🟣 Sector Brief AI Infrastructure Published Apr 28, 2026 · Q2 2025 — Ongoing

AI Infrastructure
Capex Tracker

Where $600B is going. Eight consecutive quarters of upward revisions later, the Big Five hyperscalers have committed $660–720B in 2026 AI infrastructure spend — and Wall Street consensus has been too low every single time.

Combined 2026 Guidance
$660–720B
AI-Specific Share
~75%
AI Dollar Figure
$450–500B
Consensus Missed By
~30pts × 2yrs
📊
01 / 07
The Number That Matters
$660–720B
Combined 2026 hyperscaler capex guidance · Big Five

That number didn't materialize overnight. It's the result of eight consecutive quarters of upward revisions, with Wall Street consensus running too low every single time. Goldman Sachs projected $500B. The actual figure blew past it. For two straight years, estimates implied ~20% annual growth. Actual spending exceeded 50% both times.

Here's how each company's guidance evolved — from initial FY2025 print to today:

Hyperscaler Initial FY2025 Guidance FY2025 Actual FY2026 Guidance
Microsoft (MSFT) ~$80B ~$88–90B ~$120B+
Alphabet (GOOGL) ~$75B ~$91–93B $175–185B
Meta (META) $60–65B ~$70–72B $115–135B
Amazon (AMZN) ~$100B+ ~$125B ~$200B
Oracle (ORCL) ~$16B ~$35B ~$50B
Combined ~$430B ~$400–420B $660–720B

Sources: Company earnings calls Q1–Q4 2025; CreditSights 2026 estimates; The Register Feb 2026

The single most important number: 75%. That's the share of 2026 hyperscaler capex — roughly $450–500 billion — that is AI-specific infrastructure. Not general cloud. Not office real estate. GPU clusters, AI servers, liquid cooling, and the power grid to run them.

🎤
02 / 07
What Each Hyperscaler Just Said

Verbatim from the most recent earnings calls. Every major hyperscaler described the same condition: more demand than supply.

Microsoft (MSFT) — Q2 FY2026
Oct 30, 2025
"We will continue to increase our infrastructure investments to support the growing AI demand we're seeing." — Satya Nadella, CEO
Q1 FY2026 capex: $34.9B (+74% YoY) FY2026 annualized: $150B+ Transformer lead time: 128 weeks

CFO Amy Hood acknowledged "capacity constraints" — Microsoft has more AI orders than it can fulfill. ~$80B in unfulfilled Azure orders sitting in backlog because the company lacks sufficient grid power to energize GPUs already purchased.

Alphabet/Google (GOOGL) — Q4 2025
Feb 4, 2026
"We're seeing our AI investments and infrastructure drive revenue and growth across the board." — Sundar Pichai, CEO
FY2025 capex: ~$91–93B (revised up from $75B) FY2026 guidance: $175–185B Cloud backlog: $240B

Google Cloud revenue hit $17.7B in Q4 2025, up 48% YoY. FY2026 guidance is more than double 2025 actual, exceeding all peer hyperscalers. When asked what keeps him up at night, Pichai answered: compute capacity constraints and supply chain bottlenecks.

Meta (META) — Q4 2025
Jan 29, 2026
"Demands for compute resources across the company have increased even faster than our supply." — Meta CFO
Q4 capex: $22.14B (+49% YoY) FY2026 raised to: $115–135B External Oracle capacity: ~$20B

Meta has signed ~$20B in external Oracle cloud capacity on top of its own buildout — hedging against schedule risk and peak training demand. Prometheus project: a 1 GW supercluster targeting 500,000 GPUs in 2026.

Amazon (AMZN) — Q4 2025
Feb 5, 2026
"As fast as we install this AI capacity, it's getting monetized." — Andy Jassy, CEO
FY2025 capex: ~$125B FY2026 guidance: $200B Trainium/Inferentia ARR: $10B+

Amazon is the single largest AI infrastructure spender. AWS now runs at a $142B annualized revenue rate. Free cash flow turned negative in Q3 2025; Amazon raised $12B in new debt and cut 30,000 corporate roles generating ~$6B annual savings to offset capital strain.

