By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel consensus is bearish, warning of margin compression in hyperscalers, commoditization of AI inference, and regulatory hurdles in healthcare AI. They agree that AI is maturing but face risks like capex translating into sustainable revenue and margins.
Risk: Commoditization of AI inference and margin compression in hyperscalers
Opportunity: Expansion into underpenetrated areas like healthcare AI and data-center efficiency
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Artificial intelligence has moved from buzzword to the defining investment theme of the decade — and clients are asking their advisors one question above all: where should I invest in AI in 2026?
At the Future Proof conference, Bloomberg News reporter Alexandra Semenova put that question to three experts with very different vantage points: David Wright, head of quantitative investments at Pictet Asset Management; Joe Wilson, managing director at J.P. Morgan Asset Management and manager of the JTEK tech ETF; and Gina Sanchez, CEO of Chantico. Here are the key takeaways from their 44-minute discussion on AI investment opportunities, AI stocks, and portfolio construction for 2026.
The AI Investment Ecosystem: Five Layers of Opportunity
Sanchez mapped the AI trade into distinct layers that advisors can use to structure client portfolios:
AI chip stocks — the foundational semiconductor play (Nvidia, Broadcom)
AI services and software — Palantir, ServiceNow, Adobe
Energy and data center cooling — an overlooked play driven by strained electrical grids
Late adopters and applications — industrials, real estate, and the application layer now emerging in private equity and venture capital
Hyperscaler Capex: A $1 Trillion AI Spending Wave
Four hyperscalers — Alphabet, Amazon, Microsoft, and Meta — are projected to spend roughly $650 billion on AI capex this year, about triple the level of just a few years ago. Wilson expects that figure to reach $1 trillion within three years, though he cautioned the current "hyperbuild" phase won't grow in a straight line.
One structural insight for stock pickers: Alphabet is the only company that owns its own chips, large language model, cloud infrastructure, and proprietary data. Wilson also revealed he owns no Apple in his tech portfolio, calling the company "dead silent on AI" — though its non-AI profile could make it a defensive holding if AI sentiment cools.
Are Investors Underallocated to AI? Probably Not
All three panelists agreed on a counterintuitive point: most investors already have more AI exposure than they think through broad index funds — including hidden exposure via industrials, energy stocks, and analog semiconductors. As Wilson put it, eventually "every company is an AI company," just as every company became an internet company.
The opportunity for advisors isn't adding generic AI exposure — it's identifying specific sub-themes, like the AI energy trade, that broad indexes underweight.
AI Bubble Warning Signs: Is the AI Trade Overheated?
Is AI a bubble in 2026? The panel's froth checklist:
Valuation math vs. the dotcom era: Sanchez, a portfolio manager during the 1999 boom, recalled that Cisco's price once implied growth larger than Europe's entire GDP. Today's AI valuations are elevated but not yet at that extreme.
Earnings revisions without price response: Nvidia's forward earnings estimates rose roughly 100% in six months while the stock stayed flat — a classic sign of over-ownership, per Wilson.
Speculative spillover: froth in quantum computing stocks and questionable "AI-ticker" names.
Falling compute costs: Wright, an AI practitioner, noted his models now train in a day instead of a month — efficiency gains that could mean total AI spending undershoots expectations, or shifts from training to inference with different beneficiaries.
Best Ways to Get AI Exposure: ETFs, Active Funds, or Private Markets?
Broad diversification (Wright): the index does the heavy lifting; the AI trade rotates across industries, styles, and countries too quickly for most investors to chase.
Active management with an equal-weighted benchmark (Wilson): flexibility to classify companies like Robinhood, Figure, and Tempus AI as tech — and to avoid pure AI-themed funds that may fade as AI becomes a utility.
Private markets (Sanchez): the AI application layer lives in venture capital and private equity, where the 2025 "venture winter" has left stronger, Darwinian survivors.
AI and Software Stocks: The Operating Leverage Test
Wilson delivered the panel's sharpest critique: software stocks have underperformed semiconductors for a decade, and CEOs should stop defending their AI relevance on TV and start showing AI-driven operating leverage — slower R&D hiring and better GAAP margins. He contrasted Salesforce's best-ever 19% GAAP operating margin with Texas Instruments' 35–50%+.
