Every Company Is a Software Company Now

The Setup

There is a version of the AI software story that has been told a hundred times this year, and it is told from the seller's chair. Software stocks are down. The iShares Expanded Tech-Software ETF has shed roughly a quarter of its value in 2026. Salesforce is off about 30%. Workday is down more than a third. For the first time in the sector's history, software's forward price-to-earnings multiple has slipped below that of the broad market, falling from the mid-80s during the 2020 to 2022 peak to roughly 23 times forward earnings. The market has concluded that the recurring revenue model is under structural threat, and it is repricing accordingly.

We agree with that conclusion. We made the case ourselves in The SaaS Reckoning, which walked through why AI does not have to replace software to damage it. It only has to make software optional. That piece is the supply side of the story: what happens to the companies that sell software when their customers gain the ability to build.

This piece is about the other side of the table. Because here is the thing that the selloff narrative quietly skips over: the value that is draining out of software stocks does not evaporate. It moves. When the cost of building software collapses, the unique economic advantage that software companies enjoyed for two decades, the ability to compound margins as they scaled, stops being a property of the software industry and starts becoming available to everyone else. The phrase that has defined the last fifteen years was Marc Andreessen's line that software is eating the world. The next fifteen years invert it. Every company is becoming a software company, whether it sells software or not.

That is not a slogan. It is an argument about where margins go next, and it is the kind of shift that the market tends to price late, because it does not show up in a single dramatic earnings miss. It shows up slowly, in the gap between revenue growth and headcount growth, across hundreds of companies that nobody thinks of as technology businesses at all.

The $500,000 Line Item That Became $5,000

Start with the input cost, because everything else follows from it. The price of producing working software has fallen by an order of magnitude in roughly three years, and the decline is not theoretical. It is showing up in the operating data of companies that have actually done it.

The old math was brutal and well understood. A custom internal tool, a workflow system, a reporting dashboard, a customer portal, used to require a team of five to ten engineers, six to twelve months of calendar time, and a budget that started near $500,000 and climbed past $1 million once you added a product manager, a designer, and the maintenance tail. That cost structure is exactly why the buy decision usually won. Building was a capital project. Buying was a line item.

Then the input cost fell through the floor. AppDirect, a commerce and subscription platform company, reported reaching more than 90% AI-generated code within a single year, alongside a 70% reduction in development costs and a fivefold increase in output, with internal teams shipping roughly five times more applications than a traditional engineering organization of the same size could support. That is not a demo. That is an operating result at scale.

It is not an outlier either. Microsoft and Google have both said publicly that somewhere between a quarter and a third of the new code inside their own repositories is now generated by AI. By some counts, more than half of all new code committed to GitHub in early 2026 was either generated or substantially assisted by AI. The cultural marker arrived in late 2025, when vibe coding, the practice of building working software by describing what you want in plain language, was named word of the year. When the dictionary notices, the trend is no longer early.

Put the two numbers side by side. The build that used to cost $500,000 and half a year of a funded team now lands closer to $5,000 of engineering time for a well-defined tool. That is the single most important price change in enterprise technology in a generation, and it does not only threaten the companies that sell software. It rewrites the cost structure of everyone who was ever going to buy it.

What Made Software Special

To understand why this matters beyond the software sector, it helps to be precise about what investors were actually paying for when they paid premium multiples for software companies. They were paying for operating leverage.

A traditional business has a cost that scales with its revenue. A retailer that wants to sell twice as much has to buy roughly twice as much inventory. A consulting firm that wants to bill twice as many hours has to hire roughly twice as many people. A manufacturer that wants to ship twice as many units needs more materials, more line workers, more shifts. Growth costs money in direct proportion to the growth, which caps the margin.

Software broke that link. Once the product is built, the cost of serving the ten-thousandth customer is close to the cost of serving the thousandth, which is close to nothing. That is why healthy software businesses run gross margins in the 70% to 85% range while a good retailer fights for 30%. Revenue could climb while costs stayed roughly flat, and the difference fell straight to the bottom line. That is operating leverage, and it is the entire reason the market was willing to pay forty or eighty times earnings for software when it would pay fifteen for an industrial.