Oracle (ORCL) — Q2 FY2026
Dec 2025
"We are ramping capital spending to $35 billion in 2026 to build data centers and secure GPU capacity." — Safra Catz, CFO
AI contract backlog: $455B (+359% YoY) OCI cloud revenue: $7.2B (+28%) FY2030 revenue target: $144B

Customers include OpenAI, xAI, Meta. Oracle bumped FY2026 capex to $50B in December — a massive pivot for a historically asset-light company. GPU rental gross margins run at mid-teens %, against claims of 30–40% on an "adjusted" basis. This ROIC gap is the central bear case for the stock.

🏗️
03 / 07
Where the Money Is Going

$600B+ is not a monolithic spend. Here's how it breaks down across the AI infrastructure stack:

Category Share 2026 Figure Key Context
AI Silicon (GPUs/accelerators) 40–50% $240–300B H200 → Blackwell B200 transition; Blackwell Ultra (B300) ramping Q1 2026
Networking 10–15% $60–90B InfiniBand, NVLink, Spectrum-X, 800G optical transceivers
Datacenter shells & construction 20–25% $120–150B Land acquisition, building construction, real estate
Power infrastructure 10–15% $60–90B Grid connections, transformers (128-week lead times), nuclear deals
Cooling systems 5–10% $30–60B Liquid cooling transition; Blackwell racks at 120kW vs. H100 at 40kW

Sources: CreditSights, Cerno Capital, analyst estimates

The Silicon Layer: The transition to Blackwell (B200/GB200) is underway — each Blackwell rack (NVL72) draws 120kW vs. H100's ~40kW and requires liquid cooling, making every Blackwell deployment a facility renovation. Blackwell Ultra (B300) began shipping in volume Q1 2026: 288GB HBM3e, 7.5x dense throughput vs. H100. Next on deck: Vera Rubin (H2 2026) — 50 PFLOPS NVFP4 inference, 10x lower cost per token vs. Blackwell.

The Real Bottleneck: Hyperscalers have more GPUs than they can power. Microsoft has $80B of Azure orders it can't fulfill because the electricity doesn't exist yet. Power transformer lead times have extended to 128 weeks. The IEA projects data center electricity consumption doubles to 945 TWh by 2030. The power wall is real — and it's the underpriced trade in this entire cycle.
🔍
04 / 07
Second-Order Beneficiaries

The GPU trade was 2023–2024. The infrastructure trade is 2025–2026. Here's where the money is actually flowing:

Power Utilities — Nuclear & Grid
VST · CEG · NEE

Every megawatt of AI capacity needs electrons, and hyperscalers are signing 20-year nuclear PPAs at a premium. Meta signed a massive PPA with Vistra for Comanche Peak. Microsoft reactivated Three Mile Island Unit 1 for 20 years (Constellation). Google acquired Intersect Power for $4.75B. The IEA expects U.S. electricity demand growth to be driven overwhelmingly by data centers through 2026. U.S. utility capex now exceeds $211B annually — more than every other U.S. industry.

ARM Holdings
ARM

Every custom silicon chip being built by hyperscalers runs on ARM architecture. Amazon's Graviton and Trainium, Google's TPU, Apple's chips, NVIDIA's Grace CPU — all ARM. The sovereign AI buildout (30+ nations) is also predominantly ARM-based. ARM's business model: license the architecture, collect royalties on every chip shipped. As custom silicon grows, ARM royalties compound. Consensus: $1.47B rev, $0.58 EPS.

Coherent Corp. — The Forgotten Pick
COHR

AI networking at 800G+ is impossible without advanced optical transceivers, and Coherent is one of two dominant suppliers. Transceiver shipments for 800G-and-above are projected to grow 163% in 2026 (24M → 63M units). Coherent is also the contract manufacturer for Lumentum — it captures the photonic supply chain at multiple layers. The stock has lagged NVDA and VRT, making the valuation more interesting relative to the exposure.

Vertiv Holdings
VRT

The most direct play on the Blackwell transition. When GPU rack density jumped from ~40kW (H100) to 120kW (Blackwell NVL72), air cooling became physically impossible — every new deployment requires liquid cooling. Vertiv is one of three dominant providers. Q4 2025 backlog: $15B. Organic revenue growth: 28% YoY. Evercore ISI price target: $210. YTD 2026 return: +86%. The company stopped disclosing backlog after Q4 2025 — either because the number got too large or management didn't want to anchor expectations.

