The Price of AI in 2026

The AI trade has evolved. Markets are no longer simply “buying AI” as a theme – they are underwriting a multi-year infrastructure buildout. The key question now isn’t whether AI matters. It’s whether the cash flows arrive quickly enough to justify what investors have already paid.

Valuations remain elevated. The S&P 500 forward P/E moved from roughly 19.5x in early 2024 to above 22x through 2025, and is still sitting around 21–22x in February 2026. On a cyclically adjusted basis, CAPE is hovering close to 40. That’s historically rare territory, comparable to the late-1990s period when markets were pricing in a decade of structural growth upfront.

This doesn’t automatically mean “bubble.” But it does mean expectations are extremely high.

Concentration – The Hidden Leverage in the Index

The structure of the market amplifies this risk. The top ten stocks now represent roughly 37% of the S&P 500. In a cap-weighted index, that concentration matters enormously. When leadership works, the index looks stable and strong. When leadership stalls, the entire market feels fragile.

In both 2024 and 2025, the largest tech names accounted for a disproportionate share of total returns – at times over 40–50% of index gains. Narrow breadth can persist longer than people expect, but it increases left-tail risk. Diversification becomes less effective when a handful of stocks drive performance.

There is also a valuation mismatch building beneath the surface. Technology represents roughly 30%+ of index market cap but a meaningfully smaller share of aggregate earnings. The premium rests on continued outgrowth. If that growth slows – even modestly – multiples don’t have much room to expand further.

The Capex Supercycle

The defining feature of 2026 is hyperscaler capital expenditure. The numbers are extraordinary.

Amazon is guiding toward roughly $200bn of capex this year. Alphabet is planning around $175–185bn. Meta expects $115–135bn. Microsoft’s recent quarterly capex implies an annualised run-rate north of $150bn if sustained. Combined, these four companies alone are expected to invest over $600bn in 2026, overwhelmingly directed at AI infrastructure: data centres, GPUs, networking, cooling, and power capacity.

This is the core stress test.

So far, revenue delivery has broadly justified the spending. AWS grew in the mid-20s year-on-year in recent quarters. Google Cloud has been expanding even faster. Azure remains near 40% growth. Nvidia’s data centre revenue continues to scale at extraordinary rates. Meta and Amazon are still compounding top-line growth in the teens or higher.

The issue isn’t whether AI generates revenue. It clearly does. The issue is incremental return on capital. At these spending levels, even slight deceleration in growth can compress free cash flow meaningfully. Markets are starting to differentiate between “AI spend with visible monetisation” and “AI spend without clear payback.”

Private Markets – Even More Aggressive

If public markets look stretched, private markets look even more enthusiastic. AI has captured the majority of U.S. venture deal value over the past year, with funding concentrated into fewer, larger rounds. Frontier model companies are raising at valuations that imply enormous future profitability.

Capital is effectively front-loading a decade of growth.

This doesn’t necessarily imply mispricing. But it does mean liquidity conditions and exit windows matter. If IPO markets stay open and revenue scales quickly, valuations hold. If the window shuts, multiples can reset quickly.

Productivity – Micro Evidence, Macro Lag

The productivity argument is credible. Controlled studies show double-digit improvements in task productivity when generative AI tools are deployed in customer support and knowledge work. That’s real.

But macro data hasn’t yet shown a clean step-change. Multifactor productivity has improved modestly, not explosively. That gap may simply reflect diffusion. Historically, new technologies boost productivity only after firms redesign processes and integrate complementary systems.

The bull case rests on that diffusion phase accelerating.

Dot-Com Parallels – And Differences

The comparison to 2000 is unavoidable. Then, markets priced infrastructure buildout and future demand that failed to materialise quickly enough. Today’s leaders are very different: they are profitable, cash-generative incumbents, not pre-revenue startups.

What rhymes is infrastructure overshoot risk. Fibre in 2000. Compute and power in 2026.

With the 10-year Treasury around 4% and policy rates still restrictive relative to the 2010s, discount rates matter. A 50–100bp rise in real yields would compress long-duration multiples even if earnings remain intact. At current valuations, sensitivity to rates is high.

