The four largest hyperscalers are on track to spend a combined $725 billion on AI infrastructure in 2026, up roughly 77% from about $410 billion in 2025CITE:E1. Individual 2026 guidance ranges from Microsoft at $175–190 billion, Alphabet (Google) at $195–205 billion, Amazon at $200–220 billion, to Meta at $130–145 billionCITE:E1. Meta's guidance marks a near-doubling from its 2025 actual spend of $72.2 billion, one of the sharpest increases among the fourCITE:E2.
| Company | 2025 Actual | 2026 Guidance |
|---|---|---|
| Microsoft | — | $175B–$190B |
| Alphabet | — | $195B–$205B |
| Amazon | — | $200B–$220B |
| Meta | $72.2B | $130B–$145B |
| Combined (4 hyperscalers) | ~$410B | ~$725B |
NVIDIA's own results are the clearest evidence that capex dollars are converting into chip revenue. For its fiscal 2027 second quarter (the period ended July 26, 2026), NVIDIA reported data center revenue of $89 billion, up 117% year over year, and guided next-quarter revenue to $108 billionCITE:E3.
Bain estimates the AI industry will need about $2 trillion in new annual revenue by 2030 to sustain the current pace of investment, and even after accounting for AI-driven cost savings, projects a shortfall of roughly $800 billion per yearCITE:E4. Bain's figures are a consulting estimate of a future gap, not a report of actual results.
Investor Michael Burry has accused hyperscalers of stretching out depreciation schedules on AI chips and servers to inflate reported profitsCITE:E5. By his own calculations, this understates depreciation by a cumulative $176 billion across 2026–2028, and could overstate Oracle's and Meta's 2028 profits by roughly 27% and 21% respectivelyCITE:E5. These figures are Burry's personal estimates rather than company disclosures.
NVIDIA, OpenAI, and Oracle sit at the center of a financing loop that critics compare to dot-com-era vendor financingCITE:E6. NVIDIA has announced a commitment of up to $100 billion in OpenAI, though its actual participation in one funding round was closer to $30 billion, and the deal was reported paused as of January 2026CITE:E6. Separately, OpenAI has signed a data center contract with Oracle worth up to $300 billion, while Oracle in turn purchases chips from NVIDIA, closing the loopCITE:E6. Bloomberg's compiled deal-mapping analysis notes the terms of these arrangements remain in fluxCITE:E6.
NVIDIA CEO Jensen Huang argues that demand is genuine because "compute is revenue," pointing to accelerating orders as evidenceCITE:E7. JPMorgan Asset Management separately argues today's AI spending is funded mainly from hyperscalers' own cash flow rather than external debt, and notes that data center vacancy sits near just 1.6%, indicating supply is not outrunning demand — a structural difference from the externally financed dot-com bubbleCITE:E7.
Even OpenAI CEO Sam Altman acknowledges the AI buildout could produce both large winners and large losers. Altman has said "someone is going to lose a phenomenal amount of money," while also expecting many participants to make a phenomenal amount, and has described AI adoption as still in its early stagesCITE:E8.
The evidence points in two directions at once: capex is growing 77% year over year toward $725 billion, and NVIDIA's data center revenue is growing even faster at 117%, which is the strongest near-term case against a bubble. But Bain's $800 billion annual shortfall projection for 2030, Burry's $176 billion depreciation allegation, and the circular financing linking NVIDIA, OpenAI, and Oracle all describe risks that would only surface later — set against NVIDIA's and JPMorgan Asset Management's counter-evidence of cash-funded spending and 1.6% data center vacancy. Even Altman, sitting inside the industry, frames the outcome as unresolved rather than settled in either direction.
The four largest hyperscalers are on track to spend a combined $725 billion on AI infrastructure in 2026, up roughly 77% from about $410 billion in 2025CITE:E1.
NVIDIA's own results are the clearest evidence that capex dollars are converting into chip revenue.
Bain estimates the AI industry will need about $2 trillion in new annual revenue by 2030 to sustain the current pace of investment, and even after accounting fo…
Investor Michael Burry has accused hyperscalers of stretching out depreciation schedules on AI chips and servers to inflate reported profitsCITE:E5.
The starkest number here is the gap between two growth rates: hyperscaler capex is rising 77% year over year toward $725 billion, while NVIDIA's data center revenue backing it grew 117% — near-term chip demand is, for now, outrunning even the spending itself. That's exactly why Bain's $800 billion annual shortfall projection for 2030 matters more than any single quarter: it's a claim about five years out, not about today. The nearer-term test is Michael Burry's $176 billion depreciation estimate and his claim that Oracle's and Meta's 2028 profits could be overstated by 27% and 21% — accounting assumptions that are checkable well before 2030 arrives. The single figure worth watching next is whether NVIDIA's guided $108 billion in next-quarter revenue actually lands, since a miss there would undercut the demand-side case for the entire buildout.
SK Hynix (SK海力士) is planning a ₩10 trillion plant investment in Japan, a move that coincided with gains in two Japan-focused semiconductor ETFs. The plan follows a first half of 2025 in which the company posted ₩39.87 trillion in revenue, ₩15.10 trillion in net profit, and a debt-to-equity ratio that fell to 48.13% from 62.15% at the end of 2024.
US Commerce Secretary Howard Lutnick is pushing a new round of semiconductor tariffs that would tie tariff-free import quotas to how much chip manufacturing a company commits to build inside the United States, a design that could extend duties beyond the advanced processors already taxed since January 2026 to AI servers, laptops and gaming consoles.
Taiwan's benchmark TAIEX closed at 46,331.45 points on August 28, up 356.23 points and above the 46,000 threshold, on turnover of roughly NT$1.023 trillion. TSMC rose NT$10 to NT$2,420, and 36 listed stocks hit their daily limit after NVIDIA's earnings beat lifted sentiment at the open.