Artificial intelligence (AI) with real-time reasoning is around the corner. Smaller, more transparent AI models are shaking up the competitive landscape worldwide. Rising resource demands are inviting creative chip-design solutions. Autonomous agents will work together to make people and organizations more efficient.
These four enterprise AI will drive significant change over the next few years, according to Debojyoti “Debo” Dutta, chief AI officer at Nutanix, the hybrid multicloud software company that provides a platform for managing applications and data across different IT infrastructures.
In an interview with The Forecast, Dutta explained how he expects these trends to unfold and why he thinks they are pivotal to AI’s evolution.
1. Reasoning Models Make AI Much Smarter
Reasoning models ramp up the cognitive capabilities of generative AI, which uses large language models (LLMs) trained on immense datasets. For all their capabilities, GenAI apps still can’t think like a human brain because they have a hard time responding to new information.
“A reasoning model can interact with real-time data like news articles or images or temperature and make decisions,” Dutta said. “And that is called reasoning roughly because that's how humans or any other animal try to learn from their environment and make decisions.”
Reasoning models take all the learning already gained via LLMs and fold in new information as it arises. And they can solve much more complex problems in shorter time frames.
“They can start to do more root-cause analysis,” Dutta added.
For instance, if a doctor has a patient whose symptoms are difficult to diagnose, a reasoning model could pull in current patient data and combine it with learning models based on knowledge collected from patients around the world.
Some reasoning systems have been around for decades, like the models designed to play chess, poker and other games. Reasoning models change the game because they can join forces with LLMs. Sure, there’s been plenty of hype about this kind of cognitive computing. But it’s a genuine step forward.
“Reasoning models can be used to prove math theorems,” Dutta said. “That gives you a sense of how real it is.”
2. AI Leadership Will Be Open, Small and Global
LLMs require extensive guardrails and governance to ensure that users’ queries produce useful information free of bias or inaccuracies. Ideally, these models would be open so that developers and the public can see what makes them tick and decipher their performance.
“But in reality, most big models are always closed — built inside a walled garden,” Dutta said.
This is likely to change as models move toward openness. Model sizes will shrink while their reach will expend to pretty much everywhere. The DeepSeek model from China is a prime example of the small, open and global trend.
“We now see a race where these bigger models have competition from a bunch of small models that are permissibly open in many ways,” Dutta said. “They're globally distributed and they are catching up to the bigger models.”
He said smaller models most likely can’t outperform their bigger brethren.
“But when you pick a small, open model from a geographically diverse vendor, you might get more transparency and agility, and you'll have a large community to innovate and that will give you more resilience in the longer run,” Dutta said.
Moreover, a small model distilled from larger models can be much more energy efficient and easier to manage in a private setting.
“And in certain parts of our business, many of our customers prefer models from a particular region, and that helps us serve a wider variety of customers because we are a global company,” Dutta added.
3. Inference Scaling Drives Growth of Next-Gen Architectures
Inference is the heart of AI because it uses computing power to infer a user’s intent. As more people continually place more demands on AI systems, inference costs are bound to scale upward. Thus, AI services may cost more in the future, especially with reasoning models coming online.
“That is the nature of the game, the price you have to pay for better intelligence,” Dutta said. Higher AI fees may come with even higher benefits, however.
“From an absolute point of view, it'll increase costs,” he said. “In reality, it might give you much more back in terms of benefits and productivity gains.”
Heavier demands on data centers raise the prospect of higher electricity and cooling costs. Dutta suggested that power demands might not rise as quickly as inference costs, thanks to next-generation microprocessor architectures. Conventional semiconductor designs kept microprocessor and memory chips segregated, creating a bottleneck for data traveling between the two.
“New computing architectures put the compute elements and the memory on the same chip,” he said. Compute operations can fetch data from memory at orders-of-magnitude faster rates, making these chips prime candidates for AI workloads.
Thus, the computing muscle of these new chips creates efficiencies that will give enterprises tools to rein in AI power consumption even as they adopt more AI use cases, Dutta suggested.
4. Multi-Agent Systems Make AI More Productive
Enterprises are already developing AI agents that automate manual processes and free people to find better uses for their time.
“This field is moving very fast as we speak,” Dutta said.
The next phase on the drawing board is getting multiple autonomous agents to work together.
“The proofs-of-concept are already there,” he added. “I expect that in a couple of years, we will have multi-agent systems that can negotiate with other agents — with humans in the loop as the proctor.”
How should enterprises experiment with multi-agent systems?
“A company can start by looking at their business problems and where they need efficiency,” Dutta suggested.
Those business needs should form the foundation for building individual agents. Once those agents have proven their worth, organizations can start working on coordinating their functions.
“It’s crawl, walk, run,” Dutta concluded.
Tom Mangan is a contributing writer. He is a veteran B2B technology writer and editor, specializing in cloud computing and digital transformation. Contact him on his website or LinkedIn.
© 2025 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.