The premium was never really about the code. It was about the shape of the cost curve. Software companies could grow without growing their expenses at the same rate, and that property, rare and valuable, was treated as a structural moat that only technology businesses possessed.

Hold that thought, because the build-cost collapse does something specific to it. It does not destroy operating leverage. It hands operating leverage to companies that never had it.

The Privilege Is Escaping the Industry

Here is the inversion. For two decades, if you wanted software-company economics, you had to be a software company. You needed the engineers, the capital, and the time, which meant only firms organized around building software could capture the leverage. Everyone else bought the leverage secondhand, by paying a SaaS vendor a margin to rent it.

When building drops to $5,000 and a few weeks, that requirement disappears. A regional bank can build the internal underwriting tool it used to license. A mid-sized retailer can build the inventory and pricing system it used to buy. A logistics company can build the routing and scheduling software that three vendors used to sell it at six figures a year each. None of these companies becomes a software vendor. They do not sell the tools they build. But they capture the economics anyway, because the work those tools do scales without adding the people it used to require.

This is the part the seller-side story misses entirely. When a bank replaces a licensed workflow tool with one it built, two things happen at once. The SaaS vendor loses a contract, which is the part the market is busy pricing. And the bank quietly acquires a sliver of software-company economics, a process that now runs at near-zero marginal cost and lets it handle more volume without proportionally more staff. The first effect is visible and dramatic. The second is invisible and durable.

Multiply that across an entire economy and the implication is large. The competitive advantage that justified two decades of premium software valuations is being redistributed. It is flowing out of a few hundred software vendors and into the operating models of thousands of ordinary companies that learned to build. The market knows how to price the first part of that trade. It has barely started pricing the second.

The Proof Is Sitting in Your Browser

I do not have to reach for a hypothetical here. You are reading this on the evidence.

Wealth Engine Pro is a full investment research platform. It scores thousands of stocks across financial health, trend strength, and valuation, runs a daily AI analysis pipeline, and carries brokerage integration, portfolio tracking, options tools, billing, and a conversational assistant. I described how it was built, by one person, with no engineering team and no venture funding, in The SaaS Reckoning. The point there was that the build cost had collapsed. The point here is different and, for an investor, more useful: a one-person company is now running on a software-company cost structure.

Think about what that means. The marginal cost of serving the next subscriber on this platform is close to zero. The work that would once have required a support team, a data team, and an engineering team is carried by automated pipelines that do not get tired and do not get added to payroll as the user count grows. That is operating leverage, and it now belongs to a business with one employee.

Now scale the example up rather than down. If a solo founder can assemble software-company economics out of AI tools and managed infrastructure, consider what a established company with a real balance sheet, genuine domain expertise, and an existing customer base can do with the same tools. The solo founder is not the impressive case. The solo founder is the floor. The impressive case is the unglamorous mid-cap, in a boring industry, that has quietly started building instead of buying and is watching its revenue per employee climb while its competitors keep renewing their software contracts at a 15% annual increase.

Where the Leverage Shows Up

A thesis that cannot be measured is just a feeling. So the question that matters is where this shift becomes visible in the numbers, because that is where the analysis has to live.

The cleanest signal is the gap between revenue growth and headcount growth. A company that is genuinely capturing software-style leverage will grow its revenue while its employee count stays flat or falls. That shows up as a steadily rising revenue-per-employee figure, and it tends to drag gross and operating margins up with it over several quarters. The story is not a single blowout quarter. It is a trend line that bends in the right direction and keeps bending, which is exactly the kind of pattern that systematic scoring is built to catch and that narrative-driven coverage tends to miss.

The second signal is in the cost structure itself. Watch for companies where the software and IT line is shifting from a rising subscription bill toward a smaller, flatter spend on AI tooling and infrastructure, paired with stable or shrinking technical headcount that is producing more. A business that used to pay six vendors and is now paying for model credits and a small internal build team is converting a recurring operating expense into durable internal capability.

