Ethan Gordon, CMT@EthanGordon1
The HARDWARE vs SOFTWARE Trade
by Ethan Gordon
For the better part of a decade, software was king.
Recurring revenue. Fat margins. Sky-high multiples. Software companies were the darlings of Wall Street and for good reason. They scaled without factories, printed cash without inventory, and compounded like clockwork.
Then, quietly, the market flipped the script.
In 2025, while most investors were still debating which software company would dominate the AI era, the money had already moved. It moved into chips. Into data centers. Into the physical infrastructure of artificial intelligence. And the numbers tell a story that is hard to ignore.
$SMH (VanEck Semiconductor ETF): +52.63% YTD 2026
$IGV (iShares Software ETF): −16.93% YTD 2026
That gap, nearly 70 percentage points, is not noise. It is a signal. And understanding what is driving it, where it goes next, and what it means for your portfolio is what this month's newsletter is all about.
Part One: Why Hardware Won
The answer comes down to one thing: follow the capital.
The hyperscalers: Amazon, Google, Meta, and Microsoft, have collectively committed up to $630 billion in capital expenditures for 2026. That is a 62% increase from the record $388 billion spent in 2025. Roughly 75% of that aggregate spend is flowing directly into AI infrastructure: GPUs, servers, data centers, and the networking equipment to tie it all together.
That money does not land in Salesforce's pocket. It lands in Nvidia's. In Broadcom's. In AMD's. In TSMC's. The companies making the chips and building the physical backbone of AI have been the direct and immediate beneficiaries of the largest technology investment cycle in history.
The Morningstar U.S. Semiconductors Index gained 42.8% in 2025, more than double the broader market. Nvidia alone added 6 percentage points to overall market gains. Broadcom contributed another 2.7 points.
The hardware trade was real, it was well-supported by fundamentals, and it has largely played out. Which brings us to the next chapter and the one where I think the most interesting opportunity now lives.
Part Two: Why Software Got Punished
The selloff in software stocks was not random. Investors had two very specific fears, and both of them are worth understanding clearly because one of them is legitimate, and one of them is being massively overapplied.
Fear Number One: AI Will Rebuild What Software Companies Built and Take Their Customers
The first fear is straightforward. Investors began asking a simple question: if AI can do what a software company does, why would anyone keep paying for the software?
Take Duolingo as an example. Duolingo charges users a monthly subscription to help them learn a new language through structured lessons and exercises. But today, you can open ChatGPT or Claude and have a full conversation in Spanish, get corrected in real time, and learn faster than any app could teach you. The core product Duolingo sells is being replicated, for free, by general purpose AI. That is a genuine problem for a company like Duolingo, and the market was right to reprice that risk.
Or take a smaller website builder. If OpenAI's Codex or another AI coding tool can build you a professional website in twenty minutes without any technical knowledge, the demand for simpler website-building platforms with limited functionality starts to erode. Again, a fair concern.
This fear, applied to the right companies, is legitimate. Software businesses with shallow moats, simple functionality, and no deep customer integrations are genuinely at risk of being disrupted or commoditized by AI.
But here is where the market made its mistake.
It applied that same fear to every software company regardless of size, complexity, or how deeply embedded the product is inside its customers' operations. And that is where the logic breaks down completely.
Consider Microsoft. Microsoft Office is used by 82% of Fortune 500 companies. Excel, Word, PowerPoint and Outlook are not just software tools. They are the operating language of global business. When you send a client an Excel model, the probability that they already own Microsoft and know exactly how to open it, edit it, and send it back is extremely high. The entire workflow, the format, the formulas, the compatibility, the institutional familiarity, everything is built around Microsoft.
ChatGPT is not going to build a competing spreadsheet application and convince the Fortune 500 to abandon Microsoft Office and retrain their entire workforce on something new. The switching cost alone makes that scenario nearly impossible. And that does not even account for the compliance requirements, IT infrastructure, and enterprise agreements that tie large organizations to Microsoft for years at a time.
