By Craig Muir, Head of Software, Data & Analytics
Technology dealmaking is open again — but it is more selective, more forensic, and more influenced by AI than anything I’ve seen in my career. Buyers still want to do deals, and there is plenty of capital on the sidelines. The question isn’t whether there’s appetite; it’s where that appetite concentrates, how defensibility is underwritten, and what gets filtered out quickly.
What’s Hot
1) Proprietary, workflow-embedded data
If you want a simple headline for what’s drawing attention in tech M&A, it’s this: many investors are doubling down on data. I’ve had multiple conversations recently where we barely get 20 seconds into a software description before the buyer asks, “Is this a software business?” — and if the answer is yes, the discussion can end right there. But if the next sentence is, “It’s a data business,” the tone changes immediately.
Not all data is created equal. Proprietary data sits at the top of the list; publicly data sits at the bottom. The AI test is straightforward: if it’s easy to collect, easy to replicate, and easy to stitch together with a model and a plug-in, you have a problem. If it truly is proprietary, domain-specific, and tied to a critical workflow — then you’ are in a very different category.
2) Data plus deep domain expertise—especially where “close enough” isn’t good enough.
In many professional and regulated settings, 75% accuracy is not a product: it’s a liability. The platforms that win here have decades of precedent, nuance, and expert judgment embedded in the underlying content, taxonomy, and workflows. AI can be a great assistant, but in high stakes use cases, customers will still pay for solutions that are tested, trusted and correct.
3) “Workflow” AI: narrower models trained on specific datasets
One thing I do not think gets enough attention is how quickly enterprise AI is moving from broad, general-purpose models to narrower, domain-specific solutions. Think of it like education: the best outcomes come from the best collateral, the best teachers, and the most relevant training data — applied to a specific discipline. The model matters, but the differentiation increasingly comes from what you train it on and how tightly it’s knitted into a workflow
4) Picks-and-shovels infrastructure that powers AI
You cannot have enterprise-grade AI without both the infrastructure layer and the data layer. That reality continues to support strong interest in the businesses that enable training, deployment, governance, and scaled compute — alongside the data assets that make AI truly useful.
What’s not hot — or at least, what’s harder
1) Easy-to-replace software and “low-friction” apps
AI has raised the bar for software investing. The categories that feel most challenged are the ones with low switching costs and limited workflow lock-in—where a capable team (or in some cases, a non-technical operator with the right tools) can replicate a meaningful slice of functionality surprisingly quickly. That certainly does not mean software is “dead.” It means buyers are having to underwrite a much higher standard of defensibility.
2) Broad, undifferentiated information products
If a business model is built primarily on organizing what is already public, AI can compress the value quickly. By contrast, mission-critical enterprise platforms that underpin core decisions and have real operational gravity are harder to displace. Even there, though, expectations around net retention and upsell can reset as customers satisfy fringe use cases with AI-driven point solutions.
What does this mean for deal activity?
Overall appetite is healthy, but it’s not evenly distributed. I’d frame it as a rotation: where software might have been 80/20 versus data historically, it feels closer to 50/50 today, even though there are simply more software assets in the market.
That shift drives two outcomes: more competitive processes for scarce, high-quality data assets — and more scrutiny, selectivity, and slower volume in software. If and when macro uncertainty eases and rates continue to normalize, I expect volume to increase; the heat map will still matter.
For founders and management teams, the most important work right now is being honest about where you sit on the AI-defensibility spectrum. If your product or dataset can be replicated easily, the market will price that in — and sellers will not like the outcome. If, on the other hand, you own proprietary data, have distribution and workflow integration, and can show customers are paying for accuracy and outcomes, you’re operating from a position of strength.
Bottom line
Technology M&A isn’t slowing because buyers have lost interest — it’s evolving because the definition of “quality” has changed. The winners are the assets that are hard to recreate, embedded in real workflows, and built on differentiated data and domain expertise.
That’s a constructive backdrop for sellers who can articulate defensibility clearly, and for buyers who are willing to do the work to separate what’s durable from what’s merely popular.
To learn more about Craig Muir’s views listen to his podcast interview on dealmaking in the age of AI.