Eaton Corporation
ETN

The "grid-to-chip" play. Eaton makes the switchgear, transformers, and power distribution equipment that connects utility grid to data center rack. Data center sales grew 64% in 2025. Deutsche Bank named it top pick for 2026 with a $400 price target. ETN underperformed in 2025 (consensus ran too high), which means the setup for 2026 is more favorable. Key metric: order backlog and lead times — Eaton disclosed record backlogs at the UBS industrials conference, December 2025.

⚠️
05 / 07
The Bear Case

This is not a balanced "on the other hand" section. It's the actual risk inventory that matters.

Risk 1
The Monetization Gap Is Historic
$600B+ in annual capex. ~$25B in direct AI revenue (OpenAI ended 2025 at ~$20B ARR; Anthropic at $9B). That's roughly 4 cents of current revenue for every dollar being spent. The bull case requires a decade of compounding to justify the present-value math. The bear case is that the gap never closes at the required rate — and $600B of annual depreciation hits the income statement before the revenue shows up.
Risk 2
ROIC Is Already Deteriorating
4 of the 5 hyperscalers saw cash flow decline in 2025. BofA calculates aggregate hyperscaler capex now consumes 94% of operating cash flows after dividends and buybacks. LPL Research and AdvisorAnalyst both flagged "signs of eroding returns on invested capital." ROIC has historically been volatile and front-runs the investment returns — but the window between spending and earning is longer for AI infrastructure than it was for prior tech cycles.
Risk 3
Capital Intensity at Historically Unthinkable Levels
2025 capex as % of revenue crossed 22% for the Big Five. Historical average: 11–16%. Futuriom argues that a mean reversion alone implies a potential 27% capex decline in 2026 once current infrastructure is "digested." The hyperscalers themselves acknowledge this possibility — which is why guidance language has shifted from "we'll spend as much as needed" toward more careful discussion of "ROI signals."
Risk 4
The Duration Mismatch
GPU useful life: 3 years. Equity valuations pricing these investments: 20+ years. AMOVA Asset Management calls this a "duration mismatch — short-lived assets funded by long-duration equity valuations." When $30,000 GPUs depreciate over 3 years, the income statement impact is concentrated and brutal.
Risk 5
The Power Wall Could Strand Assets
Microsoft has GPUs sitting idle because it can't get grid connections. Power transformer lead times of 128 weeks mean you can order GPUs today and not run them for 2.5 years. If the AI revenue ramp slows during that window, you have $30,000/unit depreciating assets generating zero return. This is not theoretical — it is the current state for at least one major hyperscaler.
Risk 6
Debt Structure Risk
The Big Five raised $121B in bonds in 2025 alone. JP Morgan projects $1.5T in tech debt issuance over the coming years. For the first time, hyperscalers hold more debt than cash. Moody's flagged this in a March 2026 report: "$700B capex in 2026 with investors fearing overbuild and weak returns." Oracle's 5-year CDS has more than tripled since its September bond deal.

// Signals to watch for 2026 deceleration

Azure growth missing consensus by 1+ points for two consecutive quarters
Any hyperscaler explicitly guiding 2027 capex below 2026
NVIDIA data center bookings deceleration (key leading indicator)
Power connection timelines extending further
Enterprise AI adoption data showing slower-than-expected penetration
🔭
06 / 07
What FinTwit Is Missing

Three structural dynamics underrepresented in mainstream AI capex coverage.

1. Custom Silicon Is Eating NVDA Share Faster Than the Headline Numbers Suggest

Everyone knows Google has TPUs and Amazon has Trainium. But the consensus treats custom silicon as a minor footnote. The actual number: custom ASICs now handle 15–25% of internal hyperscaler workloads (TrendForce, early 2026). That's approaching a quarter of all compute. Google's TPU v5 matches Blackwell on specific inference benchmarks. Amazon's Trainium 2 crossed $10B ARR.

The bear implication: NVIDIA's TAM growth from hyperscaler share gains is already decelerating, even as absolute dollar amounts grow. The bull case for NVIDIA increasingly depends on (a) sovereign AI, (b) enterprise deployments outside the Big Five, and (c) inference stack advantage from CUDA/NIM lock-in. NVIDIA's response: Groq acquisition ($20B), Vera Rubin architecture optimized for inference economics, and $2B equity investment in CoreWeave.

