Investors Raise Expectations As AI Industry Proves Unforgiving

By Maor Farid

Back in 2021, simply slapping the label of AI onto a presentation deck could secure you funding. Small and mid-sized businesses were eager to jump in early and purchased software rapidly. I observed companies acquire customers and raise capital, only to battle to keep anyone actually using their tools. User retention plummeted drastically.

A large portion of those offerings were generic: they merely replicated standard software functions while tossing in a layer of artificial intelligence. Other applications shaved off a little time on trivial duties that no one’s job relied upon. A 20% efficiency boost on a nonessential task yields an impressive product demonstration alongside a canceled renewal.

Easy AI triumphs never truly existed, even when effortless cash circulated widely. The market is presently correcting that imbalance. Enterprises delivering genuine utility are scaling faster than anything I have previously witnessed within enterprise software. For financiers, the key inquiry is which of these firms can retain their client base — and whether upcoming initial public offerings will highlight the disparity.

What makes an AI business defensible today

Before my co-founder and I authored a single line of code, we spoke with upwards of 900 mechanical engineers, ranging from junior staff to vice presidents. We inquired where their hours vanished and which of those hours they would happily pay to reclaim. Those 900 discussions shaped my perspective on constructing an artificial intelligence enterprise.

First: the performance enhancement must justify the financial outlay. Trimming 10% off an occasional chore rarely alters how an organization functions. Conversely, shrinking a critical operational workflow from weeks down to minutes, or generating significant financial savings, certainly will.

Second: domain expertise. Frontier models are rapidly turning into foundational infrastructure. Utilization of a capable large language model is accessible to nearly anyone holding an API key. Commercial potential resides in comprehending a specific industry thoroughly enough to resolve challenges that off-the-shelf tools fail to address.

Third: proprietary context. Taking engineering scenarios as an example, every firm manufacturing physical items has accumulated decades of technical wisdom. Much of this expertise remains concealed within legacy blueprints or trapped inside the minds of veteran personnel. Such background information is missing from a foundation model’s original training data. An AI solution unable to access this data will struggle to become indispensable to the enterprise. This advantage compounds as the tool embeds itself tightly within customer workflows, making replacement financially burdensome.

What this means for funding and IPOs

The approaching wave of AI IPOs will subject these business architectures to heightened scrutiny. Public market investors will closely analyze net revenue retention and gross margins to evaluate whether explosive expansion translates into a viable, enduring enterprise. They will additionally seek to determine whether each deployment grows cheaper to maintain as the organization scales. The benchmarks they set will shape the expectations private backers place on earlier-stage artificial intelligence startups.

I already notice this shift occurring within my own fundraising dialogues. Several years ago, hitting $1 million in annual recurring revenue represented a major milestone. In the current landscape, a fresh product can touch $1 million in ARR within roughly a year, only to shut down twelve months later. Consequently, ARR alone no longer provides investors with sufficient insight.

Investors now persistently inquire about account expansion. They want to verify whether clients escalate their expenditures following the initial rollout. Expansion serves as our closest indicator that an offering has fundamentally transformed operational workflows within an enterprise: clients have witnessed its utility and committed a larger share of their own budgets toward it.

Over the upcoming 12 to 24 months, I anticipate capital continuing to direct itself toward businesses that synthesize deep sector proficiency with substantial improvements to mission-critical duties, alongside access to clients’ proprietary knowledge. While they might expand via hands-on deployment rather than viral self-service pathways, they retain the capacity to grow organically inside accounts that place trust in them.

Firms whose products represent merely thin operational layers built on top of external models—boasting flashy client rosters alongside retention metrics nobody cares to display—will discover their subsequent financing rounds drastically more difficult than the last. I would personally opt for steady, robust growth spanning five years over spectacular expansion lasting merely five quarters.

Maor Farid is the founder and CEO of Leo AI, the first AI for mechanical engineering — a large mechanical model for physical product design. He conducted AI and mechanical engineering research at MIT as a Fulbright postdoctoral fellowship and became the youngest Ph.D. graduate in the history of the Technion – Israel Institute of Technology. Farid has built a community of more than 60,000 engineers and supports underserved youth through his nonprofit initiative.

Related Crunchbase query:

• Global Venture Funding To AI Startups In 2026

Illustration: Dom Guzman