The 2026 Stress Test

The investment debate now comes down to two variables:

1. Do incremental AI revenues scale fast enough to justify the capex wave?

2. Do discount rates stay contained?

If growth holds and productivity diffusion broadens, today’s multiples can be sustained. If capex persists while growth slows, free cash flow compresses and multiples re-rate lower – especially in a concentrated index.

The base case probably isn’t “bubble collapse” or “permanent revolution.” It’s something in between. AI is clearly transformative. But markets have already priced in a significant portion of that transformation.

In February 2026, the burden of proof sits with earnings.

Amazon.com, Inc. (2026) Amazon.com Announces Fourth Quarter Results. Press release, 5 February.

Brynjolfsson, E., Li, D. and Raymond, L.R. (2025) ‘Generative AI at Work’. Quarterly Journal of Economics.

Bureau of Industry and Security (2025) BIS announces updates to strengthen export controls on advanced computing and semiconductor manufacturing items. Press release (U.S. Department of Commerce), 2 December.

CB Insights (2026) State of AI 2025 Report. 13 January.

CoreWeave, Inc. (2025) CoreWeave Announces Pricing of Initial Public Offering. Press release, 27 March.

European Commission (2026) EU AI Act Service Desk / implementation guidance materials.

Federal Open Market Committee (2026) Monetary policy decision statement. 29 January.

Federal Register (2026) Final rule: Implementation of additional export controls: certain advanced computing and semiconductor manufacturing items; supercomputer and semiconductor end use; revisions and clarifications. 15 January.

Federal Reserve Bank of St. Louis (2026) FRED series DGS10: Market Yield on U.S. Treasury Securities at 10-Year Constant Maturity (observation 2026-02-12).

Gurufocus (2026) S&P 500 Shiller CAPE Ratio (value as of 2026-02-01).

International Monetary Fund (2026) World Economic Outlook Update, January 2026: Global Economy – Steady amid Divergent Forces. 19 January.

Microsoft (2025) Microsoft Annual Report 2025.

Microsoft (2026) Microsoft Cloud and AI strength drives second quarter results (FY26 Q2). 28 January.

Noy, S. and Zhang, W. (2023) Experimental Evidence on the Productivity Effects of Generative AI. Working paper.

Organisation for Economic Co-operation and Development (2024) ‘What we know about GenAI and productivity’. OECD Blog, 22 May.

Reuters (2025) ‘AI bubble isn’t near a peak; it’s only “base camp”’. 22 October.

Reuters (2025) ‘Complacency and cash: S&P 500’s top 10 hits 37.3% weight’. 22 July.

Reuters (2025) ‘Nvidia Q3 earnings beat estimates’. 20 November.

Reuters (2026) ‘Alphabet says capital spending in 2026 could double, cloud business booms’. 4 February.

Reuters (2026) ‘Anthropic clinches $380bn valuation after $30bn funding round’. 12 February.

Reuters (2026) ‘Big Tech’s quarter in four charts: AI splurge and cloud growth’. 6 February.

Reuters (2026) ‘Meta boosts annual capex sharply on superintelligence push’. 28 January.

Reuters (2026) ‘Microsoft capital spending jumps, cloud revenue fails to impress’. 28 January.

Reuters (2026) ‘OpenAI CFO says annualised revenue crosses $20 billion’. 19 January.

Reuters (2026) ‘US FTC ramps up scrutiny of Microsoft over AI, cloud practices’. 13 February.

Slickcharts (2026) S&P 500 holdings and weights. Accessed mid-February 2026.

U.S. Bureau of Economic Analysis (2025) Gross Domestic Product (Q3 2025 updated estimate).

U.S. Bureau of Labor Statistics (2025) Multifactor Productivity, 2024.

U.S. Bureau of Labor Statistics (2026) CPI-U January 2026 release tables.

White House (2025) Ensuring a National Policy Framework for Artificial Intelligence. Executive order, 11 December.

The 2026 Stress Test: From AI Vision to Infrastructure Reality