The fear that AI destroys Duolingo is reasonable. The fear that AI destroys Microsoft Office is not. But in 2025, the market sold both stocks like the threat was identical. That indiscriminate selling is what created the opportunity we are looking at today.
Fear Number Two: AI Agents Will Compress the Number of Seats Companies Need
The second fear is more subtle but equally important to understand. Even if AI does not replace a software platform entirely, investors worry that it would reduce how many human users companies need to pay for.
The traditional SaaS business model charges per seat, meaning per human user. A company with 5,000 employees might pay for 5,000 Salesforce licenses. But if AI agents can now handle a significant portion of the work those employees used to do, the company might only need 3,000 human users doing active work in the platform. That means fewer seats, which means less revenue for the software company.
This is the seat compression argument, and it contributed meaningfully to the selloff across the SaaS sector.
Both fears together created what Wall Street started calling the "SaaSpocalypse." And while neither fear is entirely wrong, both are being priced into software stocks in a way that ignores the much larger counter-story that is already unfolding.
Part Three: The Market's Bluff Gets Called
Both fears have now been directly addressed by one of the most credible voices in technology. In a widely circulated CNBC interview and at ServiceNow's Knowledge 2026 conference just days ago, Nvidia CEO Jensen Huang called the SaaSpocalypse narrative a fundamental misunderstanding of how AI actually works inside enterprises. His case was simple:
"Agents won't replace the tools, but agents will use tools."
Think about what that means in practice. An AI agent handling customer service does not invent a new workflow engine. It uses ServiceNow. A finance agent does not bypass SAP. It populates SAP more consistently and at greater speed than any human team ever could. A design agent does not build new chip tools from scratch. It runs the existing ones.
At Nvidia itself, Huang is already seeing this play out. The number of compilers, scripts, and software instances running across the company is growing rapidly, not because there are more human employees, but because there are more agents using the existing tools. Tool usage goes up, not down.
The market sold software on the fear of displacement. The reality is multiplication.
Part Four: The 100x Agent Thesis and the New Revenue Model
This is the insight I believe the market has not yet fully priced in, and the reason I think the software selloff represents one of the more significant mispricings we have seen in recent years.
At the ServiceNow Knowledge 2026 conference, Huang delivered a line that stopped the room:
"For the first time, service is software. Software is service and the service industry is 100x larger than the software industry."
Let that sink in.
The traditional software industry grew by selling seats, one license per human user. A company with 10,000 employees bought 10,000 seats. The market was capped, by definition, by human headcount.
Agentic AI breaks that ceiling entirely.
Huang's vision, which is not theoretical but already happening, is a world where every human employee works alongside dozens, hundreds, or even thousands of AI agents. He put a specific number on it at Nvidia's GTC conference: in ten years, he expects Nvidia to have 75,000 human employees working alongside 7.5 million AI agents. That is a 100-to-1 ratio of agents to humans.
Now ask yourself: what software does each of those agents run on?
The answer is the same enterprise software already in use. ServiceNow. Salesforce. SAP. Microsoft. The platforms companies have spent years integrating into their operations. Agents do not reinvent the enterprise stack. They run on top of it. And every agent that runs on top of it generates usage, tokens, and revenue.
This is where the business model shift becomes the investment thesis.
The legacy SaaS model charged per seat, per human user. But agents do not have seats. They run tasks. They generate tokens. They execute workflows at machine speed, around the clock, without logging off.
The smartest software companies saw this coming and have already begun converting their pricing models accordingly. Salesforce now charges per Agentforce conversation, not per user seat. ServiceNow is moving toward usage and outcome-based pricing alongside its traditional seat subscription model. Microsoft is charging for Copilot usage that scales with consumption, not headcount.
The implication is profound. When a company deploys 1,000 AI agents that each interact with ServiceNow hundreds of times per day, ServiceNow's revenue from that account is no longer capped by the number of human employees. It scales with agent activity. And if Huang is right that agents will outnumber humans 100-to-1 in the enterprise of the future, the total addressable market for software just expanded by an order of magnitude.
