Nasdaq Hits Record High: How Long Can the AI Boom Last?

The Nasdaq is at a record, and the spending behind it is already booked through 2026. The revenue that is supposed to pay for it arrives in 2028 on current plans. That leaves roughly a year, 2027, where the money has to come from lenders, and rates are going the wrong way. I'd say the boom lasts as long as credit stays open to the companies building it. A better model or a worse one matters less than that.

The spending is already in the national accounts. Four AI-related investment categories (equipment, software, research and development, and data-centre construction) added 1.30 percentage points to annualised US growth in the first quarter of 2025 and 1.16 points in the second (St. Louis Fed). That was 30% of second-quarter growth and 11% of third-quarter growth. Software's 0.57-point contribution in the second quarter was the largest on record, ahead of 0.44 points in 1997. IT and AI-related categories made up 28% of growth in 2000 (0.81 points). Over the first nine months of 2025 they made up 39% (0.97 points). AI's share of growth is now larger than IT's was at the 2000 peak.

The five largest US hyperscalers are set to spend more than $730 billion in 2026, against about $416 billion in 2025 (J.P. Morgan Asset Management). A wider group of six, adding Oracle and SpaceX, is projected to more than double its 2025 spending by 2027 (Yahoo Finance). The chart below uses S&P Global's figures for that six-firm group, so the totals are higher than the five-firm numbers above.

Hyperscaler capital expenditure, six firms, 2025-2027 (S&P Global projections)

The productivity data doesn't show a payoff yet. Task-level studies report time savings of 20% to 50%, but estimates of long-run aggregate productivity growth come in below 1% (BIS). Labour productivity in industries with high, medium and low exposure to AI has moved broadly together, so whatever gains firms see are not adding up across the economy (Federal Reserve). Adoption is wide and shallow. 51% of US businesses pay for AI tools, and 20% report using them in any business function (J.P. Morgan Asset Management).

So the funding matters more than the technology. Cumulative data-centre and AI investment is projected at about $5.3 trillion by 2030, and hyperscaler cash flow covers roughly $1.5 trillion of it (J.P. Morgan again). The remaining $3.8 trillion has to come from equity, bonds, securitised debt and private credit. Capex is already heading toward 40% of sales. On the projections, five of the six biggest spenders burn free cash flow in 2027 and only Microsoft stays positive (Yahoo Finance). The same projections put the turn in 2028, when capex flattens and revenue accelerates.

Borrowing for that gap is getting dearer. PCE inflation was 140 basis points above the Fed's 2% target at 30 June 2026 (Enterprise Bank). The Fed held at 3.50–3.75% and signalled it might raise before year end, and consensus expects one or two increases in 2026. Global inflation rose half a percentage point after the Middle East conflict, with plastics and fertiliser prices both up 50% (BIS).

Equity valuations assume none of this goes wrong. Calendar-year earnings growth is expected near 26%, against roughly 8% a year on average since 1960, and the same report calls the risk of holding that pace elevated (Enterprise Bank). The AI-exposed part of the market is about 58% of the S&P 500, with forward price-to-earnings ratios mostly in the 20s (J.P. Morgan). The top 10 names are about 41% of the index, against the high teens to low 20s for most of the past four decades. The rally has come from earnings, not rising multiples, which is better than the alternative. It also means a miss on earnings has nowhere to hide.

I see three things that would change this view. Cheaper money would help. So would faster revenue, since either lets the buildout pay for itself instead of being refinanced. A clear aggregate productivity signal would settle it in the boom's favour, and as of 2026 there isn't one (Federal Reserve). The bill may also be smaller than it looks, because hyperscaler spending includes non-AI items. If any of that changes, credit markets will show it before the technology does.

My read is that 2027 is the year to watch. Spending is committed, cash flow runs negative and the revenue is a year away.

Nothing here is financial, legal or investment advice. Every figure is published by its named source and carries the date it was published; figures may since have changed.