The third signal is the one that is easiest to overlook: it will appear first in industries nobody calls technology. Distribution, insurance, specialty finance, logistics, healthcare administration, regional banking. These are businesses drowning in repetitive, rules-based process work, which is precisely the work that collapses in cost when it can be built rather than bought and run by software rather than staff. The most interesting candidates for this thesis will not screen as technology companies. They will screen as ordinary businesses with quietly extraordinary margin trends.

The analysis does not point to a single ticker, and that is the point. It points to a screen: revenue climbing faster than headcount, margins expanding without a pricing story to explain it, in companies that were never supposed to have software economics. That is where the value flowing out of the SaaS sector is quietly landing.

What Could Go Wrong

This thesis has a real bear case, and ignoring it would be the same narrative-over-data error we are warning against. Here is where the argument can break.

The build is harder to keep than to start. Klarna offers the cautionary tale. The company built an in-house AI system to replace licensed customer relationship software, watched customer satisfaction slip, and reversed course. Building the first version is now cheap. Maintaining it, securing it, and keeping it from rotting is not. Gartner projects that 40% of AI-augmented coding projects will be canceled by 2027 over escalating costs, unclear business value, and weak risk controls. The leverage is real, but so is the graveyard of half-built internal tools.

The productivity gains may be partly an illusion. A METR study found that developers expected AI to speed them up by around 20% but were measured running roughly 19% slower on familiar, complex codebases, because the time saved typing was eaten by time spent reviewing and correcting what the model produced. If the real-world gains are smaller and slower than the demos suggest, the margin expansion this thesis depends on arrives later and weaker than the bulls expect.

Some things should never be built. Systems of record, payroll, core financial ledgers, compliance and medical data, remain firmly in the buy column, and for good reason. Where the output has to be correct every single time, there is no room for a model that occasionally invents an answer. A company that misreads the boundary and builds where it should have bought can do real damage, and the first major breach traced to AI-generated internal code could chill the entire trend overnight.

The leverage may accrue to the tool layer instead. There is a credible argument that the durable winner is not the company doing the building but the company selling the picks and shovels, the model providers and AI tooling that every builder pays. We explored where that value concentrates in Who Is Winning the AI Race? If the tool layer captures most of the surplus, the operating leverage that ordinary companies think they are capturing could get competed away into their AI bills.

None of these risks cancel the thesis. They shape it. The shift is real, but it will reward discipline over enthusiasm, the companies that build the right things well over the ones that build everything badly. The build-cost collapse is a tool, and tools do not improve a business that uses them carelessly.

The Bottom Line

The SaaSpocalypse is the loudest financial story of the year, and the market has done a thorough job pricing it. Software multiples have compressed below the broad market, more than $2 trillion in value has been erased, and the reasons are well documented. That trade is largely understood.

The trade that is not understood sits on the other side of the same shift. The collapse in the cost of building software did not just threaten the companies that sell it. It detached the most valuable property in technology, operating leverage, from the technology industry and made it available to any business willing to build. That is a slower, quieter, and far larger story than the selloff, and it will play out over years rather than quarters.

The reason it is hard to trade is the same reason it is worth trading. It does not announce itself in a single earnings miss. It accumulates in the space between revenue and headcount, in margin lines that creep upward for reasons the sell-side has not yet bothered to explain, in companies that no screen tags as technology. The headlines are watching the building burn. The more interesting question is who is quietly moving into the neighborhood.

At Wealth Engine Pro, we follow the numbers, not the narrative. The narrative says AI is killing software. The reality is that AI is handing software economics to everyone else, and the evidence for it will not be in a press release. It will be in the data: revenue per employee, margin trends, the widening gap between what a company earns and what it has to spend to earn it. That is where this story gets told, and it is exactly the kind of signal a systematic platform is built to find before the narrative catches up.