Sanchez and Wright pushed back on the AI job-replacement narrative: coding tools like Claude Code supercharge senior engineers rather than replace them, and companies that gut junior talent pipelines risk a weaker foundation of future expertise.
The Most Underowned AI Opportunity: Healthcare
The consensus pick for the biggest underowned AI beneficiary over the next five years: healthcare. Records management, billing, insurance, diagnostics, and AI drug discovery (including DeepMind's protein-folding breakthroughs) all stand to gain. Sanchez added sectors undergoing private equity roll-ups — including wealth management itself — where technology stacks determine which consolidations succeed.
AI Investing in 2026: What's Overestimated vs. Underestimated
Overestimated: the smoothness of the AI trade (expect volatility, and possibly a current Mag 7 name dropping out within three years); the idea that every quant fund will become an AI fund; expectations of flawless AI output today.
Underestimated: how much AI empowers users and accelerates learning; and how dramatically AI lowers the barriers to starting a business — pointing to small-business job creation, not just enterprise job cuts.
FAQ: Investing in AI in 2026
What are the best AI stocks to watch in 2026? The panel highlighted chip makers (Nvidia, Broadcom), cloud hyperscalers (Microsoft, Amazon, Alphabet), AI services companies (Palantir, ServiceNow, Adobe), and overlooked energy and cooling plays supporting data center growth.
Is AI a bubble in 2026? Panelists say valuations are elevated but not yet at dotcom extremes. Warning signs to watch: earnings upgrades that no longer move stock prices, and speculation in unproven AI-adjacent names.
Should investors buy AI-themed ETFs? The panel was cautious: narrow AI thematic funds may underperform as AI fades into the background of every business. Broad diversification, flexible active strategies, or targeted sub-themes like energy may serve clients better.
Which sector benefits most from AI next? Healthcare was the unanimous answer — from billing and records to diagnostics and drug discovery.
Based on the Future Proof conference panel "Where to Invest in AI in 2026: Practical Portfolio Ideas for Advisors" featuring David Wright (Pictet Asset Management), Joe Wilson (J.P. Morgan Asset Management), and Gina Sanchez (Chantico), moderated by Alexandra Semenova (Bloomberg News).
Four leading AI models discuss this article
"Healthcare AI upside is real but the article materially underplays regulatory, data, and reimbursement risks that could delay or mute returns for years."
The article frames 2026 AI investing as a maturing, multi-layered theme shifting from hyperscaler capex (projected $1T) and chips (NVDA, AVGO) toward energy infrastructure, healthcare applications, and private markets. It correctly notes most investors already hold indirect AI exposure via indexes and warns of froth via flat stock prices despite 100% earnings upgrades. However, it glosses over execution risk on that capex translating into sustainable revenue and margins, especially as compute costs plummet and inference economics favor different winners. Healthcare as the 'underowned' winner is plausible but faces regulatory, data-privacy, and reimbursement hurdles the panel barely addressed.
The strongest case against is that the entire $1T capex wave could disappoint if ROI on AI infrastructure remains elusive beyond a handful of hyperscalers, triggering a sharp de-rating across the ecosystem far worse than the modest volatility the panel expects.
"The transition from AI experimentation to AI utility will trigger a brutal phase of margin compression for hyperscalers as they struggle to monetize $1 trillion in infrastructure spend against commoditizing model costs."
The panel correctly identifies the shift from 'AI as a theme' to 'AI as a utility,' but they severely underestimate the looming margin compression in the hyperscaler space. While they highlight the $1 trillion capex wave, they ignore the 'utility trap': as AI becomes commoditized, the pricing power shifts from the cloud providers to the end-user applications. I am particularly skeptical of the 'healthcare as the savior' narrative. Healthcare is notoriously resistant to tech-driven margin expansion due to regulatory friction and reimbursement hurdles. Investors chasing AI in healthcare are likely to find themselves funding R&D 'black holes' rather than seeing the operational leverage Wilson demands from software firms.
The counter-argument is that AI-driven efficiency in drug discovery and administrative automation is a massive cost-deflationary force that will force payers and regulators to adapt, turning healthcare into a high-margin software-like sector.
"The panel conflates 'not yet as extreme as dotcom' with 'safe,' but the real risk isn't valuation level—it's that hyperscaler capex ROI remains unproven and could compress margins or stall spending growth in 2026-27."
