The Second Great Tech Bubble

By Will Melton · March 19, 2026 · 6 min read read

AI is thriving, but a bubble is forming in bloated, overpriced SaaS software. Companies built on complexity face disruption as new, specialized tools emerge.

Everyone keeps waiting for the AI bubble to pop.

Wrong bubble.

AI isn’t in a bubble. AI is in a heyday. The companies building the foundational models, the inference engines, the compute infrastructure — they’re printing real utility. Real value. Real staying power.

The bubble is in software. Specifically: the bloated, overpriced, workflow-heavy, multi-seat SaaS layer that’s been extracting rent from businesses for the last two decades on the premise that software is hard to build. It isn’t anymore. And the companies that built their moats on complexity are about to find out what happens when complexity becomes cheap.

The Math Was Always Broken

Here’s the scam hiding in plain sight.

The actual compute cost of running most SaaS tools is dirt cheap. The markup has always been astronomical — tens of thousands of percent in some cases — because the pitch was simple: “We built the thing so you don’t have to.” That worked when building was expensive, when engineering talent was scarce, when integrating data systems required an army.

That era is over.

Now the question every savvy operator is asking isn’t “how do we get better software?” It’s “why are we paying this company to house our own data and charge us to access it?”

If you’re an investor in software companies, ask yourself a hard question: does the software I’m backing have immense utility to the customer — utility so specific, so deeply embedded, so irreplaceable that the customer has absolutely no reason to build their own version?

If the answer is no, you’re holding a depreciating asset.

N=1 Software Is Coming

I’ve been building what I call N=1 software — tools built for one specific use case, one specific problem, one specific person or team — and sharing them where they make sense to share.

Sound Rewrite is an example. I built it because I had a specific problem as someone who works with ideas, drafts, and long-form content. It doesn’t write for you. It helps you assemble the parts of what you’re already writing faster — so your ideas actually make it to the page instead of dying in a disorganized file structure. Fifty-five features, all of them designed to help writers finish.

A challenge to the incumbent tools? Not as a crusade. As a proof of concept. I built something that solved a real problem, for a real person, faster and cheaper than the alternative would have cost me in ongoing subscriptions.

That’s the model. And it scales.

We can now leverage the APIs of existing platforms to ingest data, own the outputs, and build purpose-built tools on top of infrastructure that already exists. You don’t have to start from scratch. You just have to be specific.

MCPs Are the Wrecking Ball

The platforms that built their value proposition on workflow complexity — the ones that charged you to build automations, maintain integrations, move data from one place to another — are staring down the barrel of something they didn’t see coming.

Model Context Protocol changes the game. When AI can connect directly to the tools and data sources that matter, the elaborate middleware layer those platforms monetized becomes redundant. We’re done lifting information manually. We’re done rebuilding workflows every time something breaks. The information is abundant. The connections are real-time. The decisions — which pieces matter, which actions to take, which tools to deploy — can now be augmented by AI in ways that make manual orchestration look like hand-cranking an engine.

The platforms that don’t adapt get replaced. Not by a competitor. By nothing. By absence. By someone deciding they don’t need the platform anymore and spending an afternoon building a better version for a tenth of the price.

A Warning to Investors

The companies that are probably safer for longer are the ones with real monopolies — Adobe, Intuit. They have leverage that most can’t replicate. But the new wave of SaaS companies, especially the ones vibe-coded into existence over the last twelve months? They are going to face real challenges.

If your app isn’t specialized, if it doesn’t provide immense value to the specific audience you serve, if it’s overpriced, if you’re marking up AI compute credits 10,000%, if a capable operator can replace it in a day — your model is at risk.

Overpriced tools that keep adding features to justify inflated invoices are a thing of the past. The SaaS-as-cash-cow era is closing. The companies that survive this moment will be the ones that put people on the front lines to build real relationships with customers, that pair service with their tools, that work continuously to make their customers’ lives easier. Software alone won’t close deals anymore. Buyers are hesitant. They’re looking everywhere for answers. Confidence in a path forward is the only thing that will shorten a sales cycle, and confidence comes from quality service paired with quality tools — not from another feature release.

What We’re Actually Doing About It

We’re not just watching this shift. At Xponent21, we’re building into it — for ourselves and for our clients.

Earlier this year we built a trade show game for a client. In the first hour of using it, they saw an immediate lift in LinkedIn followers. That was a secondary goal. The primary goal was also met. One custom tool, built for one specific moment, delivered on two fronts before lunch.

We’re doing this for nonprofits. For small and medium businesses. For startups with sound models that aren’t going to get wiped out by AI in six months. We’re helping organizations transition away from expensive software that was built for the crowd and tuned for nobody in particular. Software that does too much for everyone else and too little for them.

The work is to meet clients where they are and build what they actually need — tools that fit the way they work, priced in a way that respects their budget, paired with the expertise to make them effective. That combination is available now in a way it has never been before, and Xponent21 is delivering it.

I’m About to Drop Something

Later this week I’m going to share something specific.

One piece of software I use costs me around $24,000 a year. I replaced it in two days. My version will run me about $500 a year.

It’s a tool that nearly everyone in my industry uses and everyone overpays for. I haven’t decided how I’ll get it out there yet — open source is on the table. Stay tuned.

The markup is obscene. It’s been obscene for a long time. The only thing that kept it going was the assumption that building was out of reach for most people.

It isn’t anymore.