CFO Anat Ashkenazi tied Alphabet’s increase to 'an acceleration in the delivery of capacity to meet growing demand' in the company’s earnings call, with roughly 60% of this quarter's spend going into servers and the rest into datacentres and networking.  

Amazon's increase was pinned by chief financial officer Brian Olsavsky largely on 'the higher cost of memory', which pushed 2026 cash capex from around US$200bn to roughly US$220bn without any change to the underlying build plan.  

Meta's gap between actual H1 capex and full-year guidance is the widest of the four, with its implied second half of US$79bn to US$94bn close to double its US$50.9bn first-half actual figure. This suggests the company's AI infrastructure rollout remains more back-end loaded than its peers.  

TMT Finance is currently tracking at least 11 rumoured, live or recently closed raises for projects linked to Meta across the US and Canada, representing what could amount to near US$80bn paper in the market. Management attributed the increase to higher component costs and datacentre construction schedules, although the guidance also implies capacity additions will accelerate materially in the second half.

Microsoft extended the useful life of its datacentres and office buildings from 15 to 25 years, shifting more future leases off the capex line entirely so its apparent reduction reflects this accounting change rather than a meaningful reduction in investment. CFO Amy Hood said spending plans 'remain unchanged' in economic terms in the company’s earnings call on 29 July. The change applies only from Microsoft's 2027 financial year, so the US$72.9bn first-half actual is unaffected.  

Funding the gap  

Operators are drawing on an increasingly broad range of financing sources to fund AI infrastructure, although there seems to be no template for the balance of debt, equity and structured financing as hyperscalers vary in their usage.  

Alphabet reported the largest financing programme among the group; its debt balance has increased from about US$16bn a year ago to roughly US$100bn. The company also announced an US$80bn equity raise in June, comprising a US$30bn underwritten offering of common stock and mandatory convertible preferred shares, a US$40bn at-the-market  programme and a US$10bn private placement with Berkshire Hathaway, according to an SEC filing.  

Meta generated free cash flow of US$784m against US$31.9bn of operating cash flow. It took on US$24.9bn of new debt and stopped buying back shares entirely, while leaning on partnerships such as its new BlackRock-backed venture for a 1GW datacentre in El Paso, Texas.  

Microsoft, however, is still guiding to positive free cash flow, relying less on external financing than its counterparts; it booked a US$3.2bn gain this quarter on its Anthropic stake alone.  

In comparison to the top four, Oracle is raising around US$40bn in debt and equity in its 2027 financial year but is also using prepayments and 'bring your own hardware' deals, where the customer supplies its own chips and Oracle provides the building, power and operations, to keep its own cash outlay down. Those contracts totalled US$75bn at the company's latest earnings update.  

Backlog appears to be growing faster than capex across the sector. Oracle's remaining performance obligations increased 363%; Microsoft's backlog rose 84%, or only 25% excluding its OpenAI contract. The expanding use of customer prepayments, vendor financing and long-term capacity agreements suggests hyperscalers are increasingly securing customer commitments before committing additional capital.  

Cash flow under pressure  

The debate has moved beyond the durability of hyperscaler capex to the pace at which those investments are translating into financial returns. Microsoft generated US$19.6bn of free cash flow even as capex rose roughly 70% year-on-year and the company expects to remain free cash flow positive through its 2027 financial year, according to the latest earnings call.  

In contrast, Alphabet and Meta both went negative or near zero, per their earnings calls. Alphabet reported its first negative free cash flow since its 2004 IPO, US$5.9bn, with Meta generating a marginal US$784m positive. Both companies raised capex guidance in the same breath; in Alphabet’s earnings call, Ashkenazi said the company’s free cash flow 'will remain under pressure' while it keeps investing.  

Amazon does not provide explicit free cash flow guidance, but its own reported measure has also turned negative. Trailing-twelve-month free cash flow swung to a US$7.6bn outflow in the second quarter, its first negative reading since 2023, even as trailing operating cash flow rose 33% to US$161.4bn. The company attributed the shift entirely to the capex increase rather than any weakening in the underlying business. Chief executive Andy Jassy instead framed returns over a longer horizon: datacentres, once built, generate revenue for '30-plus years', while servers and networking equipment break even in less than three years against a five-to-six-year useful life.  

The divergence underscores how differently hyperscalers are managing the trade-off between elevated AI investment and free cash flow. Microsoft continues to generate sufficient cash flow to support its spending, while Alphabet, Amazon and Meta are accepting greater near-term pressure on free cash flow as AI spending accelerates.  

The outlook for 2027  

Management commentary suggests AI infrastructure spending will remain elevated into 2027, with little indication on the earnings calls that investment is nearing a peak.  

Amazon's Jassy said the company's capacity constraints are likely to persist through 2027, with projected demand in 2028 already informing infrastructure planning. Oracle has provided the clearest outlook, indicating capital expenditure of roughly US$70bn in fiscal 2027. Alphabet also signalled a further increase in spending, while Meta and Microsoft have declined to provide specific guidance, with Meta CFO Susan Li describing capital planning as ‘highly dynamic’.  

Nvidia's [NASDAQ: NVDA] CFO Colette Kress gave the most bullish framing in the company's earnings call, citing third-party analyst forecasts of hyperscale capex topping US$1 trillion in 2027 and total AI infrastructure spending reaching US$3 trillion to US$4 trillion per year by 2030.  

For now, hyperscalers continue to raise investment guidance. How the scale implied by current spending plans through 2027 will be financed remains to be seen, as the market arms itself with an ever-increasing array of tools and structures to fulfil demand.