After two years in which investor enthusiasm centred on chipmakers and cloud providers, attention is beginning to turn to the companies expected to deploy AI at scale and to the physical and political constraints that could slow its expansion.
“AI has been one of the biggest themes in the market, and AI’s diffusion through the economy is going to be one of the biggest themes in 2026,” said Michelle Weaver, head of US Thematic Research at Morgan Stanley at RF Bank & Trust’s Cayman Economic Outlook conference on 11 Feb.
Weaver discussed the evolution and impact of AI, breaking it into three parts: the underlying technology, enablers and adopters. In her presentation, she highlighted the rapid improvement of AI models, the significant infrastructure needs, and the labour market changes AI will bring.
AI models expected to improve
Generative AI models, trained on vast amounts of computing power, have improved at an astonishing pace. The next generation of models is trained on roughly ten times the compute of earlier systems, a scale-up that analysts expect will deliver another step change in capability this year, Weaver said.
“We’re fundamentally bullish on the underlying technology. We think that the models are going to continue to improve at a really rapid and really surprising rate, and we think the models are going to continue to become ever more capable,” she noted.
Usage is rising just as fast as performance. The volume of “tokens”, or units of data processed by AI systems, has surged as companies and consumers move beyond simple chat functions towards image generation, coding assistance and more complex, multi-step tasks.
Energy bottlenecks
But the infrastructure needed to sustain that growth is under strain. Data centres, the facilities that house AI servers, are major consumers of electricity. Morgan Stanley estimates that projected chip sales tied to AI would require around 74 gigawatts of additional power, equivalent to more than seven times New York City’s average daily usage.
Even allowing for capacity already under construction and spare grid supply, the bank sees a shortfall of roughly 49 gigawatts. Alternative sources, such as gas turbines, fuel cells and limited nuclear generation, could narrow the gap, but analysts still anticipate a deficit of 10-20%.
The power constraint has become both an engineering and political issue. In parts of the United States, particularly in Virginia and the Mid-Atlantic region where data centre clusters are concentrated, voters increasingly blame large facilities for higher household electricity bills. Local opposition has led to project delays and heightened scrutiny from policymakers.
In response, major technology groups are exploring ways to generate power on site rather than relying solely on local utilities, Weaver said. Such “off-grid” solutions may reduce friction with communities but highlight the growing divide between private capital and public infrastructure.
Aside from energy, geography is another factor in the AI landscape. The United States controls the bulk of global computing capacity for advanced AI models, far outpacing China and the EU. Washington’s strategy has focused on maintaining leadership in frontier model development, while Beijing has emphasised broader diffusion of AI across priority industries.
AI adoption to accelerate
Beyond the underlying AI models and infrastructure development, adoption is going to determine the economic impact of artifical intelligence, Weaver argued.
“The last chapter of the story, the biggest chapter, the broadest, and the one that’s going to cause the most real societal level change is the AI adopters.
“These are the companies that will be adopting AI, using it in their businesses to make themselves more efficient, to make their employees more productive.”
Surveys suggest that AI tools are already embedded in daily life. Around three-quarters of Americans report using AI at least once a week, while a fifth is using it daily for work or personal tasks. This familiarity is lowering barriers for corporate deployment.
A Morgan Stanley survey of chief information officers found that 81% expect to have at least one AI-enabled tool live in their organisations by the end of this year. Early applications have centred on IT operations, marketing and customer service, where automation can deliver quick efficiency gains. Now AI is expected to move from isolated pilot projects to cross-functional deployment.
Retailer Walmart, for example, uses AI to improve demand forecasting, manage inventory and operate increasingly automated warehouses. In agriculture, John Deere has developed autonomous tractors and computer-vision systems that identify weeds and apply chemicals precisely.
AI-supported robotics presents another disruptive frontier. In Amazon’s warehouses, the ratio of robots to human workers has risen sharply over the past decade, driving efficiency gains. Similar trends are emerging in retail logistics and food preparation.
Automation and augmentation effects still unclear
The labour implications remain contentious. Analysts distinguish between “augmentation”, where AI enhances human productivity, and “automation”, where it replaces workers altogether. In white-collar sectors, augmentation is likely to dominate in the near term, with AI tools assisting rather than eliminating roles.
Morgan Stanley modelled the potential labour value of tasks that could be automated or augmented by AI, estimating up to $920 billion in possible gains for S&P 500 companies alone. The figure is theoretical, but it shows the scale of expectations embedded in the markets.
Weaver told delegates, “I don’t think the jobs five years from today are going to look like the jobs of today, but there will be a shift in the composition of the labour market, rather than necessarily a huge spike in unemployment.”
Still, recent volatility in technology stocks shows how quickly sentiment can shift, as investors are grappling with how rapidly productivity gains and disruption could materialise. Shares in some software groups fell sharply after demonstrations suggested new AI tools will encroach on tasks once thought insulated from automation, including legal research and coding.
“I think you’re going to continue to have these big moments of AI‑driven disruption, AI‑driven volatility, and they’re only going to continue as these models become more capable and the ability of these models continues to grow,” Weaver said.
Evidence of the financial impact is beginning to emerge. Morgan Stanley found that companies identified as significant AI adopters have outperformed peers where AI plays a marginal role.
Meanwhile, early signs of labour market adjustment are visible. A Stanford study found a decline in entry-level software developer roles in recent years. Corporate surveys suggest that while some jobs have been eliminated or left unfilled due to AI, new roles are also being created, leading to smaller net changes overall.
“We ran a survey of companies around the globe to see how AI is already impacting hiring and labour decisions, and people reported over the last 12 months AI has contributed to an elimination of 11% of jobs and 12% of jobs not being backfilled. But it’s also added 18% new jobs, so that leads us to around a 4% net loss figure,” Weaver said. However, the results are mixed by country. “In the US, we actually saw a net gain of 2% thanks to AI, but we saw net losses in the UK and Germany.”
The central question is whether AI can lift labour productivity growth, which has been steadily falling. The answer will depend not only on model breakthroughs but on energy supply, regulatory responses and how effectively companies integrate AI into everyday workflows.
