Shifting Enterprise AI Investment Toward Measurable Data Infrastructure

The current AI investment cycle is shifting from a focus on growth at any cost toward a demand for measurable ROI. While hyperscalers spend hundreds of billions on infrastructure, the broader enterprise market is hitting a wall of diminishing returns. The result is a looming reckoning where projects lacking clear, data-backed outcomes will be shuttered by 2027. Investors who prioritize companies providing foundational plumbing, such as data connectivity, quality, and governance, over those chasing flashy, loss-making frontier models will gain an advantage. This conversation provides a roadmap for identifying which businesses are building durable, high-margin infrastructure and which are burning capital to mask a lack of fundamental utility.

The ROI Reckoning and the End of the Blank-Check Era

In this conversation, Boomi CEO Steve Lucas outlines a shift in how enterprise software budgets are allocated. For the past two years, boardrooms have been driven by fear, forcing executive teams to launch AI initiatives without clear strategic outcomes. This created a blank-check environment that is now rapidly closing.

The market is responding to the unsustainable economics of frontier models. As Lucas notes, the cost to train these models has exploded from $50,000 for GPT-2 to over $1 billion for current frontier models. Because these costs cannot be sustained by negative economics in perpetuity, they will eventually be passed to the enterprise and the consumer.

There is no doubt that that is the case today. I just think it is the enormity of the pressure on that left-hand side coupled with rushing into a lot of AI projects. We are not seeing the high rates of return that you would expect from businesses.

-- Steve Lucas

Why Easy AI Leads to Systemic Failure

The ease of starting an AI project, which Lucas describes as cracking your knuckles and asking Claude or OpenAI to do things, is a trap. Because implementation is deceptively simple, teams often bypass rigorous business requirements and strategic planning. This creates a feedback loop where projects are launched for the sake of experimentation, only to be abandoned when they fail to produce measurable results.

The implication is that the 40% failure rate predicted by Gartner is not a technical failure, but a failure of process. When organizations prioritize the buzz of AI over the energy of data, they create systems that lack trust. Lucas emphasizes that trust is the ultimate bottleneck; if a system produces one inaccurate result, the entire effort is often discarded by the organization.

After 30 years in software, I know one thing. And that is if humans do not trust something, it will never be used. And I forget AI. The reality is, I have seen thousands of business intelligence or analytics or data projects that fail because the data was not accurate and no one trusted it.

-- Steve Lucas

Identifying the Real Moats: Data as the Energy Source

Systems thinking reveals that while chip manufacturers like NVIDIA capture the immediate value, the long-term winners are the companies providing the infrastructure that makes AI usable for the enterprise. Lucas argues that AI without data is meaningless.

The competitive advantage lies in knowledge graphs and unique data sets that cannot be replicated or generated by AI models alone. While many software companies are currently using AI as a convenient way to explain away layoffs or market stagnation, the truly transformative companies are those seeing accelerated new customer acquisition. New logo growth is the primary signal of a product that solves a real business problem, rather than one that is simply being forced onto existing customers.

Key Action Items

• Audit AI Spend for ROI (Immediate): Stop funding experimental AI projects that lack a defined business outcome. If a project cannot demonstrate efficiency gains within the next six months, it should be a candidate for termination.
• Prioritize Data Infrastructure (Next 6-12 Months): Shift investment focus from model-layer companies to data-activation and governance platforms. Durable value is moving toward the infrastructure that cleans and connects data to models.
• Look for New Logo Growth (Ongoing): Use new customer acquisition as your primary filter for identifying AI winners. If a company is struggling to win new clients, their AI strategy is likely a marketing layer rather than a transformative engine.
• Demand Evidence, Not Spin (Ongoing): When companies cite AI efficiency as a reason for layoffs, require specific, data-backed proof of productivity gains (e.g., code output per engineer, finance cycle time) rather than accepting the narrative at face value.
• Evaluate Ontology and Hardware-Plus Strategies (12-18 Months): Watch for companies that combine AI with proprietary data or unique physical hardware. These full-stack approaches are harder to disrupt than pure SaaS-wrapper AI companies.