Alphabet capex: watch the build, not the guide
A reasonable chance Alphabet raises its 2026 capex this week (about 40%); base case reaffirm at $180-190B. Either way, the deployment matters more than the guide. Locked before the print, graded Wednesday.
We believe there is a reasonable chance Alphabet raises its 2026 capex projection to $190-200B ($195B midpoint) at this week's earnings call. Our model puts the chances at roughly 40%. The forecast weighs four inputs.
| Model input | Current signal |
|---|---|
| Historical raise rate at this print | Favors reaffirmation |
| Spending headroom within the current range | Favors reaffirmation |
| Latest management signal (June reiteration) | Favors reaffirmation |
| Last year's Q2 precedent | Favors a raise |
Our base case is that it reaffirms at $180-190B.
Alphabet reports Wednesday, July 22, after market close. The call begins at 4:30pm ET.
Either way, an unchanged guide is not bearish for the AI infrastructure chain unless deployment language also weakens.
The key question is not whether the guide moves by $5B. It is what management says the spending is buying.
Alphabet established the scale and direction of its multiyear plan in June. It raised $84.75B to finance AI infrastructure, reiterated $180-190B for 2026, and flagged a significant increase in 2027. Wednesday mainly tells us how much of that plan enters reported 2026 capex.
1. Will Alphabet change the guide? We lean toward reaffirmation, though a raise to $195B or above is a reasonable possibility (about 40%). This is the cleanest question to grade, but the least consequential economically: the gap between the current ceiling and a $195B midpoint is small relative to the total program.
2. Is Alphabet executing a larger multiyear plan than the 2026 range alone implies? High confidence: yes. The evidence already points beyond the current range: $84.75B of new equity financing, Cloud backlog near $460B, Cloud revenue of $20B (up 63%), and explicit guidance for a significant increase in 2027. This barely depends on Wednesday's headline number.
3. Which suppliers actually receive the spending? The real investing question. A dollar of Alphabet capex can flow into accelerators, memory, advanced packaging, networking, power, land, or construction, and those categories carry very different read-throughs for suppliers and for Alphabet's own margins.
We read the build from both ends: what Alphabet says it will spend, and what the physical supply chain shows can actually be deployed.
| Evidence layer | What moves first |
|---|---|
| Company-specific | financing, Cloud backlog, management language |
| Supplier-level | TPU and NVIDIA demand, HBM and DRAM, CoWoS packaging, networking and power |
| Hard volume confirmation | Taiwan monthly revenue, shipped systems, foundry utilization |
By the time the build reaches reported capex, the physical supply chain has already moved. That is why we do not stop at the guidance number.
| Outcome | Classification | Investment read-through |
|---|---|---|
| ~$200B, driven by servers or faster compute | Volume-led compute | Positive for NVDA, MU, SK Hynix, AVGO |
| ~$200B, driven by land, campuses, power, or construction | Long-duration infrastructure | Positive for VRT, GEV, utilities; weaker near-term semis |
| Raise driven by component prices or depreciation | Cost inflation | Positive for supplier pricing; risk to Alphabet margins |
| Unchanged guide, weaker server mix or softer capacity language | Genuine deceleration | Broad negative for the AI hardware chain |
The headline delta is modest; management's explanation determines the read-through. A raise driven by land or construction is less bullish for near-term semiconductor demand than an unchanged guide paired with stronger deployment, supply-constraint, and 2027 spending language. Classify the spending before trading the headline.
- Bullish: server or compute deployment pulled forward; capacity remains supply-constrained; 2027 spending steps up further.
- Neutral: guide unchanged with no material language shift; incremental spending in land or construction; accounting or leasing treatment changes.
- Bearish: deployment delayed or deferred; capacity language eases; utilization or server mix weakens.
Our constraint map still reads the AI infrastructure cycle as tight in level, though the incremental signal is mixed.
| Node | Score | Current read |
|---|---|---|
| DRAM (DDR5) | 86 | Critical, tightest node, no visible relief |
| HBM3e | 80 | Critical, eased from a higher peak |
| N3 wafers | 79 | Tight, elevated |
| N2 wafers | 70 | Tight, capacity build underway |
| CoWoS packaging | 60 | Tight, elevated, indirect read |
These corroborate industry-level deployment pressure but do not isolate Alphabet's share of demand.
The implication for Wednesday is straightforward. If Alphabet raises capex without attributing it to higher memory or component costs, the more likely explanation is additional deployment volume, not inflation. The constraint picture today supports expansion, but it does not support a broad cost shock.
The earliest confirmation will not come from Alphabet's reported capex. It will appear first in Taiwan monthly AI-server revenue, our earliest hard read on shipped systems, and in leading-edge foundry utilization, the clearest evidence that demand is converting into wafer starts.
The broader supply chain remains consistent with that view. TSMC raised its 2026 capex guide to $60-64B in July while continuing to describe advanced packaging as a customer constraint, a signal that leading-edge demand remains healthy. Recent hyperscaler history points in the same direction: Alphabet finished 2025 materially above its initial capex guide, and Meta followed a similar pattern.
Tessara read: the physical evidence still supports an expanding AI build, but not a fresh acceleration.
- The guidance forecast is wrong if Alphabet raises its midpoint to $195B or above. Binary, graded Wednesday.
- The Alphabet thesis (sticky guidance is not slowing deployment) is wrong if management cuts planned deployment, weakens capacity or utilization language, or shifts spending away from compute more sharply than expected.
- The broader house view (the AI build keeps accelerating) is wrong if two or more hyperscalers cut planned 2026 deployment, or several independent volume indicators roll over together. Lower prices alone do not falsify it: prices can fall because supply expands faster than demand while deployment keeps growing.
By the time the build reaches reported capex, the physical supply chain has already moved.