AI Revolution model basket

Hyperscaler Capex Sponsors

A book built around the balance sheets funding AI buildout and the second-order suppliers drafting behind them.

What is the thesis for Hyperscaler Capex Sponsors?

Hyperscaler capital spending has roughly tripled off its 2019 base and is still accelerating. The question is not whether the spend is real but whether it is already in the price; we think the sponsors and a narrow slice of suppliers still offer a defensible expected value.

This is a curated QuantLink model basket. It is not a filed portfolio, not a fund, and not investment advice.

Published Apr 14, 2026. Updated Apr 14, 2026. Source: QuantLink curated model basket and FastAPI ideas endpoint.

Holdings
9
Benchmark
SPY
Status
New
1Y model return
-7.4%

Performance as of Nov 22, 2026.

Thesis narrative

The question

The right question for this book is not whether hyperscaler capex is large. It is whether the market's imputed probability of sustained high returns on that capex is reasonable, and whether the companies writing the checks are priced as though the returns will hold. Aggregate capex at the top four US hyperscalers has moved from roughly $90B in 2019 to an expected $340B+ in 2026. The implied reference class is something closer to telecom 1999 than to industrial compounders of the 1990s, and that comparison is the consensus worry. Our view is narrower: the sponsors with pre-existing cashflow engines can absorb a wide range of return outcomes without impairing intrinsic value, while a selected group of suppliers offers asymmetric payoff if utilization holds, and option value if it does not.

Base rates

The base rate for multi-year capex super-cycles is not encouraging. Studies of capital-intensive buildouts (rail, fiber, shale, wind) show median incremental ROIC compressing roughly 300-500bps from peak to trough, with a five-to-seven year lag between spend and return normalization. The base rate for large-cap platform businesses that already earn mid-twenties operating margins is materially better: companies entering a capex cycle from a position of high free cash flow conversion have historically preserved 70-80% of pre-cycle ROIC, because the incremental asset base is funded from retained earnings rather than debt or equity issuance. That distinction — who funds the build — is the crux.

Why the consensus view is wrong (or incomplete)

The prevailing narrative treats hyperscaler capex as homogeneous. It is not. Alphabet and Amazon are funding buildout with roughly 100% of spend covered by operating cash flow; several mid-tier compute providers are funding via a mix of preferred equity, vendor financing, and receivable-backed debt. The market is pricing both cohorts off a shared multiple framework, which creates dispersion we can exploit. Second, the consensus view assumes token-economics will mean-revert quickly. The evidence from 2024-2025 is that inference unit costs are falling faster than demand elasticity is absorbing, which lengthens the window of positive unit economics for capacity owners. Third, the consensus treats suppliers as commodities. A handful of ODMs and rack integrators have locked in multi-year, cost-plus frameworks that look less like cyclical hardware and more like contract manufacturing with escalators.

Position construction

The book is organized into three sub-books.

Sponsor balance sheets (40%) — GOOGL and AMZN at 20% each. These are the companies writing the largest checks with the least dilution risk. We own them for the combination of (a) cashflow optionality to slow the build if returns disappoint and (b) the embedded cloud franchises whose unit economics improve as capacity comes online. GOOG Class C is held at zero weight and is listed for ticker consistency with composite feeds.

Application-layer beneficiaries (20%) — APP at 20%. The thesis here is that the companies monetizing inference at the ad-tech and consumer-discovery layer capture a disproportionate share of the economic surplus, because their cost-of-revenue falls as inference prices fall while pricing power is anchored to advertiser willingness-to-pay. APP is the cleanest expression we found.

Suppliers and neoclouds (40%) — CRWV 16.2%, CLS 7.3%, HPE 6.9%, NBIS 5.8%, IREN 3.8%. This is the asymmetric sleeve. CRWV and NBIS are pure-play compute providers with contracted backlogs that extend past the near-term capex peak; CLS and HPE are rack and system integrators with book-to-bill above 1.2x; IREN is a lower-conviction option on power-adjacent compute. Weights decline with visibility of cashflow; the tail is small on purpose.

