What to Expect From AI in 2026: Personal Agents, Mega Alliances, and the Gigawatt Ceiling

Artificial intelligence (AI) models are becoming more than just chatbots—an important step in their evolution that will have repercussions for the global economy in 2026 and beyond, says Marco Argenti, Goldman Sachs’ chief information officer.

“In my 40 years in technology, 2025 saw the biggest changes I have seen in my career,” Argenti says. “And what’s crazy is we haven’t seen anything yet—in fact, I predict 2026 will be an even bigger year for change.”

AI has emerged as a critical driver for financial markets and potentially for the broader economy. Wall Street analysts, who have consistently underestimated the amount of investment going into AI, expect the largest hyperscale cloud computing companies to pour more than half a trillion dollars into capital expenditures in 2026. The seven biggest tech companies now account for more than 30% of the S&P 500’s market capitalization and roughly one quarter of the index’s earnings, according to Goldman Sachs Research.

Argenti, the former vice president of technology of Amazon Web Services, says AI is rewiring everything from the traditional workforce to the traditional software stack. He makes seven predictions about how AI could evolve in the near future:

AI models will be the new operating system

The traditional paradigm for software engineering is changing: Rather than functioning as one-dimensional applications, AI models are becoming operating systems that independently access tools in order to perform tasks.

In turn, computing is evolving from static, hard-coded logic to outcome-based assistants that reprogram themselves. This makes AI agents much more capable of handling complex problems. As a result, those who own the models will own the new operating systems that power AI agents.

Context is the new frontier

AI engineers’ focus will shift from building “larger models” to “better memory.” Think of it this way: The models have been built from vast pools of data—they’ve scoured essentially the entire internet and then some in the form of synthetic data for model-training purposes. However, the immediate context available to models—what they remember from previous discussions and tasks—is relatively tiny. Already some newer models are able to reason and inject much larger contexts into processes to provide far more bespoke, customized responses.  

The rise of the personal agent

AI personal agents will arrive, which is something companies have been chasing with varying degrees of success. What we do now with apps—manually, and in piecemeal fashion—will be done automatically soon. For example, if a flight is cancelled because of the weather, an AI agent will know to rebook the flight, reschedule meetings, and will order food for afterwards (since restaurants will be closed). This is very possible with AI with agentic capabilities.

The agent-as-a-service economy

Companies will shift from deploying human-centric staff to tackle tasks to deploying human-orchestrated fleets of specialized multi-agent teams. Instead of calculating billing by hours worked, these hybrid teams of humans and machines will charge clients by the amount of tokens—the units of data used by AI models—that are consumed.  

Learning becomes the most important skill

The workers who thrive will be the ones with expertise who are also the most willing to adapt.

For those workers, the single biggest differentiator will be their ability to reimagine—in an age where AI will help them to do their job—something they’ve been doing for many years. There’s recent precedence for this: With the introduction of computers, people had to rethink many aspects of their work. AI is generating a change of that magnitude, which makes learning the most important skill.

Winner-takes-most mega partnerships

AI is a game of scale, and there are going to be network effects from the very large upstream and downstream partnerships that are forming. Headline partnerships and strategic alliances of unprecedented scale will reshape the AI landscape. These networks will create a self-reinforcing cycle where only a handful of major players are capable of competing. In this way, AI may come to resemble complex major industries like aerospace that are characterized by duopolies.

Power is the new capital

Scaling to meet the AI demand will hinge not just on capital, but on access to the utility grid: Goldman Sachs Research’s base case is that power consumption from data centers will jump 175% by 2030 from 2023 levels (our analysts' previous forecast was for an increase of 165%). Capacity constraints, from access to new gas turbine power plants to electrical grid connectivity, mean access to electrical power will require the right set of relationships. 

The sheer scale of the infrastructure necessary for AI data centers, the multi-year lead time to bring new power facilities online, and the rapid evolution of AI models will exacerbate the need for power in 2026, resulting in a gigawatt ceiling. Companies will obsess over allocating every megawatt of power to activities with the highest return.

This article is being provided for educational purposes only. The information contained in this article does not constitute a recommendation from any Goldman Sachs entity to the recipient, and Goldman Sachs is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient. Neither Goldman Sachs nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this article and any liability therefore (including in respect of direct, indirect, or consequential loss or damage) is expressly disclaimed.