Compounding Waves — How Each Tech Era Built the Substrate, and the Skills, for the Next

Narrative

The dominant story in the financial press from 2023 to 2026 is a capital-allocation event of historical scale whose productivity justification remains contested. Bloomberg Intelligence projects hyperscalers will commit more than $3.5 trillion in AI-related capital expenditure through 2030, and first-quarter 2026 earnings placed Amazon, Alphabet, Microsoft and Meta on track for a combined $650–700 billion in capex for the year — nearly double 2025 figures. Goldman Sachs, in its foundational 2023 research, estimated that generalised AI adoption could raise global GDP by 7 percent and lift US labour productivity by 1.5 percentage points annually; its 2024 'Too Much Spend, Too Little Benefit?' report then staged the counter-case, featuring MIT's Daron Acemoglu, who estimates the realistic TFP gain at 0.5–0.6 percent over a decade. Apollo chief economist Torsten Slok's observation — 'AI is everywhere except in the macroeconomic data' — has become the financial press's shorthand for this productivity paradox, and an NBER survey of 6,000 executives published in early 2026 found the vast majority see little operational impact, with AI usage averaging just 1.5 hours per week.

The financing architecture around the buildout is itself a subject of financial-press scrutiny. Man Group's November 2025 analysis documents how hyperscalers began moving liabilities off their balance sheets in 2025, into special-purpose vehicles, private credit funds and data-centre REITs — a structure that transfers risk to institutions that 'do not see themselves as betting on GPU cycles'. Bloomberg reported at least $178.5 billion in US data-centre credit deals struck in 2025 alone. CoStar analysis using company filings shows that Alphabet's 2026 capex guide outpaces its trailing twelve-month free cash flow, implying increasing reliance on external debt — the clearest available evidence that the question 'who pays for the next substrate layer?' is no longer hypothetical.

The training-data enclosure story is the financial press's most concrete illustration of the 'externalised harvest' frame. OpenAI's $60 million licensing deal with Reddit in May 2024, Google's equivalent agreement, and Reddit's subsequent lawsuits against Anthropic and Perplexity in 2025 establish a two-tier system: well-capitalised incumbents license wave-one data legally while smaller entrants face litigation for accessing the same public corpus. CNBC's October 2025 coverage of the Reddit v. Perplexity case quotes Reddit's chief legal officer describing 'an industrial-scale data laundering economy' — language that frames the issue not as copyright technicality but as a structural rent extraction from wave-one creators by wave-three operators. The IMF's March 2026 Finance and Development article adds the measurement dimension: official statistics record massive AI capital outlays but miss productivity spillovers, simultaneously overstating and understating AI's economic contribution.

On labour markets, the financial press reflects a genuinely contested empirical picture. Bloomberg Intelligence's 2025 bank-sector report estimates global investment banks could shed 200,000 jobs over three to five years; Goldman's own CIO simultaneously argues the human workforce will be 'amplified'. Brookings' March 2026 synthesis of the labour-economics literature finds no aggregate displacement through 2024–2025 but a 16 percent employment decline for workers aged 22–25 in AI-exposed occupations — an early-career concentration effect that Brynjolfsson, Chandar and Chen at Stanford describe as 'canaries in the coal mine'. The debate maps closely onto the earlier internet and cloud waves: both generated role-creation (webmaster, cloud engineer, data scientist) alongside displacement of routine clerical work, but the net employment effect of each took years to appear in aggregate data, consistent with Brynjolfsson's Productivity J-Curve framework.