Asymmetric payoff

Our rough expected-value frame assumes three states. In a base case (50% probability), hyperscaler capex plateaus in 2027 and the sponsor cohort delivers 10-12% annualized; the supplier sleeve delivers 15-20%. In an upside case (25%), utilization holds through 2028 and the supplier sleeve delivers 40%+. In a downside case (25%), capex is cut 30% and the supplier sleeve draws down 35-45% while the sponsor cohort protects to a 10-15% drawdown. Probability-weighted, the book offers a mid-teens expected return with roughly 2:1 upside-to-downside asymmetry, driven more by the sponsor cashflow floor than by the supplier beta.

Three things that would change our mind

First, a sustained inflection in cloud gross margins lower rather than flat; that would indicate the incremental capex is not earning a competitive return. Second, a shift in hyperscaler financing mix toward off-balance-sheet structures or vendor-financed supplier receivables; that would signal return-on-capital pressure is being masked. Third, a credible entrant at the model layer that commoditizes inference pricing faster than capacity providers can amortize their builds.

What we are explicitly NOT betting on

We are not betting on a particular model architecture winning. We are not betting on semiconductor market share shifts at the accelerator layer. We are not betting on sovereign AI spend filling any capex gap left by the hyperscalers. And we are not betting that the suppliers can hold current margins through the next down-cycle; we are betting the backlog carries them through the window we care about.

Model basket holdings

Model basket: curated equal or target weighting, not a filed portfolio. Weights are the target basket weights returned by the live ideas endpoint.

NameSymbolModel weight
Alphabet Inc.GOOGL19.99%
Amazon.com, Inc.AMZN20.00%
AppLovin CorporationAPP20.00%
CoreWeave, Inc. Class A Common StockCRWV16.23%
Celestica Inc.CLS7.27%
Hewlett Packard Enterprise CompanyHPE6.93%
Nebius Group N.V.NBIS5.81%
IREN LimitedIREN3.77%
Alphabet Class CGOOG0.00%

Backtested performance vs SPY

Performance is backtested from the returned tearsheet series. It reflects the model basket methodology and benchmark series, not live fund returns or a filed portfolio track record. Performance as of Nov 22, 2026.

Total Return

-7.4%

↓ SPY +3.2%

Ann. Return

-7.7%

↓ SPY +3.3%

Ann. Vol

44.2%

↑ SPY 14.5%

Sharpe

-0.18

↓ SPY 0.23

Max Drawdown

-37.4%

↓ SPY -11.5%

Alpha vs SPY

-8.0%

↓ hit rate 47.5%

Performance as of Nov 22, 2026. Risk-free rate for Sharpe and Sortino: 0% (no risk-free rate applied).

Rolling Performance vs Benchmark

Portfolio Holdings

Holding
Weight
Country
Exchange
Sector
Industry
Mkt Cap
Price
1Y
1Y Trend
AMZN
AMZNAmazon.com, Inc.
20.0%
—
—
—
—
—
—
—
APP
APPAppLovin Corporation
20.0%
—
—
—
—
—
—
—
GOOGL
GOOGLAlphabet Inc.
20.0%
—
—
—
—
—
—
—
CRWV
CRWVCoreWeave, Inc. Class A Common Stock
16.2%
—
—
—
—
—
—
—
CLS
CLSCelestica Inc.
7.3%
—
—
—
—
—
—
—
HPE
HPEHewlett Packard Enterprise Company
6.9%
—
—
—
—
—
—
—
NBIS
NBISNebius Group N.V.
5.8%
—
—
—
—
—
—
—
IREN
IRENIREN Limited
3.8%
—
—
—
—
—
—
—
GOOG
GOOGAlphabet Class C
0.0%
—
—
—
—
—
—
—

SSR performance series fallback

The table below is the server-rendered reference series behind the interactive chart. Values show the wealth index level from a 1.00 starting value, not a second 1Y return figure. Series as of Nov 22, 2026.

