August 7, 2026, (Inside AI) — The Ai4 conference in Las Vegas this week drew more than 12,000 attendees from over 90 countries, all confronting a stark reality: after years of pouring billions into AI infrastructure, corporate returns remain largely theoretical. The shift from “tokenmaxxing” — the race to maximize model compute — to scrutinizing chatbot budgets dominated hallway conversations.
Despite the sobering talk, the allure of hypothetical payoffs held the crowd. Robot dogs roamed the exhibition floor, and sessions on artificial genomes promised revolutionary breakthroughs. Yet the gap between spending and tangible outcomes was the elephant in the room.
Corporate AI investment has surged, with global spending projected to hit $632 billion by 2028. But recent estimates show that fewer than 30% of AI projects deliver measurable ROI. The Ai4 crowd knew this well, yet the promise of AI-driven transformation remains a powerful narrative.
“We’re past peak hype but not yet at peak value,” said Sebastian Pellejero, a Reuters Breakingviews columnist attending the event. “The challenge is moving from pilots to production without burning cash.”
Read: OpenAI Proposes ‘Useful Intelligence per Dollar’ as the New AI Scorecard
One major theme was the hidden costs of scaling AI. Beyond model training, companies face ballooning expenses for data preparation, integration, and ongoing maintenance. A 2025 McKinsey survey showed that 70% of AI budgets go to non-model activities.
Vendors pitched solutions: automated MLOps, cheaper fine-tuning, and “AI-in-a-box” products. But skepticism lingered. A panel of CFOs admitted that AI’s financial impact often gets lost in accounting, with benefits spread across departments and hard to isolate.
Meanwhile, the generative AI boom has created a new class of spending: chatbot licenses. Companies are paying per-seat fees for tools like Microsoft Copilot, but productivity gains are anecdotal. “We’re seeing a lot of experimentation, not transformation,” noted one attendee from a Fortune 500 firm.
The conference also highlighted a shift toward smaller, domain-specific models that cost less to run. This “right-sizing” trend could curb the infrastructure arms race, but it requires in-house expertise that many firms lack.
Regulatory uncertainty added another layer of caution. With the EU AI Act now in force and U.S. rules evolving, legal teams are slowing deployments. “Compliance is becoming a bigger line item than compute,” said a healthcare AI executive.
Still, the visionaries on stage argued that today’s spending is a down payment on tomorrow’s breakthroughs. They pointed to early successes in drug discovery and materials science, where AI has cut research timelines by years.
Read: Turn AI Adoption into Real Business Value with Four Steps
But for most companies, the path from theoretical benefit to bottom-line impact remains murky. As the conference wrapped, the message was clear: the AI revolution is real, but the bill is coming due, and the payoff is still a moving target.
Shamil has a low tolerance for AI that only works in controlled demos. He covers agentic systems and generative AI with the skepticism of someone who has seen too many “revolutionary” tools quietly disappear six months after launch.
