Investors trying to capitalize on artificial intelligence should stop treating it as a narrow technology trade and examine the entire stack, from the infrastructure supplying computing power to the models running on it and the applications that could ultimately reshape industries, according to Morgan Stanley’s Jitania Kandhari.
The deputy chief investment officer of the investment bank’s Solutions and Multi-Asset group identified her top 10 investment themes in AI that investors should understand when building a global portfolio. Her framework highlights investable components across and within the AI stack, encouraging investors to look beyond the most visible chipmakers and U.S. technology giants.
Need more clues? Ask the Sherlock chatbot in the lower right corner to summarize this story, explain technical concepts or answer other questions.
Her report, “Artificial Intelligence: Ten Investment Truths,” frames AI as a full-stack capital cycle spanning sectors, countries and asset classes. It explores how value migrates across the AI stack as infrastructure bottlenecks shift, agentic and physical AI emerge, software business models evolve, geopolitical competition intensifies and AI moves from a technology story to an economy-wide investment theme.
“These were the top 10 things that I felt an investor and an asset owner should know,” she told The AI Innovator. The report “lays out the whole landscape in place, separating the signal from the noise.”
Among the different layers of the AI stack, she looks for bottlenecks – or shortages. With more demand than supply, companies can charge higher prices, potentially leading to higher revenue that could ultimately lift stock prices.
Investing in picks and shovels
Among her 10 investment themes, she favors the “picks and shovels” supplying the AI infrastructure boom in the near term. Over the intermediate term, she sees the next capital cycle emerging in autonomy and physical AI. Longer term, she expects value to migrate toward businesses across industries that use AI to raise productivity, improve margins and create new products and services.
“The most important investment chronology from here is the picks and shovels, which is the more immediate opportunity,” she said.
While chips attracted initial investment interest, Kandhari said AI bottlenecks are shifting quickly among chips, memory, advanced packaging, networking, cooling and power. Within the power system, transformers can become a greater constraint than the grid itself, while growing bandwidth demands are pushing data centers toward silicon photonics as copper interconnects struggle to keep pace.
Energy widens the opportunity further. Kandhari pointed to gas turbines as a bridge while renewable and nuclear generation expand, as well as uranium mining as nuclear power becomes a larger part of the electricity mix. The underlying logic is that hyperscaler capital spending becomes revenue for the companies supplying these scarce components.
“When bottlenecks create pricing power, bottlenecks create business moats,” she said.
Some of those trades, however, are no longer inexpensive. Kandhari said certain gas-turbine and electrification companies have order books extending to 2030, but much of that optimism is already reflected in their valuations. She continues to see potential in pockets of memory and packaging outside the U.S., uranium, the more energy-efficient tensor processing unit (TPU) ecosystem and the components supporting autonomy.
Medium- and long-term bets
The intermediate opportunity begins when AI moves from generating information to operating in the physical world. Kandhari sees autonomy developing across factory floors, drones and in space. Yet she would rather invest in the common components these systems require — sensors, actuators, detectors, LIDAR and specialized materials — than try to predict which robot manufacturer will prevail.
“I want to buy that rather than buying the robot companies,” she said, noting that manufacturers must invest heavily while competing over product quality and pricing. “That supply chain, which is the picks and shovels of autonomy, is the more medium-term opportunity.”
The longer-term opportunity may be less conspicuous. Kandhari expects investment value to shift from AI enablers to adopters across retail, health care, financial services, industrials and other sectors.
“The value creation will move from the AI enablers to the adopters,” she said.
That shift also complicates predictions of what Wall Street fears would be a “SaaSpocalypse,” in which generative AI destroys much of the software industry by making code cheap and easy to produce. Kandhari agreed that a company whose principal advantage is writing code could be vulnerable, but said the market has indiscriminately punished even software businesses with more durable advantages.
“The world is just painting software with the same brush,” she said. “People believe that AI will eat software. I don’t think that’s going to be the case.”
She looks for software companies with proprietary data, domain expertise, workflow integration, orchestration capabilities and knowledge of industry-specific legal and compliance requirements. In those cases, AI could strengthen the business rather than displace the vendor.
“You can’t just buy the software index,” she said. “You really need to be discerning.”
Bearish about AI model companies
Kandhari is considerably more cautious about the foundation model layer. “I’m most worried about the models here,” she said.
The proliferation of U.S., Chinese and open-source models raises the prospect that the technology will commoditize, just as overbuilding in previous technology cycles eventually led to falling prices and broader adoption. China, constrained by access to advanced chips but benefiting from ample power, is developing a parallel and more efficiency-focused AI stack. The U.S. has leading chips but faces tighter power constraints.
“There are models all over … that someone’s launched, and every day I can’t keep up,” Kandhari said. “Model (companies) are where I see the most commoditization risk.”
She also questions how much model developers will ultimately be able to charge and which will deliver the best ratio of tokens to power consumed – or tokens per watt. “That’s the space where there’s the most uncertainty.”
The economics of token consumption pose a more immediate test. Kandhari recounted a meeting with a European industrial software company whose executives said some inference workloads require so many tokens that hiring experienced software engineers in Poland, Hungary and the Czech Republic would be a cheaper option — and they supply human judgment as well.
“That’s going to be a very interesting thing to study over time,” she said.
AI capabilities to rise 250x by 2028
The pace of technical progress nonetheless remains startling. Kandhari’s report cites industry estimates that AI capabilities are doubling every four months, which would make systems 250 times more capable by 2028. She said the figure came from company representatives she met at a Morgan Stanley technology conference.
AI follows a familiar investment pattern: infrastructure is built, prices fall, usage expands and applications emerge that were difficult to envision at the start. Telegraph networks and the internet followed that trajectory, but Kandhari said AI is moving considerably faster and could permeate society at least as broadly as the internet.
“It’s not just hardware, it’s not just the Mega Seven (tech giants), it’s not just software,” she said. “It’s like industrials, it’s like materials. So even the lines are blurring here.”
One underappreciated area is agentic commerce, in which AI systems act and transact for users. Kandhari also sees overlooked applications in drug discovery, clinical trials, agent workflows and other uses embedded inside company operations rather than marketed as stand-alone AI products.
Asked to compare AI’s boom today to the dot-com rise and crash, she said today’s largest technology companies driving the boom differ from many dot-com-era businesses in one key aspect: They generate substantial cash.
The principal bubble risk, she said, lies in excessive capital spending and the financial interdependence developing among cloud hyperscalers, model companies and suppliers. This “circular financing” occurs when cloud companies pay their suppliers and then receive orders back from them.
If the cloud giants decrease their spending, suppliers could take a hit because shares are already priced for continued spending growth. “That would look like a bubble if the capex is cut,” Kandhari said.
To stay a step ahead of a potential bubble bursting, she monitors “under-the-hood” indicators such as GPU rental rates because investors watching only hyperscaler budgets would spot red flags when it’s already too late.
She is less concerned about the still-developing application layer, however, where the winners are not yet clearly defined.
“Amazon was a bookstore until the internet came, and then that was the application layer of the internet where it became this highly consumer digitized platform,” Kandhari said. “That application layer of AI, which is the Amazon of AI, is unknown but will happen eventually.”
Be First to Comment