DateModel basket wealth indexSPY
Nov 25, 20251.0000x1.0000x
Nov 26, 20251.0213x1.0069x
Nov 28, 20251.0304x1.0124x
Dec 1, 20251.0434x1.0078x
Dec 2, 20251.0404x1.0096x
Dec 3, 20251.0578x1.0131x
Dec 4, 20251.0846x1.0139x
Dec 5, 20251.0928x1.0158x
Dec 8, 20251.0889x1.0128x
Dec 9, 20251.1134x1.0119x
Dec 10, 20251.1064x1.0186x
Dec 11, 20251.1013x1.0210x
Dec 12, 20251.0426x1.0100x
Dec 15, 20251.0170x1.0085x
Dec 16, 20251.0085x1.0057x
Dec 17, 20250.9721x0.9946x
Dec 18, 20251.0020x1.0021x
Dec 19, 20251.0703x1.0083x
Dec 22, 20251.0901x1.0145x
Dec 23, 20251.0824x1.0192x
Dec 24, 20251.0807x1.0228x
Dec 26, 20251.0658x1.0227x
Dec 29, 20251.0548x1.0190x
Dec 30, 20251.0481x1.0178x
Dec 31, 20251.0321x1.0102x
Jan 2, 20261.0424x1.0121x
Jan 5, 20261.0536x1.0188x
Jan 6, 20261.0615x1.0249x
Jan 7, 20261.0624x1.0216x
Jan 8, 20261.0600x1.0215x
Jan 9, 20261.0854x1.0282x
Jan 12, 20261.1249x1.0298x
Jan 13, 20261.1279x1.0278x
Jan 14, 20261.1024x1.0227x
Jan 15, 20261.1092x1.0255x
Jan 16, 20261.1121x1.0247x
Jan 20, 20261.0744x1.0038x
Jan 21, 20261.0655x1.0154x
Jan 22, 20261.0556x1.0207x
Jan 23, 20261.0655x1.0211x
Jan 26, 20261.0801x1.0262x
Jan 27, 20261.1262x1.0303x
Jan 28, 20261.1275x1.0302x
Jan 30, 20261.0424x1.0251x
Feb 2, 20261.0501x1.0302x
Feb 3, 20261.0416x1.0215x
Feb 4, 20260.9722x1.0165x
Feb 5, 20260.9348x1.0039x
Feb 6, 20260.9828x1.0231x
Feb 9, 20261.0321x1.0280x
Feb 10, 20261.0212x1.0253x
Feb 11, 20261.0033x1.0251x
Feb 12, 20260.9476x1.0093x
Feb 13, 20260.9675x1.0100x
Feb 17, 20260.9503x1.0116x
Feb 18, 20260.9795x1.0167x
Feb 19, 20260.9895x1.0140x
Feb 20, 20260.9849x1.0213x
Feb 23, 20260.9635x1.0109x
Feb 24, 20260.9917x1.0183x
Feb 25, 20261.0087x1.0269x
Feb 26, 20261.0096x1.0212x
Feb 27, 20260.9708x1.0163x
Mar 2, 20260.9619x1.0168x
Mar 3, 20260.9455x1.0079x
Mar 4, 20260.9983x1.0150x
Mar 5, 20260.9939x1.0093x
Mar 6, 20260.9681x0.9961x
Mar 9, 20260.9954x1.0048x
Mar 10, 20260.9807x1.0032x
Mar 11, 20261.0017x1.0019x
Mar 12, 20260.9836x0.9867x
Mar 13, 20260.9900x0.9811x
Mar 16, 20261.0176x0.9911x
Mar 17, 20261.0146x0.9937x
Mar 18, 20260.9993x0.9799x
Mar 19, 20260.9982x0.9775x
Mar 20, 20260.9863x0.9608x
Mar 23, 20261.0058x0.9709x
Mar 24, 20260.9954x0.9676x
Mar 25, 20261.0152x0.9730x
Mar 26, 20260.9523x0.9557x
Mar 27, 20260.9185x0.9394x
Mar 30, 20260.8867x0.9362x
Mar 31, 20260.9506x0.9634x
Apr 1, 20260.9573x0.9707x
Apr 2, 20260.9700x0.9716x
Apr 6, 20260.9879x0.9762x
Apr 7, 20261.0044x0.9766x
Apr 8, 20261.0284x1.0015x
Apr 9, 20261.0472x1.0072x
Apr 10, 20261.0876x1.0066x
Apr 13, 20261.1316x1.0164x
Apr 14, 20261.1793x1.0288x
Apr 15, 20261.2044x1.0369x
Apr 16, 20261.2098x1.0395x
Apr 17, 20261.2179x1.0520x
Apr 20, 20261.2276x1.0499x