2. Sovereign AI Is a Fundamentally Different Market That Consensus Ignores

30+ nations are now building domestic AI infrastructure — Saudi Arabia, UAE, India, Japan, France, Canada. The key word is sovereign: these projects aren't price-sensitive the way enterprise contracts are. They're geopolitically motivated. A government building a national AI compute layer doesn't negotiate the price of Blackwell GPUs the way AWS does.

NVIDIA has visibility into $500B of cumulative Blackwell/Rubin revenue through end of 2026 — a significant portion sovereign. Telcos in 5 continents are partnering with NVIDIA to build sovereign AI infrastructure. Vertiv called out sovereign AI as a "significant and unexpected" growth driver in its EMEA and APAC order books in late 2025. The market models NVIDIA against hyperscaler capex. The sovereign channel is an entirely separate demand source that appears nowhere in most sell-side models.

3. The Training-to-Inference Mix Shift Changes Who Wins

2023–2024 was the training era. 2025–2026 is the inference era. OpenAI processes billions of requests daily. Meta serves AI recommendations to 3B+ users. That traffic is inference, not training.

Why this matters for capital allocation: Training is bursty, GPU-hungry, and benefits from NVIDIA's Blackwell. Inference is continuous, latency-sensitive, and benefits from networking (Arista, Coherent), memory bandwidth, and cost efficiency per token. The market still assigns most of the AI capex narrative to GPU suppliers. But if inference is now 75% of total AI compute (NVIDIA's own projection for 2030), the compounders are Arista (networking), Coherent (optical), Vertiv (thermal), and NVIDIA Dynamo. This transition is why NVIDIA launched Vera Rubin — its first architecture explicitly optimized for inference economics over training throughput.

👁️
07 / 07
The Watchlist

Five tickers most directly exposed to the AI infrastructure capex cycle — across silicon, power, and thermal.

Ticker Our Take Consensus Next Print
NVDA The toll booth on AI inference. Rubin launch H2 2026 is the next catalyst. Custom silicon risk is real but embedded; sovereign AI is the surprise upside. Watch: inference token economics vs. Groq/Rubin thesis. ~$180B FY2026 rev · 68–73% gross margin Late May 2026
VRT Most direct Blackwell-transition play. Every rack going from 40kW to 120kW passes through Vertiv's order book. $15B backlog (Q4 2025). Risk: stopped disclosing backlog; concern is whether growth is being pulled forward. ~$9.5–10B FY2026 rev · ~25% EBITDA margin Late Apr/May 2026
AMD → MI400 series gaining ground in sovereign AI and inference. AMD's "fastest-ramping product in company history." CUDA moat is real but inference workloads are more ASIC-agnostic than training. Next print is the setup for H2 custom silicon narrative. $7.25B Q1 rev · $1.09 EPS · Data Center ~$3.3B May 5, AMC
ETN Grid-to-chip. Underperformed in 2025, setting up for 2026 with more realistic expectations. Deutsche Bank top pick. Watch Q1 order growth commentary vs. backlog expansion. Tariff drag easing. ~$27–28B FY2026 rev · ~21% operating margin Late April 2026
ORCL $455B AI contract backlog, cloud growing 28%. Bear case: margin compression on GPU rentals (mid-teens vs. claimed 30–40% adjusted). $50B capex in 2026 is a big bet for a historically asset-light company. Bull case: OCI becomes preferred sovereign cloud. ~$75B FY2026 rev · $7.70 EPS June 2026 (FY Q4)
Data current as of April 28, 2026. Sources: Company Q4 2025 / Q1–Q2 FY2026 earnings calls (Microsoft, Alphabet, Meta, Amazon, Oracle); CreditSights hyperscaler capex estimates; Moody's "US hyperscaler capex to top $700bn in 2026" (March 2026); TrendForce AI server shipments data; Vertiv Q4 2025 earnings; NVIDIA GTC 2026 and Q3 FY2026 earnings; IEEE ComSoc Sovereign AI analysis; Futuriom, LPL Research, AMOVA Asset Management, BofA, JP Morgan research. Forward-looking statements are Vektor analysis, not investment advice.
// Companies in this brief
⚡ Vektor Intelligence

Get briefs like this
for any company — $29/mo

Companies, markets, and trends. Drop in any name and get a structured intelligence brief in 40 seconds — earnings previews, sector trackers, competitive deep dives.

Start for $29/month → — or get your first brief free —
✓ You're in — we'll send your free brief shortly.

No credit card. Cancel anytime.