When Alphabet and Tesla report second-quarter earnings after the US market closes on Wednesday, with Intel following on Thursday, the headline numbers are almost the least interesting part. Investors have stopped debating how much the technology giants are spending on artificial intelligence and started pressing a harder question: when does the spending begin to pay for itself?

That question has already moved markets. AI chip stocks sold off sharply through the week, and on 17 July Apple briefly overtook Nvidia to reclaim its place as the world’s most valuable company — not because Apple surged, but because investors rotated away from the firms building AI towards one that sells finished products to customers. The reckoning is not confined to Wall Street. In boardrooms, finance chiefs are applying the same test to their own AI budgets, and marketing, where a large share of the early generative-AI money was spent, sits squarely inside it.

Why has the market stopped rewarding AI spending?

Wall Street’s four biggest spenders — Alphabet, Microsoft, Meta and Amazon — are on course to commit close to $700bn to AI infrastructure this year, according to CNBC, and the scale of that figure is no longer the story. The story is what comes back. Free cash flow across the largest cloud builders has fallen even as their earnings have grown, because the capital is leaving faster than the returns arrive; the research group Epoch AI has calculated that the combined free cash flow of the biggest hyperscalers reaches zero this summer.

The market has read the signal. Nvidia, whose chips absorb much of that infrastructure spending, slid as investors moved into steadier names, a rotation rather than a wholesale retreat from equities. Apple’s 2026 gain of roughly 23%, against Nvidia’s 7%, captures the mood in miniature: investors are separating the companies that sell AI from the companies that still have to prove they can earn a return on it. This week’s results are the first close look at whether three of the largest capital-spending programmes in corporate history can describe a credible route from outlay to income.

Why is the same question landing on marketing?

CFOs have pulled marketing’s AI spending into the same frame. Deloitte’s 2026 marketing trends report describes finance chiefs placing marketing under heavier scrutiny and demanding measurable returns, and recommends that marketing leaders agree a single ROI model with finance and review the weakest fifth of spend each quarter. Marketers describe the shift in blunter terms: budgets now arrive “with an asterisk”, as one marketing leader put it to UserTesting, with boards asking a single question — where is the AI? — and expecting a number in reply.

The evidence that the money has worked so far is mixed, and marketing is not flattered by it. MIT’s Project NANDA study, The State of AI in Business 2025, found that 95% of enterprise generative-AI pilots produced no measurable impact on profit or loss, and noted that budgets clustered in sales and marketing even though returns there lagged back-office uses. Boston Consulting Group’s 2026 survey of 1,800 executives found just 26% had generated meaningful financial value from AI, and Gartner expects roughly 30% of the generative-AI projects begun in 2024 to be abandoned by the end of this year. The common thread in the research is not that the technology fails, but that most organisations deploy it without the data, integration or defined outcome needed to prove it worked. Where marketing gets those conditions right, the payoff is real: McKinsey has put the uplift for companies using AI well in sales and marketing at 10 to 20%.

What does the reckoning ask of marketers?

Marketers are being asked to do exactly what the hyperscalers are being asked to do: show the return, not the roadmap. That is a demand about evidence, not a verdict on the profession — a distinction the louder AI-and-jobs coverage tends to blur. When executives say AI is replacing work, the reporting rarely supports them. A Harvard Business Review survey of 1,006 global executives, led by Babson and MIT’s Thomas Davenport, found that only 2% of large headcount reductions were tied to AI actually being implemented, while about a fifth were made in anticipation of gains that had not yet arrived. Gallup’s first-quarter data for 2026 found that just 1% of laid-off US workers gave AI or automation as the reason.

What is often labelled an AI decision is, on inspection, a budget one. Meta guided to between roughly $115bn and $145bn of AI infrastructure spending this year while cutting around 8,000 jobs, redirecting salaries towards compute rather than replacing them with it. Goldman Sachs has found that only 11% of companies are cutting staff because of AI, while 47% are using it to raise productivity and revenue rather than to shed cost; the bank’s chief economist, Jan Hatzius, has framed AI’s effect so far as tilted towards lifting output rather than cutting costs. For a marketing manager, the practical reading is steadier than the headlines: the pressure now falling on marketing budgets is the same discipline being applied to Alphabet’s and Nvidia’s, and it rewards the same answer — a specific, measurable line from AI spend to pipeline or revenue.

Alphabet and Tesla report after Wednesday’s closing bell, and the market will be listening past the quarterly beats and misses for one thing: a credible account of when AI spending turns into AI earnings. Marketing teams will be asked for the same account on a far smaller budget, and — by Deloitte’s estimate that AI investments now take two to four years to pay back, against the seven to twelve months technology projects once did — rather sooner than the technology itself may allow.