Apr 21, 20261.2116x1.0431x
Apr 22, 20261.2420x1.0536x
Apr 23, 20261.2177x1.0495x
Apr 24, 20261.2140x1.0577x
Apr 27, 20261.2264x1.0595x
Apr 28, 20261.1838x1.0543x
Apr 29, 20261.2057x1.0542x
Apr 30, 20261.2391x1.0646x
May 1, 20261.2737x1.0676x
May 4, 20261.3105x1.0637x
May 5, 20261.3299x1.0722x
May 6, 20261.3640x1.0871x
May 7, 20261.3464x1.0838x
May 8, 20261.3118x1.0927x
May 11, 20261.3058x1.0952x
May 12, 20261.2905x1.0936x
May 13, 20261.3080x1.0997x
May 14, 20261.3440x1.1084x
May 15, 20261.3198x1.0950x
May 18, 20261.2952x1.0943x
May 19, 20261.2621x1.0870x
May 20, 20261.2822x1.0981x
May 21, 20261.3181x1.1003x
May 22, 20261.3166x1.1046x
May 26, 20261.3405x1.1120x
May 27, 20261.3731x1.1118x
May 28, 20261.4019x1.1179x
May 29, 20261.4269x1.1207x
Jun 1, 20261.4804x1.1237x
Jun 2, 20261.4802x1.1253x
Jun 3, 20261.4277x1.1174x
Jun 4, 20261.4201x1.1216x
Jun 5, 20261.3535x1.0926x
Jun 8, 20261.3627x1.0951x
Jun 9, 20261.3225x1.0919x
Jun 10, 20261.2769x1.0747x
Jun 11, 20261.2915x1.0929x
Jun 12, 20261.3205x1.0989x
Jun 15, 20261.3759x1.1182x
Jun 16, 20261.3910x1.1116x
Jun 17, 20261.3546x1.0977x
Jun 18, 20261.3650x1.1062x
Jun 22, 20261.3254x1.1028x
Jun 23, 20261.3024x1.0868x
Jun 24, 20261.2855x1.0862x
Jun 25, 20261.2542x1.0878x
Jun 26, 20261.2521x1.0800x
Jun 29, 20261.2893x1.0977x
Jun 30, 20261.3188x1.1063x
Jul 1, 20261.3020x1.1048x
Jul 2, 20261.2532x1.1033x
Jul 6, 20261.2922x1.1130x
Jul 7, 20261.2683x1.1077x
Jul 8, 20261.2940x1.1043x
Jul 9, 20261.2997x1.1136x
Jul 10, 20261.2888x1.1184x
Jul 13, 20261.2300x1.1098x
Jul 14, 20261.2285x1.1138x
Jul 15, 20261.2340x1.1182x
Jul 16, 20261.1707x1.1121x
Jul 17, 20261.1599x1.1011x
Jul 20, 20261.1756x1.0994x
Jul 21, 20261.2163x1.1085x
Jul 22, 20261.2103x1.1072x
Jul 23, 20261.1694x1.0936x
Jul 24, 20261.1223x1.0947x
Jul 27, 20261.1390x1.0949x
Jul 28, 20261.1322x1.0975x
Jul 29, 20261.0811x1.0806x
Jul 30, 20261.1675x1.0988x
Jul 31, 20261.2041x1.1067x
Aug 3, 20261.2899x1.1224x
Aug 4, 20261.3281x1.1427x
Aug 5, 20261.3015x1.1404x
Aug 6, 20261.2107x1.1386x
Aug 7, 20261.2361x1.1455x
Aug 10, 20261.2277x1.1452x
Aug 11, 20261.2064x1.1415x
Aug 12, 20261.2716x1.1444x
Aug 13, 20261.2801x1.1524x
Aug 14, 20261.2785x1.1501x
Aug 17, 20261.2728x1.1447x
Aug 18, 20261.2225x1.1369x
Aug 19, 20261.2142x1.1393x
Aug 20, 20261.2005x1.1297x
Aug 21, 20261.1938x1.1344x
Aug 24, 20261.1831x1.1310x
Aug 25, 20261.2064x1.1346x
Aug 26, 20261.1978x1.1349x
Aug 27, 20261.1982x1.1423x
Aug 28, 20261.1931x1.1397x
Aug 31, 20261.1806x1.1363x
Sep 1, 20261.1594x1.1285x
Sep 2, 20261.1663x1.1335x
Sep 3, 20261.1964x1.1454x
Sep 4, 20261.2150x1.1410x
Sep 8, 20261.2494x1.1347x
Sep 9, 20261.2266x1.1294x
Sep 10, 20261.2100x1.1227x
