What Is the Data Center Index? An Equal-Weighted AI Supply-Chain Basket, Explained

The short version

DataCenterIndex publishes two indices. The DCI-AI (AI Infrastructure Leaders) is the complete picture: hyperscalers, semiconductors, power, cooling, operators, neoclouds, construction, utilities, networking, server OEMs. Forty-four names that together represent the entire public-market stack of the AI buildout. The DCI-SC (Supply Chain) is the same basket with Amazon, Google, Meta, Microsoft, Oracle, AMD, Broadcom, and Nvidia taken out. What remains is the companies that sell into the buildout rather than the companies doing the buying.

Both are equal-weighted, both compound from a base of 100, both update daily. The reason there are two is that these are two different bets. DCI-AI is a bet that the AI buildout continues. DCI-SC is a bet that the buildout continues and that the margin is going to flow downstream, toward the people selling transformers and switchgear and fuel cells and cooling equipment, rather than staying with the hyperscalers capturing the AI workloads themselves.

As of April 17, 2026, DCI-SC (Supply Chain) sits at 206.86, up from a base of 100 on April 7, 2025. DCI-AI (AI Leaders), which adds the hyperscalers and pure-semiconductor names, sits at 198.53. That is 106.9% and 98.5% respectively over a year in which the S\&P 500 returned roughly 20%. Both indices corrected meaningfully off summer highs (peak drawdown around -36% on each), then rebuilt. You can watch both live on the dashboard.

This page explains how the indices are built, what is in each, and what the two-index structure is meant to reveal.

The argument for two indices

The simplest way to play data centers is to buy a fund that holds Nvidia, Broadcom, Amazon, and Microsoft. Whatever happens at the edges, those four companies will capture most of the economic value, and the market-cap-weighted ETFs make sure they dominate the basket. You know the trade. You have probably already made the trade.

The interesting question is what happens at the edges. In any infrastructure cycle, railroads, electricity, fiber, there is a period where the attention and the returns are on the headline companies. Then there is a longer period where the returns migrate to the suppliers, because the headline companies keep outspending each other and the suppliers keep having pricing power they did not have before. You can watch this play out over the past eighteen months in Vertiv, whose margins have been quietly expanding every quarter while their customers post capex guidance that would have been unthinkable in 2023. More recently you can watch it play out in fuel cells, where Bloom Energy has become one of the most-cited AI power stories on earnings calls despite being a company most investors had written off in 2022. The Q1 2026 capex scorecard has more.

DCI-AI captures the whole cycle: buyers and sellers, compute and power, hyperscalers and gensets. If you think the AI buildout is real and accelerating, DCI-AI is a clean way to express that.

DCI-SC drops the buyers. What you are left with is a sharper instrument: if capex is really growing 50 to 80 percent year over year (it is), and if the power and cooling supply chains really are running at all-time capacity (they are, see the power-bottleneck piece), DCI-SC should outperform DCI-AI. And it has. Not by a lot. Just by enough to matter.

We track both because the spread between them tells you something that neither index tells you alone.

How the math works

Both indices are calculated the same way. Each constituent contributes equally to daily return. On any given day the index value equals yesterday's value times one plus the simple average of today's constituent returns. Compounded forward from a base of 100.

Nothing fancy. No survivorship cleaning, no float adjustment, no sector caps, no volatility targeting. We are not running a fund. We are publishing a number.

Equal weighting is the only interesting methodology choice, and it is a deliberate corrective to the weirdness of market-cap weighting in a concentrated market. A market-cap-weighted version of the DCI-AI basket would be well north of half the basket in Nvidia, Broadcom, and the hyperscalers. You would be buying a mega-cap tech ETF with some power-equipment decoration. Equal weighting forces the small names to actually vote. If Primoris Services has a 15% week, Primoris Services has a 15% week. The index does not quietly round them down to irrelevance. Which is important when a $4B fuel-cell company like Bloom Energy can triple in a quarter on data center orders; a market-cap index would barely register it, and equal weight puts it directly into the return stream.

This has a cost. The index drifts daily (each constituent's weight bobs around as prices move), and a proper tracking product would need to rebalance. We do not trade this. We publish it.