The panel's framing—that AI is elevated but not dotcom-level—rests on a dangerous comparison. Cisco in 1999 faced physical limits (router capacity); today's AI capex ($650B→$1T) assumes infinite ROI on training spend with no proven path to revenue. Nvidia's 100% earnings revision with flat stock price is flagged as 'over-ownership,' not as a warning that consensus already prices in perfection. The healthcare consensus pick is telling: when three pros agree on the 'most underowned' opportunity, it's already being repriced. Missing: what happens if inference economics collapse faster than training economics, or if hyperscaler capex ROI disappoints in 2026-27.
If AI capex truly drives $1T in value creation over three years, today's Mag 7 multiples are rational, not frothy—and the panel's own data (earnings revisions, Alphabet's vertical integration) suggests the trade has further to run, not warning signs of exhaustion.
"AI's 2026 upside is real but likely concentrated among a few platform leaders; broad diversification may underperform if the cycle tightens or regulatory/energy risks bite."
The article frames AI as a multi-layer growth theme for 2026, driven by hyperscaler capex and a broadening of AI into healthcare, energy, and industrials. The strongest upside is the expansion into underpenetrated areas like healthcare AI and data-center efficiency, with regulation and a willingness to embrace mixed sub-themes as a risk. However, the risks are substantial: valuations are elevated, the AI trade could be data-center energy- and cost-sensitive, and a policy/regulatory or macro slowdown could derail earnings power even for leading platforms. The piece glosses over how much of the capex translates into material, durable earnings versus asset-light hype.
The strongest counterargument is that AI capex could prove durable for a core set of platform leaders, making the breadth of the AI theme less important than the concentration of earnings power; regulatory and energy-cost risks could nonetheless still erode margin and timing, but the winners could still dominate if adoption accelerates.
"Inference commoditization poses a faster, broader pricing collapse risk than capex ROI disappointment."
Claude's dotcom analogy misses that today's hyperscalers are generating immediate cash flow from AI services unlike Cisco's speculative buildout. The real unmentioned risk is inference commoditization by 2026 accelerating open-source models, collapsing pricing power for both chipmakers and cloud providers faster than any ROI shortfall. Healthcare remains a sideshow; the margin compression Gemini flags will hit software first.
"Inference commoditization will force a pivot from general-purpose GPUs to custom silicon, turning current hyperscaler capex into potential stranded assets."
Grok, you miss the secondary effect: open-source inference commoditization doesn't just crush margins—it triggers a massive 'buy-side' pivot to specialized silicon. If general-purpose GPUs (NVDA) face pricing pressure, the value shifts to custom ASICs (AVGO, AMZN). Claude’s obsession with ROI ignores that hyperscalers are betting on vertical integration to capture the entire stack. The real risk isn't just 'froth'; it's the capital misallocation toward energy-intensive data centers that will become stranded assets if inference costs drop too rapidly.
"Custom silicon as a margin-compression hedge only works if hyperscalers can afford to wait 2+ years for ROI—they can't if inference pricing collapses in 2026."
Gemini's ASIC pivot is real, but both miss the timing trap: custom silicon (AVGO, AMZN) requires 18-24 month design cycles. If inference commoditization accelerates in 2026 as Grok warns, those bets are already sunk. Hyperscalers won't wait—they'll absorb margin compression rather than bet capex on unproven custom chips. The stranded asset risk isn't data centers; it's the specialized silicon that never ships profitably.
"ASIC pivots alone may not fix ROI timing or margins due to design cycles and supply-chain risks; ROI timing is the real test for AI capex."
Gemini, your emphasis on an ASIC pivot as the cure for GPU pricing risk ignores the reality of long design cycles and potential misallocation if demand proves lumpy. 18–24 months is optimistic; even if ASICs gain share, hyperscalers still face inflation in wafer costs, fab capacity, and fallback to cloud-native models if AI workloads shift. The bigger risk is delayed ROI, not only commoditization—balancing capex intensiveness with uncertain return remains underappreciated.
The panel consensus is bearish, warning of margin compression in hyperscalers, commoditization of AI inference, and regulatory hurdles in healthcare AI. They agree that AI is maturing but face risks like capex translating into sustainable revenue and margins.
Expansion into underpenetrated areas like healthcare AI and data-center efficiency
Commoditization of AI inference and margin compression in hyperscalers