Artificial intelligence (AI) has cemented its role as a key sector for investors, but its path forward is shifting.
Several catalysts, including sustained AI infrastructure spending and US Federal Reserve interest rate cuts, are poised to drive tech sector growth in 2026; however, massive capital expenditure digestion by hyperscalers, alongside increasing demands for a return on investment and persistent power supply limitations, are influencing a rotation in focus, with risks like high valuations and policy uncertainty potentially capping AI industry gains.
Overall, experts are calling for the technology sector to navigate a delicate balance between aggressive expansion and necessary financial discipline in 2026, with AI at the heart of these matters.
Capex digestion and AI verticalization
AI capital expenditures by hyperscalers are projected to fuel demand for semiconductors, data centers and related infrastructure in the year head, as per Nicholas Mersch, portfolio manager at Purpose Investments.
According to notes from multiple analysts, the Big Four — Meta Platforms (NASDAQ:META), Microsoft (NASDAQ:MSFT), Alphabet (NASDAQ:GOOGL) and Amazon (NASDAQ:AMZN) — are slated to spend over US$300 billion on AI infrastructure. Mersch cited forecasts that see hyperscaler capex hitting roughly US$600 billion in 2026.
“Over the next 12 to 24 months, the narrative likely shifts from who can build fastest to who can drive the highest revenue and margin per dollar of AI infrastructure,’ Mersch added. “This is where verticalization matters. The companies that can capture the full stack, from silicon to applications, look like they will win.’
His top pick in this arena is Google, followed by Microsoft.
While the cloud layer remains a high-stakes game of concentration among a few platforms, Mersch said the hardware layer underneath is beginning to fragment as the chip stack quietly diversifies.
“Large multi-year AI chip deals are broadening the market beyond NVIDIA (NASDAQ:NVDA), with Advanced Micro Devices (NASDAQ:AMD) and custom application-specific integrated circuit (ASIC) programs winning meaningful share. High-bandwidth memory (HBM) has become the real bottleneck and profit pool, with tri-sourced HBM3E, an emerging HBM4 race and surging HBM demand from ASICs,’ the expert said.
‘The result is a more plural, multi-vendor accelerator ecosystem. Looking out to the second half of the decade, total AI silicon spend can keep growing even if individual GPU vendors see more competition and pricing pressure, with memory, packaging and custom silicon capturing a larger share of the economics.’
Chip diversification, however, is now colliding with HBM and packaging shortages, constraining output from 2026 to 2027. BMI’s Cedric Chehab notes that rapid capex growth is outpacing supply, ruling out near-term oversupply, but warns of volatility if data center investments fail to deliver profitability amid persistent infrastructure shortages.
Power as a binding constraint for AI
Power limits are a specter looming above AI expansion heading into 2026.
“Individual campuses are pushing past 1 gigawatt, utilities in key regions are scrambling to add generation and transmission and Big Tech is signing multi-gigawatt nuclear and long-term power deals, including restarts of previously shuttered plants,” explained Mersch. US data center demand is now poised to triple by 2030, thrusting utilities, nuclear operators and grid infrastructure into prime investment orbits.
“Even Google has acknowledged that serving capacity needs to double roughly every six months,” he added.
Alphabet, the parent company of Google, and other hyperscalers became active infrastructure developers in 2025, inking high-profile strategic deals designed to secure 24/7 — and carbon-free — energy for AI data centers.
Google’s deal with Elementl Power in May to provide capital to develop three advanced nuclear sites in the US represents a shift toward nuclear energy that is perhaps the most significant structural change in the AI landscape today, further extending the verticalization narrative into the power grid itself.
The shift toward energy-backed AI is being institutionalized at the highest levels of finance. In late 2025, JPMorgan Chase (NYSE:JPM) launched its US$1.5 trillion Security and Resiliency Initiative, a decade-long plan specifically targeting the intersection of AI, grid infrastructure and nuclear energy.
By earmarking US$10 billion in direct equity for US firms, the initiative effectively underwrites the full-stack transition.
Are AI stocks in a bubble?
The path for AI is moving from building technology to proving its value. While many experts remain optimistic, the transition from deployment to execution introduces new risks that could define the industry’s next winners and losers.
As organizations fully embed AI into their core workflows, the operational stakes are shifting. Infrastructure strategies are diversifying as security-conscious businesses seek more control over their high-value AI workloads.
Simultaneously, the rise of agentic AI, which automates full workflows, combined with cost and complexity issues on major hyperscalers, will lead to a trend of cloud repatriation toward regional and bare-metal platforms.
Despite concerns over a potential bubble, the industry will continue to receive massive institutional backing. B2BROKER’s John Murillo rejects the idea of an AI bubble, comparing OpenAI to Edison’s plants amid giants’ resilience.
‘In the case of dot-coms, everyone was investing just to invest; it didn’t matter what exactly to choose and some of the projects didn’t have a solid foundation. With AI, it’s not like this. The technology proves its worthiness every day, and it has already swept away many junior analysts,’ Murillo emphasized.
Nevertheless, high AI valuations risk corrections if adoption disappoints or energy constraints emerge.
The success of the current capex cycle will depend on whether these investments translate into measurable operating leverage and cost savings through the back half of the decade.
“The bubble scenario is very unlikely,” Murillo added. “I think in the current economic situation, there are problems much worse than a potential bubble.”
For example, geopolitical tensions, sticky inflation and US midterm elections could spark volatility, prompting sector rotations away from overvalued mega caps.
Investor takeaway
The investment focus in AI is shifting from the initial narrative to tangible execution and quantifiable profitability. While the challenges of elevated valuations and geopolitical instability persist, some experts dismiss comparisons to a technology bubble, arguing the sector’s demonstrated value offers a stable underpinning.
Future leaders in the AI industry will be distinguished by their capacity to convert infrastructure spending into significant operating leverage and cost efficiencies.
Securities Disclosure: I, Meagen Seatter, hold no direct investment interest in any company mentioned in this article.