What's in each index

DCI-AI (44 constituents across 11 sectors):

| Sector | Count | Names |

| -------------------- | ----- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

| Power equipment | 11 | ABBNY, CAT, CMI, EMR, ETN, GEV, GNRC, HUBB, NVT, POWL, SBGSF |

| Energy / utilities | 7 | BE, CEG, D, NEE, NRG, TLN, VST |

| Construction | 5 | EME, FLR, J, PRIM, PWR |

| Hyperscalers | 5 | AMZN, GOOGL, META, MSFT, ORCL |

| Cooling | 4 | JCI, MOD, TT, VRT |

| DC operators (REITs) | 3 | AMT, DLR, EQIX |

| GPU cloud / neocloud | 3 | CRWV, IREN, NBIS |

| Semiconductor | 3 | AMD, AVGO, NVDA |

| Networking | 2 | ANET, CIEN |

| IT hardware | 1 | SMCI |

DCI-SC (same list, minus the hyperscalers and semiconductors). Which leaves you with power, cooling, construction, operators, neoclouds, utilities, networking, and the server OEM. Pure supply chain. No buyers, no chips. Thirty-six names.

The thing to notice is that even in DCI-AI, the supply chain names outnumber the hyperscalers seven to one. This is not because we tilted the basket. It is because the AI buildout has a long tail of critical vendors and a short head of customers, and when you try to capture the whole buildout honestly, you get a lot of boring mid-cap electrical companies with order books that look like science fiction.

Cummins, for example, is in this index because every new hyperscale data center needs backup gensets, and the number of hyperscale data centers under construction has doubled. Cummins did not know this was coming in 2022. Their stock is telling you they know now. Bloom Energy is in for a different reason: on-site fuel cells went from a decade-long curiosity to a bankable alternative to waiting 128 weeks for a grid transformer, once Oracle's Abilene site made the math work. The Oracle fuel-cell piece walks through how that happened.

What's excluded from both

Three categories are deliberately out of both indices.

Chip fabs (TSM, UMC). The AI compute supply chain does start at the fab. But the fab layer has its own cycle, driven by iPhone demand and Chinese geopolitics as much as by hyperscaler demand. If we were running a "semiconductor supply chain" index we would include them. We are running a data center infrastructure index, and the fabs sit upstream of that.

Fab equipment (AMAT, LRCX, ASML, KLAC). Same logic. The WFE complex is its own beast. Style drift.

QTS. Taken private by Blackstone in 2021. Keeping it out of the rules file would be an embarrassing oversight; we keep it listed as excluded so anyone auditing can confirm we did not forget.

The interesting exclusion is the philosophical one. We include three public GPU-cloud names (CoreWeave, IREN, Nebius) even though one could argue they are really hyperscalers-in-waiting, and therefore customers of the supply chain rather than suppliers. We decided they belonged in because their public-market story is specifically a bet on the supply chain: debt-financed GPU clusters, hyperscale real estate, capex that rhymes with the incumbents. The Nebius neocloud piece and the CoreWeave customer-mix piece walk through the thesis. If they succeed, the supply chain wins. If they fail, the supply chain still wins, because someone else gets the orders. Either way, the index captures the trade.

What the spread tells you

Here is the reason there are two indices, and not one. The Supply Chain index (DCI-SC) drops the buyers. The AI Leaders index (DCI-AI) keeps them. Everything else is the same: same math, same start date, same equal weighting. The only variable is whether you count the hyperscalers and semiconductor names as part of the trade.

Since launch on April 7, 2025, DCI-SC is up 106.9%. DCI-AI is up 98.5%. The supply chain index has beaten the AI Leaders index by roughly 8.3 percentage points over the same window, with essentially identical drawdown (both bottomed around -36% during last summer's correction, then rebuilt). That spread is not a small number. It is not a huge one either. It is, we think, about the right size to confirm something meaningful without being suspicious.