Sep 11, 20261.2413x1.1322x
Sep 14, 20261.2185x1.1272x
Sep 15, 20261.2012x1.1220x
Sep 16, 20261.2047x1.1171x
Sep 17, 20261.2130x1.1297x
Sep 18, 20261.2158x1.1284x
Sep 21, 20261.2595x1.1459x
Sep 22, 20261.2599x1.1457x
Sep 23, 20261.2339x1.1375x
Sep 24, 20261.2513x1.1365x
Sep 25, 20261.2390x1.1427x
Sep 28, 20261.2201x1.1342x
Sep 29, 20261.2214x1.1321x
Sep 30, 20261.2178x1.1298x
Oct 1, 20261.2109x1.1318x
Oct 2, 20261.2229x1.1402x
Oct 5, 20261.2258x1.1479x
Oct 6, 20261.2487x1.1542x
Oct 7, 20261.2407x1.1514x
Oct 8, 20261.2030x1.1465x
Oct 9, 20261.2158x1.1534x
Oct 12, 20261.1928x1.1527x
Oct 13, 20261.1877x1.1526x
Oct 14, 20261.1134x1.1401x
Oct 15, 20261.1333x1.1434x
Oct 16, 20261.0704x1.1307x
Oct 19, 20261.0880x1.1357x
Oct 20, 20261.0902x1.1393x
Oct 21, 20261.0880x1.1453x
Oct 22, 20261.0297x1.1318x
Oct 23, 20261.0156x1.1258x
Oct 26, 20261.0153x1.1338x
Oct 27, 20261.0568x1.1402x
Oct 28, 20261.0836x1.1441x
Nov 5, 20261.0028x1.0859x
Nov 9, 20261.0910x1.1151x
Nov 10, 20261.0795x1.1110x
Nov 11, 20261.1508x1.1256x
Nov 12, 20261.1266x1.1216x
Nov 13, 20261.1046x1.1176x
Nov 16, 20261.0683x1.0962x
Nov 17, 20261.0539x1.0851x
Nov 18, 20261.0532x1.0817x
Nov 19, 20261.0115x1.0605x
Nov 20, 20260.9265x1.0219x

Themes and category

AI RevolutionAI InfrastructureQuality

Methodology and caveats

QuantLink fetches this idea from the live FastAPI ideas endpoints and renders the returned title, thesis, holdings, themes, benchmark, and tearsheet fields directly. Missing fields are left unavailable rather than fabricated.

Holdings are a curated model basket. They are not 13F filings, not insider filings, not adviser holdings, and not a claim that any person or fund owns the basket.

Backtested performance depends on the returned basket weights, benchmark, rebalancing assumptions, available price history, and calculation choices in the tearsheet endpoint. Backtests can differ materially from live results and do not include every cost, tax, capacity, liquidity, or execution constraint an investor may face.

Equal-weight and target-weight baskets can drift between rebalance points. Rebalancing can increase turnover, and concentrated thematic baskets can have higher drawdowns than a broad market benchmark.

Frequently asked questions

Research the stocks behind this idea

Use QuantLink's screener and company pages to inspect fundamentals, valuation, and market data after reviewing the public thesis.

Related links

QuantLink is a research tool, not investment advice. This page shows a curated model basket and backtested performance, not a filed portfolio, fund return, or recommendation to buy or sell securities.