ServiceNow CEO Bill McDermott put a number on it: the company expects to cross nearly $16 billion in subscription revenue this year and double to $30 billion by 2030. Much of that growth is being driven not by adding human seats but by usage from AI agents running on the platform.
Huang framed the full opportunity at the ServiceNow conference this way: the entire industrial economy, manufacturing plants, warehouses, logistics networks, has largely been untouched by software until now. He called it a $50 trillion opportunity that barely existed for the tech industry before agentic AI arrived. The agents that unlock that opportunity will run on software.
Part Five: The Coming Rotation
Here is the forward-looking thesis.
The market sold software stocks on two fears. The first was that AI would rebuild and replace what software companies had built. The second was that AI would compress the number of human seats companies needed to pay for. Both fears caused a sweeping selloff that treated Duolingo and Microsoft as though they faced identical risks.
They do not.
The companies worth paying attention to are the ones with deep moats, mission-critical integrations, and switching costs so high that displacement is not a realistic outcome. Microsoft has 82% of Fortune 500 companies running on its platform. ServiceNow has a 98% customer renewal rate, doubled its free cash flow over three years, and is on track to grow subscription revenue from $16 billion this year to $30 billion by 2030, driven largely by AI agent usage on its platform. Salesforce has decades of CRM data that no AI startup can replicate overnight and has already pivoted its pricing model to charge per agent conversation rather than per user seat. These are not companies being disrupted. They are companies being handed a larger revenue model than the one they had before.
The companies best positioned to monetize the AI era span the entire stack. Anthropic and OpenAI benefit at the model layer every time an agent runs a query. Enterprise platforms like Microsoft, ServiceNow, and Salesforce benefit at the application layer every time an agent uses their tools. The pie is not being divided, it is growing.
Huang said it plainly: "Every single enterprise software company will also be a value-added reseller of tokens from Anthropic, OpenAI, and others." That is not a disruption story. That is a revenue expansion story.
Think about the early days of the internet. First, the money went into laying cable and building servers, the hardware layer. Then, once the infrastructure was in place, the real fortunes went to the companies that figured out how to use it productively. Google. Amazon. Netflix. We are approaching that exact inflection point with AI. The hardware trade played out. The software trade is setting up.
The current consensus is that AI replaces software users and therefore software revenues shrink. The emerging reality is that AI agents multiply software usage and usage-based pricing means revenues can grow faster than they ever could under a seat-based model. When the market internalizes that distinction, and it will, the rotation back into quality software names will be fast and significant.
And right now, the market is still pricing these companies like the wrong story is true.
The Behavioral Angle
I want to close with something that matters more than the sector analysis.
The biggest risk in this environment is not picking the wrong ETF. It is reacting emotionally to a rotation that has already happened, selling software after a 25% drawdown, or chasing hardware after a 50% run.
Two-thirds of all undervalued stocks in Morningstar's technology coverage are currently software companies. That number was under 9% just one year ago. The fear is real. The prices reflect that fear. And in most cases, the businesses themselves have not changed nearly as much as their stock prices suggest.
The hardware trade felt uncomfortable in early 2025, when everyone was still debating whether AI spending would hold up. The software opportunity feels uncomfortable now, when the fear of disruption is loudest and the price declines feel most painful.
That is almost always how the best opportunities feel.
The investors who benefited from the hardware trade were the ones who understood the narrative before it became obvious to everyone. The investors who will benefit from the software rotation will be the ones who distinguish between businesses that are structurally impaired and businesses that are temporarily mispriced because the market is telling the wrong story.
The market is still reading chapter one. Chapter two is where the software trade lives.
Disclaimer:
The information provided in this newsletter is for general informational and educational purposes only and does not constitute personalized investment, financial, legal, tax, or other professional advice. The views expressed are those of the author as of the date of publication and are subject to change without notice. This material should not be construed as a recommendation to buy or sell any security or to adopt any investment strategy. Readers should conduct their own research and consult with their financial, legal, or tax advisors before making any investment decisions. Past performance is not indicative of future results. All investing involves risk, including the possible loss of principal.