Here is the comparison across the periods most readers care about:

| Period | DCI-SC (Supply Chain) | DCI-AI (AI Leaders) | Spread |

| ------------ | --------------------- | ------------------- | -------- |

| 1 month | +18.98% | +15.04% | +3.94 pp |

| 3 months | +33.60% | +28.70% | +4.90 pp |

| 6 months | +55.66% | +52.03% | +3.63 pp |

| YTD | +34.27% | +29.89% | +4.38 pp |

| Since launch | +106.86% | +98.53% | +8.33 pp |

The direction is consistent. Over every window we track, the Supply Chain index has done a little better than the AI Leaders index. The gap is widening slightly, not narrowing, which is mildly noteworthy. You do not need to believe a strong version of the picks-and-shovels thesis to find this interesting. You just need to notice that the version of this trade without the headline names has, so far, produced more return per unit of risk.

Why this might be true: hyperscaler capex has a faster reaction function than supplier revenue. When a hyperscaler announces a $200 billion capex year, the stock often sells off on margin concerns. The transformer company that just took the order does not sell off. They report the backlog next quarter and the market reprices them. The supply chain names ride the wave with less immediate drawdown and a longer runway of visibility.

The spread also tells you something about future tape. If DCI-AI starts outperforming DCI-SC, the market is rewarding the narrative layer again. That has not happened yet in any sustained way. It would be a real signal. If DCI-SC keeps outperforming, the market is agreeing with the boring version of the thesis: the margin is going downstream, and the people with the long backlogs are going to keep the pricing power for a while. The Oracle fuel-cell piece is one version of this same trade playing out at a single-project level.

What the indices do not do

They capture non-US exposure partially, not fully. ABB (Swiss) and Schneider Electric (French) are in there now, which is an improvement over a year ago when the basket was essentially US-only. Legrand, Mitsubishi Electric (Japanese), and various Korean heavy-electrical names are still missing. Korean and Japanese power-equipment exposure is a real gap given how much equipment actually ships from Asia.

They do not capture private market exposure. The fastest-growing data center operators are private (CyrusOne, Stream, Aligned). The basket catches only the public sliver.

They do not capture credit. A growing share of the neocloud buildout is financed with GPU-collateralized debt. That is an equity index problem only in the sense that an equity index will tell you when the credit story breaks, probably violently and after a delay.

If any of those gaps close, the indices will update. Methodology is versioned and published.

How to use this site

DataCenterIndex is structured around these two indices. The dashboard shows live constituent performance for both. The insights explain what moved and why, with the kind of detail that a ticker page cannot. Each constituent has its own page showing capex, filings, price history, and the related insight coverage — start with Vertiv, Eaton, Constellation, or Bloom Energy if you want a feel for the format. The pipeline map shows every announced US data center project and the supply chain tracker shows commodity prices and equipment lead times.

The indices are the thesis. Everything else on the site is support.

Frequently asked questions

Why two indices?

Because the spread between them is a signal. DCI-AI tells you whether the AI buildout is happening. DCI-SC tells you whether the suppliers are capturing the margin. Running both lets you see which.

Which one should I watch?

If you want one number that represents the AI buildout as a whole, DCI-AI. If you want to watch the picks-and-shovels trade specifically, DCI-SC. Most readers end up watching both and paying attention to the spread.

How is the index calculated?

Daily equal-weighted average of constituent returns, compounded from a base of 100 on April 7, 2025. No market-cap weighting. No free float. No dividend reinvestment.

Can I buy the indices?

Not directly. No ETF currently tracks this specific methodology. The constituent lists are public; a sufficiently motivated reader can replicate them.

Why equal-weighted?

Because market-cap weighting in a concentrated market produces a basket that is dominated by the top two names. Equal weighting is the simplest corrective. It also lets the mid-caps, which are where the actual levered plays on data center capex live, carry real weight.

How are constituents chosen?

Three tests: meaningful data-center-related revenue (roughly 10% or more), publicly traded with adequate liquidity, and not in one of the excluded categories. Reviewed quarterly. The last review added Bloom Energy, Talen Energy, EMCOR, Modine, GE Vernova, nVent, Schneider Electric, Arista, and Super Micro.

How often does it update?

Constituent prices update at every market close. The daily return feeds into the index calculation the next business day. Constituent membership is reviewed quarterly.

Where is the raw data?

Index history is available via /api/basket-index?index=supply_chain and /api/basket-index?index=ai_leaders. The full constituent lists are at /api/dc-pulse-index/constituents.

Methodology current as of Q2 2026. Reviewed quarterly.

Tags: index, methodology, pillar