The Real Challenge of Enterprise AI Isn’t the Model

AI models are impressive. But deploying them in enterprises is a different beast. The real problem lies beyond the model itself.

Nvidia’s CUDA transformed their chips into AI powerhouses. CUDA—short for Compute Unified Device Architecture—was created by Nvidia executive Ian Buck. This software foundation turned Nvidia hardware into the backbone of AI development.

Yet frontier models like Fable 5, Opus 4.8, and GPT-5.5 stumble when faced with complex, regulated workflows. Out of the box, these models don’t cut it. They lack the domain expertise and governance that enterprises demand.

Bratin Saha, CEO of NTT DATA AIVista, called this hurdle “the last mile” at VB Transform 2026. This stage involves wrapping models in proprietary data, workflows, and guardrails. “It’s not just a model,” Saha said. “You’re building a system around the model.”

This last mile requires deep domain knowledge. Companies guard their workflows closely, and these are rarely documented. The challenge is to translate this tribal knowledge into AI systems that can safely and effectively operate in regulated environments.

“Most enterprise AI projects fail during implementation because of poor integration, domain specialization gaps, lack of governance, and unclear ownership of outcomes,” Saha explained. This failure rate is a cautionary tale for anyone chasing AI magic.

Success demands more than technology. It needs domain expertise and change management. “You need technology, you need the domain expertise, and you need the change management expertise,” Saha said.

He stressed that deploying AI in enterprises means moving workflows from point A to point B—not just dropping a model into production. The intelligence must live in the surrounding system, not solely in the model. This approach allows companies to swap models and use open-source solutions without rebuilding everything.

NTT DATA’s advantage lies in its 20 years of insurance tribal knowledge. This trust and experience can’t be replicated overnight. It forms the foundation for effective AI systems in regulated fields.

Meanwhile, in China, Alibaba’s Qwen3.8-Max has emerged as a top AI contender. Eddie Wu Yongming, Alibaba’s CEO, stands behind it as a potential leader in the market. This was noted in early August 2026, highlighting global competition in AI development.

The takeaway is clear: AI models grab headlines, but enterprises win with systems built around those models. The last mile is the real grind—and the real value.

Stay connected via Google News

Follow us for the latest AI News