The AI Funding Reckoning: Why Mere Hype No Longer Closes Rounds

During 2021, mentioning artificial intelligence in a funding pitch was often enough to attract capital. Smaller enterprises and mid-market organizations rushed to adopt early, purchasing quickly and broadly. Yet many of these newly funded startups faced a harsh reality: customers acquired in the initial wave disappeared just as fast. The churn was dramatic and widespread.

The underlying problem was straightforward. Numerous products simply replicated what existing software already accomplished, merely incorporating AI as window dressing. Others delivered marginal time savings on tasks that had no bearing on anyone's core responsibilities. A 10% or 20% efficiency gain on a peripheral activity looks impressive in a demo but rarely translates into contract renewals.

The reality is that genuinely valuable AI businesses have never been easy to build, despite the abundance of capital flowing into the space. The market is now recalibrating to reflect this truth. Organizations that deliver real, measurable value are expanding at rates previously unseen in enterprise software. For those backing these companies, the critical challenge ahead involves identifying which ones can sustain their customer bases—and whether upcoming public offerings will reveal the gap between sustainable and unsustainable models.

Building defensible AI businesses in today's market

Before developing any software, my co-founder and I conducted interviews with more than 900 mechanical engineers across all seniority levels. We systematically explored how they allocated their time and which hours they would most value reclaiming. These conversations became foundational to our approach to constructing an AI company.

The first principle is straightforward: the gains must outweigh the costs of implementation. Trimming 10% from an infrequent task rarely reshapes organizational operations. However, compressing a mission-critical workflow from weeks into minutes, or generating substantial cost reductions, absolutely does.

The second principle centers on specialized knowledge. As large language models transition into foundational infrastructure, access to quality models has become nearly universal through APIs. The real business opportunity emerges from deep industry understanding—the ability to tackle problems that off-the-shelf tools simply cannot address.

The third principle involves proprietary information. Consider manufacturing and engineering: every organization producing physical goods has accumulated years of accumulated technical knowledge. Much of this remains trapped in archived documentation or embedded in the experience of long-tenured staff. Foundation models have no access to this information because it never appeared in their training datasets. An AI system lacking this context will struggle to become indispensable. The competitive moat strengthens as the product becomes woven into customer operations, making replacement increasingly difficult and costly.

What this means for funding and IPOs

Forthcoming AI public offerings will subject these business models to unprecedented scrutiny. Institutional investors will dissect net revenue retention and gross margins, determining whether explosive growth can translate into lasting profitability. They will also investigate whether serving each customer becomes more efficient as the company scales. The benchmarks they establish will cascade down, shaping expectations among private investors evaluating earlier-stage AI ventures.

The shift is already visible in funding discussions I participate in. A few years back, reaching $1 million in annual recurring revenue represented a significant milestone. In today's AI landscape, a new product can hit that figure within roughly a year—and then vanish within the next twelve months. Revenue figures alone no longer provide sufficient insight for investors.

What backers consistently probe for is expansion within existing accounts. They seek evidence that customers are deepening their financial commitment beyond the initial purchase. Account expansion serves as the strongest indicator available that a product has genuinely transformed how an organization operates: the customer has recognized the value and allocated additional resources accordingly.

Throughout the coming year to two years, I anticipate capital will concentrate on companies merging deep sector knowledge with substantial improvements to business-critical operations and access to proprietary customer information. Growth may occur through hands-on, customized deployment rather than self-serve scaling, yet these companies can expand within accounts that have developed trust. Conversely, companies offering thin wrappers around third-party models—those with impressive client rosters and retention metrics they prefer not to disclose—will encounter substantially tougher fundraising environments. Steady, sustained growth over five years beats spectacular growth that lasts five quarters.

Maor Farid is founder and CEO of Leo AI, an artificial intelligence platform designed specifically for mechanical engineering and large-scale model development for physical product design. He conducted research in AI and mechanical engineering at MIT as a Fulbright postdoctoral fellow and holds the distinction of being the youngest Ph.D. graduate in the Technion – Israel Institute of Technology's history. Farid has assembled a community exceeding 60,000 engineers and operates a nonprofit dedicated to supporting underrepresented youth.