There is this MIT research quote (State of AI in Business 2025) that many experts keep referencing: 95% of AI proof-of-concepts fail. It comes up in talks, podcasts, articles, and frankly, we quote it as well, both in our pitches and on our website at SolveStack.ai. Because the statistics say a lot.

This number became so familiar to us that, when we recently watched a podcast discussing this data again, our first thought was that it must be old. That’s how fast the AI industry moves: anything that feels familiar already feels outdated. But it turned out the episode was just two months old. So, what does it tell us? That, clearly, the problem hasn’t gone away.

This reflection made us want to understand why the failure rate is still so high, and why so many experts see startups as part of the solution.

Internal enterprise systems are rusty

One of the core arguments was surprisingly simple: Enterprise IT systems are old.

They are slow. They are rarely refreshed. Teams work on the same frameworks for years, sometimes decades, and over time they stop upgrading their skill sets in meaningful ways.

This creates a situation where companies simply don’t have the internal capacity to build new AI products. Not because they lack smart people, but because their environment doesn’t support modern product development anymore.

So, when the pressure to innovate appears, enterprises naturally look outside. Very often, this means hiring consultancies to introduce new tools, methods, and technologies.

Scale makes change harder, not easier

Another important factor is scale.

In large organizations, systems are shared across teams. Any new solution needs alignment, explanation, documentation, onboarding, and support. Introducing new AI tools is not just a technical task, it is an organizational challenge.

Explaining complex solutions to many teams, aligning workflows, and changing habits is not something enterprises are eager to do. They are not lean. They are not particularly agile. And they are definitely not structured like startups.

This is why delegating innovation becomes the default choice and that is when either external consultancy companies or startups are addressed. We think startup innovation is the way to go.

Why startups work differently?

Startups are good in this space for a very specific reason.

Their products are still simple. They don’t have too many layers. They are not overloaded with features designed to satisfy everyone. And because their survival depends on the product, they are forced to think carefully about usability, value, and speed. Their vulnerability often makes them more eager to experiment and work closely with the client.

Startups build entire businesses around products. Enterprises rarely do.

That difference alone explains a lot about why startups are often better positioned to deliver working AI solutions, especially in early and mid-stage implementations.

Transformation is hard in established environments

All of this leads to a broader point.

In well-established organizations, transformation is simply harder to implement. Internal products – especially those meant only for internal use and not direct commercial gain –  are often given low priority. Combined with limited hands-on experience in AI product development, it becomes very easy for companies to give up early.

On top of that, enterprises tend to be risk-averse when it comes to product design. They prefer improving what already exists rather than betting on something new. Startups don’t have that luxury. Risk is built into how they operate.

When enterprises and startups meet

When enterprises and startups come together, something interesting happens.

Enterprises are excited by the ambition of startup founders, their belief in the product, their optimism, and their willingness to take the ownership. There is also comfort in knowing that responsibility for implementation does not sit entirely inside the organization.

But for this to work, trust is required. And time. Finding startups that an enterprise can truly get excited about is not trivial.

This is where many initiatives slow down or fail before they even start.

No magic, only cooperation

Yes, startups are eager to improve their solutions, but enterprises should not expect magic. Successful AI projects require cooperation. They demand openness, feedback, and support from both sides.

The most successful outcomes come from unions, not purely transactional relationships.

And this is exactly the gap SolveStack.ai aims to close: making it easier for enterprises and startups to find each other, communicate better, and build something meaningful together.

Because when that union works, AI stops being a failed experiment and starts becoming a real capability.

This piece was inspired by a recent episode of the Lightcone Podcast, where Y Combinator experts discuss why AI adoption remains so challenging for enterprises. We highly recommend giving it a listen and we’d love to hear your thoughts as well.