Most AI Projects Will Fail, Boomi CEO Says—Spotlight on ROI, Not Hype

Steve Lucas, a 30-year enterprise software veteran who previously served as CEO of Marketo before its $4.75 billion（約7600億円） acquisition by Adobe, now leads Boomi, a data activation platform serving over 30,000 global customers. In a July 12, 2026 podcast recorded for Motley Fool, he sat down with analyst Rachel Warren to discuss the inflection point AI spending has reached across corporate America.

The central diagnosis is blunt: after two years of board pressure to adopt AI and build AI strategies, companies are now demanding proof that these investments generate returns. Lucas reports walking into board meeting after board meeting where CEOs are asking the same question: "Help me show return" from the AI projects already deployed. The grace period for spending without accountability is closing.

The numbers underscore the magnitude of the shift. Four major U.S. hyperscalers—the foundational cloud infrastructure providers—increased AI capital expenditure from around $400–410 billion last year to over $700 billion（約110兆円） this year. At the same time, broader market trends show software application spending declining while infrastructure and AI investment rise sharply. A critical cost dynamic compounds this tension: training frontier AI models now costs over $1 billion（約1600億円）, up from roughly $50,000 for GPT-2. OpenAI, a bellwether, burns $3 billion（約4800億円） a month. These losses cannot be sustained indefinitely by investors or companies, Lucas argues, meaning costs will eventually shift to consumers and enterprises.

Lucas estimates that as many as 40% of enterprise AI projects will ultimately be abandoned. He traces failures not to technical limitations but to rushed implementation and weak governance. Companies can start an AI project in minutes—calling an API, asking ChatGPT to process business data—without defining business requirements or strategic outcomes. They iterate freely, forgetting the discipline that enterprise software demands. Trust is the hidden killer. Lucas draws on three decades of experience: if humans do not trust data or results, the system fails. One person in a boardroom asserting "this is inaccurate" triggers mistrust that spreads through the entire initiative. Gartner has flagged that many agentic AI projects—AI agents working autonomously inside businesses—are expected to either fail or be shut down by the end of 2027.

The near-term implication is tightening. Within the next six months, Lucas predicts most CEOs will impose internal caps on AI spending. After that, boards will demand measurable ROI before approving tens, hundreds of millions, or billions in future AI investment. When funding for frontier model companies starts to dry up—a possibility given the unsustainable burn rates—those organizations will pass costs to consumers, further disciplining enterprise demand. The winners, Lucas suggests, will be infrastructure and chipmakers like Nvidia, which are already monetizing AI directly. The next wave of enterprise beneficiaries will likely be companies that help businesses extract genuine returns from AI, not those that build or promote AI models themselves.

The shift Lucas describes reflects a fundamental maturation in how corporate America evaluates AI investments. For two years, companies faced board pressure simply to adopt an AI strategy and use the right buzzwords on earnings calls. That era is ending. The proxy for this transition is the explosion in hyperscaler capex—a $300 billion（約48兆円） year-over-year increase—coupled with widespread CEO complaints that their AI spending has jumped 10X to 20X without corresponding revenue gains. Even executives like Elon Musk, known for aggressive investment, have begun capping AI spending at their companies.

Lucas's core argument is that most failures stem not from technology shortcomings but from implementation discipline. Companies start AI projects in five minutes, borrowing cloud APIs or ChatGPT, without defining business requirements or strategic outcomes. They then iterate without rigor, hoping something sticks. This mirrors decades of failed enterprise software rollouts—not because the software was broken, but because adoption failed. Lucas emphasizes that "change only happens at the speed of trust," and trust erodes the moment one person in a boardroom calls out inaccurate results. Once trust breaks, the entire effort collapses.

The timing matters. Lucas identifies a "grace period for now, not for long." Within six months, he predicts most CEOs will impose internal caps on AI spending. By the end of 2027, Gartner expects many agentic AI projects to be shut down. The constraint is not innovation but cost: OpenAI burns $3 billion（約4800億円） monthly, and those losses will eventually transfer to enterprises and consumers. Until companies learn to tie AI projects to measurable ROI—not just experimentation—the 40% failure rate will likely hold or worsen.