The AI opportunity is being driven by accelerating contracted enterprise revenue, and the dispersion between potential winners and losers has become the defining feature of the landscape.
The comparison to the 1990s technology bubble is misleading on the most critical metrics:
Valuations. If this period were anything like 1999, the major AI winners would likely be trading at multiples in the 60–70 times range. Today, many AI beneficiaries trade at a 20–25 times forward P/E multiple, which, in some cases, values them at a discount to widely held quality consumer names.
Adoption. AI crossed into mass adoption ahead of the capex peak, not after. Most companies we speak with are already using AI internally to improve productivity. That is the reverse of the 1999 dynamic, when spending ran ahead of demand that had not yet materialised.
See also: China helped stabilise global markets from oil shock: Asia Group
Funding model. The vast majority of AI capex is being funded by the operating cash flows of the largest hyperscalers, supplemented by private credit to a new generation of cloud infrastructure providers. This is fundamentally different from the speculative IPOs that characterised the late 90s technology bubble. In the current environment, demand continues to outpace the most optimistic expectations. As a result, capex budgets continue to increase, with the additional funding requirements getting disproportionate attention.
‘It’s all just hype; the end-demand isn’t there’
The empirical evidence does not support this critique. AI demand is increasingly visible in contracted revenue. By the end of 1Q2026, the combined backlog of contracted cloud commitments across the major hyperscalers had exceeded US$2 trillion ($2.58 trillion). Several hyperscalers have publicly stated they cannot keep pace with current demand.
‘It’s too late: We’re already at the peak’
We believe we are still in the early innings of the AI investment cycle. Hyperscaler capex is expected to exceed US$700 billion this year and is on track to approach US$1 trillion in 2027. We anticipate that approximately 50% of next year’s expenditure will be allocated to semiconductors, representing a significant shift in profit distribution towards AI computing at an unparalleled scale.
See also: Expect a period of ‘choppy rotation’ in equities: Schroders
‘The AI investment opportunity is too narrow’
The acceleration in AI demand has exposed multiple supply-chain chokepoints. These represent substantial incremental investment opportunities in our view.
The first is CPUs. As we move from AI model training to inference and to agentic AI, CPU chips play a bigger role in these AI workloads. Companies delivering CPUs for AI workloads have guided for materially higher revenue growth as inference demand scales.
The second is memory. Currently, there is a shortage of memory, and producers are gaining pricing power through new long-term and take-or-pay contracts for the first time in years.
Finally, data centre demand requires a multifold increase in power generation over the next few years. Natural gas turbine manufacturers are largely sold out through the end of the decade. This creates room for innovative solutions, including off-grid, behind-the-meter, fuel-cell technologies, which offer better time-to-power and improved environmental outcomes.
‘AI monetisation will never meet optimistic expectations’
This pushback is fascinating, as it comes despite evidence that some of the fastest-growing companies in corporate history are AI companies.
Roughly half of agentic AI usage relates to software engineering. Coding has been the natural first use case: it is structured and well-defined, and AI can learn it faster than a human can. That has made the technology industry’s own adoption frictionless. Technology companies are already using AI to drive down coding costs and increase output. Use cases are now expanding into other enterprise verticals, with companies re-engineering workflows around agentic AI.
Aside from technology, most users are unlikely to learn AI coding on their own. They will need packaged applications. There are hundreds of AI-native venture-funded private companies attacking individual verticals with agentic products; a few of them are likely to grow into very large businesses.
For more stories about where money flows, click here for Capital Section
‘SaaSpocolypse and the challenge of investing in tech right now’
The “SaaSpocolypse” refers to the recent sell-off in software companies as the market began to worry about the sustainability of their investment moats, with AI allowing companies to code their own software applications quickly, easily, and cheaply.
Firstly, we do not expect customers to remove their enterprise software systems in the near term. However, the value-add layer on top of that data is increasingly being delivered by agents rather than by the incumbent software, which creates downward pressure on renewal pricing over time.
Second, enterprises are reallocating their internal software-development budgets away from new installations and toward agentic AI products. We think it is unlikely that a typical large enterprise will evaluate a new CRM installation this year. They are far more likely to divert their spending toward agentic tools that span their existing software stack.
Third, with modern AI models used to build new applications, competitors may be able to replicate years of accumulated engineering work in a fraction of the time, eroding what were once durable competitive moats.
The market is forward-looking and is already discounting these pressures. Most software companies are not missing earnings today, but the terminal value equation has changed. With more compelling opportunities elsewhere in the AI value chain, it is important to be decisive.
‘Nothing is ever all plain sailing’
We believe the key risks lie in supply and execution rather than in demand. There are constraints throughout the AI supply chain, which paradoxically makes this less of a gold-rush dynamic and more of a measured, multi-year buildout.
That said, there are specific risks worth monitoring closely. The first is the financial condition of the major AI labs. The second is political pushback on electricity prices and water consumption near data centres, a risk that off-grid power solutions may help to mitigate. The third is the unit economics of tokens.
The most common error we see in client portfolios today is the assumption that AI is a single trade, fully priced, late-cycle and analogous to past bubbles. Yet we believe this is a multi-layered and multi-year cycle, funded by cash flows of some of the most profitable companies in history and validated by contracted enterprise revenue.
The opportunity set is wider than the market appreciates, and the dispersion between winners and losers is now the defining feature of the landscape. In our view, the AI investment opportunity is a transformational, multi-phased investment opportunity that we expect to remain fertile ground for active managers in the years to come.
Mark Baribeau is head of global equity at Jennison, an affiliate manager of PGIM
