AI models, so you bought Nvidia. That trade worked spectacularly from 2023 through early 2026, turning Nvidia into a $4 trillion company. But phase two is different. The infrastructure is largely built (or at least the first wave is), and the question has shifted from "can we build the compute?" to "can we generate returns on the compute?"

Software companies are answering that question. Salesforce closed over 22,000 Agentforce deals in Q4, with combined annual recurring revenue for Agentforce and Data Cloud reaching approximately $1.8 billion. ServiceNow's "Now Assist" agentic AI product passed $600 million in annual contract value. These aren't pilot programs or proof-of-concepts. They're production deployments generating real subscription revenue.

The spread between semiconductor and software performance hasn't been this tilted toward software since the DeepSeek-driven AI unwind 13 months ago. Investors who rode Nvidia from $50 to $190 are now looking at software as the next leg of the AI trade.

What "Agentic AI" Actually Means for the Market

The word "agentic" gets thrown around a lot, but the market rotation is grounded in a specific technological shift. Traditional AI tools (chatbots, copilots, summarizers) augment human work. Agentic AI replaces human work. An AI agent doesn't just answer a customer service question; it resolves the ticket, updates the CRM, escalates if needed, and follows up. It doesn't just draft an email; it monitors your inbox, prioritizes messages, schedules meetings, and executes workflows.

This distinction matters enormously for software business models. Under the old "per-seat" licensing model, AI that automates human tasks is a threat: fewer seats, fewer licenses, less revenue. That's what crushed software stocks in the "SaaSpocalypse" of early February, when the Nasdaq plunged over 4% in a single session on fears that AI would hollow out the SaaS business model.

But the companies that survived the SaaSpocalypse are the ones pivoting to outcome-based pricing. Salesforce introduced the "Agentic Enterprise License Agreement," shifting from charging per human login to charging per "Agentic Work Unit," essentially billing for the work an AI agent does rather than the humans who use the software. It's a fundamental business model transformation that turns AI from a revenue threat into a revenue multiplier.

The Winners So Far

The rotation has clear winners. Salesforce surged after demonstrating that its Agentforce platform is gaining traction at enterprise scale. Palantir Technologies has been a standout, with U.S. commercial revenue growing 137% in recent quarters, driven by its AI platform deployments. Cadence Design Systems rallied after releasing what it calls the world's first agentic AI solution for chip design, claiming a 10x improvement in engineering productivity.

ServiceNow is the other heavyweight. Its Now Assist platform has crossed the $600 million ACV threshold, proving that IT service management, one of the most boring corners of enterprise software, is fertile ground for AI agents. The pitch is straightforward: instead of a human reading a support ticket, categorizing it, routing it, and resolving it, an AI agent does all four steps in seconds.

On the losing side, generic SaaS companies without an AI agent strategy are getting hammered. The market is bifurcating between "AI-native software" (companies that have built AI agents into their core product) and "AI-vulnerable software" (companies whose products could be replaced by AI agents). The gap between these two groups is widening every week.

Why Nvidia Fell on a Blowout

The Nvidia sell-off is central to understanding this rotation. On Thursday, Nvidia reported $68.1 billion in quarterly revenue (beating estimates by $2 billion), guided to $78 billion for next quarter (beating estimates by $5.4 billion), and unveiled the Rubin chip platform with 10x inference cost improvements. The stock dropped 5.5%, its worst day since April.

The drop wasn't about Nvidia's results. It was about what happens next. If Rubin delivers 10x cheaper inference, the cost of running AI workloads plummets. That's great for software companies deploying AI agents (their costs fall dramatically) but raises questions about the pace of future GPU purchases. Hyperscalers who are currently "depleting their cash flows" on AI capex, as one analyst put it, might slow down purchases if each new chip is 10x more efficient than the last.

The VanEck Semiconductor ETF fell 3.3% in sympathy. Broadcom, Lam Research, Western Digital, and Applied Materials all slid. The market is sending a clear message: the easy money in AI semiconductors has been made, and the next wave of value creation is in the software layer.

Google's Contribution: Gemini 3.1 Pro

Adding fuel to the software rotation, Google released Gemini 3.1 Pro on February 19, achieving a 77.1% score on the ARC-AGI-2 benchmark, more than double the reasoning performance of its predecessor. It also hit 80.6% on SWE-Bench Verified, a benchmark for AI-assisted software development.

These benchmarks matter because they measure exactly the capabilities that power agentic AI: solving novel problems, maintaining context across multi-step tasks, and writing production-quality code. Better models mean more capable agents, which means more compelling products from the software companies building on top of these models. Google's model improvements flow directly into the revenue streams of every company building agentic applications.

What to Watch

The rotation from chips to software is still in its early stages, but the direction is clear. Enterprise AI spending is projected to accelerate by 14.7% in 2026, and the bulk of that spending is shifting from infrastructure (GPUs, data centers) to applications (agent platforms, workflow automation, AI-native software).

The key metric to track is agentic deal volume at the major software companies. Salesforce's 22,000 Agentforce deals and ServiceNow's $600 million ACV are the benchmarks. If those numbers accelerate in Q1 earnings reports (due in May), the rotation has legs. If they plateau, the market will question whether agentic AI is another overhyped category.

For investors, the takeaway is that the AI trade hasn't ended. It's evolved. The value is migrating from the companies that build the compute to the companies that monetize it. Wall Street is repricing that shift in real time.