Will aggregate hyperscaler AI capex growth decelerate below +20% YoY by Q2 CY2026?

The $120B upward revision to CY2026 CSP capex expectations (from ~$600B to ~$700B) is the most material update. For capex growth to decelerate below +20% YoY in Q1 or Q2 CY2026, quarterly spending would need to fall well below the annualized $700B trajectory. CY2025 aggregate capex was in the $550-600B range, so a +20% threshold implies ~$660-720B annualized — the $700B trajectory is right at or above this. NVIDIA's $95.2B in supply commitments extending into CY2027 represents forward-committed hyperscaler demand that directly feeds through as capex. Jensen's 'compute is revenues' framing and Meta's commitment to 'millions of Blackwells and Rubins' signal no near-term capex discipline. With only Q1 and Q2 CY2026 remaining (reports in April and July), and hyperscaler earnings guidance already pointing to sustained acceleration, deceleration below 20% YoY is unlikely.

The question resolves based on Q1 and Q2 CY2026 hyperscaler earnings. Q1 CY2026 ends March 31, meaning capex decisions for this quarter are already locked in — data center construction and GPU procurement operate on multi-quarter lead times. Q2 CY2026 ends June 30 and is similarly constrained by committed infrastructure plans. The $700B aggregate figure represents analyst consensus built from individual hyperscaler guidance and disclosed commitments. Microsoft guided CY2026 capex above $80B, Meta guided $60-65B, Amazon and Google both increasing. These individual commitments sum to well above +20% YoY growth. The only scenario for deceleration below 20% would require a synchronized, unannounced pullback by multiple hyperscalers — contradicted by every available data point. The prior context's concern about ROI pressure (10:1 infrastructure-to-revenue ratio) remains structurally valid but has not manifested in any hyperscaler reducing capex guidance.

This market's resolution is now only ~5.5 months away (August 15, 2026), and the data window is even tighter — Q1 CY2026 reports in April, Q2 CY2026 reports in July. The committee's original concern about unprecedented capex-to-revenue ratios (45-57%) has been partially addressed by the agentic AI inflection narrative — Jensen cited Anthropic's 10x revenue growth as evidence that AI revenue is inflecting upward, which would improve the infrastructure-to-revenue ratio. More concretely, NVIDIA's Q4 DC revenue of $62.3B (+75% YoY) and $78B Q1 FY2027 total revenue guidance are direct proxies for hyperscaler spending. These figures show acceleration, not deceleration. The Anthropic $10B investment, Groq $13B deal, and Meta's 'millions of Blackwells and Rubins' commitment all represent incremental demand on top of already elevated baselines. The mathematical bar for below +20% YoY growth given the $700B trajectory and $550-600B CY2025 base is not achievable absent a macroeconomic crisis.

The updated data overwhelmingly supports continued hyperscaler capex acceleration through Q2 CY2026. The $120B upward revision to CY2026 capex expectations since January is the clearest signal — expectations are rising, not falling. NVIDIA's own results serve as a leading indicator: $62.3B in Q4 DC revenue and $78B Q1 FY2027 total guidance represent hyperscaler orders already placed. The $95.2B supply commitment figure is unprecedented and extends into CY2027, indicating hyperscalers have locked in spending well beyond the resolution window. Individual hyperscaler signals are uniformly expansionary: Meta committed to 'millions of Blackwells and Rubins,' Microsoft guided $80B+ CY2026 capex, and Alphabet and Amazon continue accelerating. No major hyperscaler has signaled capex optimization or efficiency pivots. The resolution criteria require EITHER Q1 or Q2 CY2026 to show <20% YoY growth — but both quarters appear solidly above this threshold based on current trajectories.

The prior prediction at 11% was made with Q3 FY2026 data showing aggregate hyperscaler capex expectations at ~$600B. Since then, expectations have risen by $120B to ~$700B, representing a 20% upward revision in the denominator. This shift alone should reduce the probability — the base rate of deceleration below +20% is lower when the trajectory is accelerating. The committee's original concerns about ROI pressure (only 25% of AI initiatives delivering expected ROI) and unprecedented capex-to-revenue ratios remain valid structural risks, but they are medium-term concerns that have not translated into near-term capex cuts by any hyperscaler. Jensen's 'compute is revenues' framing in Q4 represents the strongest permanence argument yet — positioning AI capex as operating expense rather than discretionary investment. With Q1 CY2026 already underway and Q2 CY2026 planning locked in, the window for a sub-20% outcome has effectively closed.

While I agree the probability is low, I weight the remaining uncertainty slightly higher than my peers. The resolution depends on actual reported capex from 4 companies (Microsoft, Alphabet, Amazon, Meta), and there are measurement subtleties: how capex is categorized (total vs. AI-specific), whether capitalized leases are included, and whether one-time land/building acquisitions inflate the CY2025 base (making YoY comps harder). If CY2025 had an unusually high base quarter due to one-time infrastructure buildouts, CY2026 might show slower growth even with absolutely higher spending. Additionally, there is a non-trivial tail risk of a macroeconomic shock (tariff escalation, financial market disruption) that could cause one or more hyperscalers to delay capex in Q2 CY2026. The probability is low but not negligible — I anchor at 8% to account for measurement and macro tail risks.

CSP capex approaching $700B for CY2026 vs estimated $550-600B in CY2025 implies aggregate growth of ~17-27%. Even at the low end of this range, growth is near the 20% threshold. But the $120B upward revision since January shows the trajectory is accelerating, not decelerating. NVIDIA's $78B Q1 FY2027 guidance and $95.2B supply commitments confirm hyperscaler ordering continues at record levels. With only Q1 and Q2 CY2026 remaining before resolution, and all major hyperscalers guiding capex higher, the probability of deceleration below 20% is very low.

This is the most straightforward of the NVDA markets to assess. The resolution asks whether combined capex from 4 hyperscalers grows <20% YoY in Q1 or Q2 CY2026. Every available data point — $700B aggregate trajectory, individual hyperscaler guidance, NVIDIA's record backlog, Meta's 'millions of Blackwells' commitment — points to growth well above 20%. The prior aggregate of 11% already reflected this as unlikely; the Q4 earnings update with $120B upward capex revision makes it even less likely. A YES resolution now requires a black swan macro event within 4 months.

The capex deceleration thesis was the committee's primary structural concern for NVIDIA's revenue durability, but the Q4 data shows the opposite trend — capex expectations are accelerating. The $95.2B in NVIDIA supply commitments extending into CY2027 is particularly significant as a forward indicator of hyperscaler intent. Jensen's framing of compute as equivalent to revenue ('without compute, there's no way to generate tokens') positions AI capex as non-discretionary, reducing the likelihood of cuts. The agentic AI inflection narrative, if validated by Q1/Q2 CY2026 data, extends the capex cycle further. Probability is low, reflecting only tail risk of exogenous macro shock or measurement artifact.