AI equities scorecard, Q3 2026: megacaps carried a split quarter
A 20-stock AI basket rose 7.7% cap-weighted in Q3 2026, but the median stock fell 5.4% as hyperscalers led and AI-hosting miners dropped 30%.
Key findings
- A 20-stock AI equity basket returned +7.7% cap-weighted in Q3 2026 (July 1 to September 30), ahead of the S&P 500 proxy (SPY, +2.1%) and the Nasdaq-100 proxy (QQQ, +0.5%), but the equal-weighted mean was -3.3% and the median stock fell 5.4%.
- Hyperscalers and platforms led with a +12.2% equal-weighted Q3 return, driven by Microsoft (+37.5%) and Meta (+28.7%); bitcoin miners turned AI hosts lagged at -29.8%, with all five names lower.
- Three stocks (Microsoft, Nvidia, Meta) added an estimated $2.13T of market value in the quarter, more than the basket’s net gain of about $1.78T, as Alphabet, Broadcom, Micron and TSMC declined.
- Dispersion was wide: Q3 returns ran from -40.1% (TeraWulf) to +40.0% (Super Micro), an 80-point spread and a cross-sectional standard deviation of 22.8 points; only 8 of 20 stocks beat SPY.
- Concentration is high: the five largest names (Nvidia, Alphabet, Microsoft, Amazon, TSMC) were 73.1% of the basket’s $24.89T market cap on September 30; Nvidia alone was 22.2%, while the five miners together were 0.2%.
- Year to date, chips still lead on an equal-weight basis (+106.6%), lifted by Micron (+273.2%) and AMD (+185.7%), despite a flat Q3 for the layer.
Scope and headline result
This scorecard tracks 20 US-listed equities with direct exposure to the build-out of artificial intelligence, grouped into four layers of the stack: chips and semiconductors, hyperscalers and platforms, AI infrastructure and power, and bitcoin miners that have repositioned as AI and high-performance computing (HPC) hosts. Each layer holds five names. Returns are share price returns from closing prices on June 30, 2026 to September 30, 2026 for Q3, and from December 31, 2025 to September 30, 2026 for the year to date.
The headline depends heavily on weighting. Weighted by market capitalisation, the basket gained 7.7% in Q3, more than three times the 2.1% price return of SPY and well ahead of the 0.5% return of QQQ. Weighted equally, the same 20 stocks lost 3.3% on average, and the median stock fell 5.4%. Twelve of the 20 stocks ended the quarter lower. In other words, a small number of very large companies carried the group while the typical AI-exposed stock underperformed the broad market.
Scorecard by stock
The table below lists the sourced closing prices used for every calculation, the resulting returns and an end-of-quarter market capitalisation for each company. Market capitalisations are derived (see methodology) and are rounded.
| Company (ticker) | Close Dec 31, 2025 | Close Jun 30, 2026 | Close Sep 30, 2026 | Q3 2026 | YTD to Sep 30 | Market cap Sep 30 (derived) |
|---|---|---|---|---|---|---|
| Chips and semis | ||||||
| Nvidia (NVDA) | $186.50 | $200.09 | $228.38 | +14.1% | +22.5% | $5.52T |
| Broadcom (AVGO) | $346.10 | $377.75 | $351.19 | -7.0% | +1.5% | $1.68T |
| AMD (AMD) | $214.16 | $580.91 | $611.76 | +5.3% | +185.7% | $997.4B |
| TSMC (ADR) (TSM) | $303.89 | $477.57 | $456.19 | -4.5% | +50.1% | $1.97T |
| Micron (MU) | $285.41 | $1,154.29 | $1,065.11 | -7.7% | +273.2% | $1.20T |
| Hyperscalers and platforms | ||||||
| Microsoft (MSFT) | $483.62 | $373.02 | $512.90 | +37.5% | +6.1% | $3.81T |
| Alphabet (GOOGL) | $313.00 | $357.37 | $344.08 | -3.7% | +9.9% | $4.21T |
| Amazon (AMZN) | $230.82 | $238.34 | $249.15 | +4.5% | +7.9% | $2.69T |
| Meta (META) | $660.09 | $563.29 | $725.18 | +28.7% | +9.9% | $1.85T |
| Oracle (ORCL) | $194.91 | $146.55 | $137.30 | -6.3% | -29.6% | $416.2B |
| AI infrastructure and power | ||||||
| Vertiv (VRT) | $162.01 | $334.82 | $241.31 | -27.9% | +48.9% | $92.9B |
| Constellation Energy (CEG) | $353.27 | $248.37 | $254.02 | +2.3% | -28.1% | $90.0B |
| Arista Networks (ANET) | $131.03 | $169.88 | $203.59 | +19.8% | +55.4% | $256.8B |
| CoreWeave (CRWV) | $71.61 | $99.54 | $87.12 | -12.5% | +21.7% | $48.1B |
| Super Micro (SMCI) | $29.27 | $29.33 | $41.07 | +40.0% | +40.3% | $27.0B |
| Bitcoin miners turned AI hosts | ||||||
| IREN (IREN) | $37.77 | $45.73 | $40.88 | -10.6% | +8.2% | $16.1B |
| Cipher Digital (CIFR) | $14.76 | $24.50 | $15.83 | -35.4% | +7.2% | $6.6B |
| TeraWulf (WULF) | $11.49 | $24.70 | $14.80 | -40.1% | +28.8% | $7.4B |
| Core Scientific (CORZ) | $14.56 | $25.59 | $15.98 | -37.6% | +9.8% | $5.1B |
| Hut 8 (HUT) | $45.94 | $115.44 | $86.10 | -25.4% | +87.4% | $10.6B |
| SPDR S&P 500 ETF (SPY), S&P 500 proxy | $681.92 | $746.77 | $762.63 | +2.1% | +11.8% | n/a |
| Invesco QQQ (QQQ), Nasdaq-100 proxy | $614.31 | $736.40 | $739.77 | +0.5% | +20.4% | n/a |
Which layer led and which lagged
| Layer | Q3 equal-weight mean | Q3 median | Q3 cap-weighted | YTD equal-weight mean | YTD cap-weighted | Market cap Sep 30 (share of group) |
|---|---|---|---|---|---|---|
| Hyperscalers and platforms | +12.2% | +4.5% | +11.9% | +0.9% | +6.5% | $12.97T (52.1%) |
| AI infrastructure and power | +4.3% | +2.3% | +1.9% | +27.6% | +25.1% | $514.7B (2.1%) |
| Chips and semis | +0.0% | -4.5% | +3.8% | +106.6% | +39.6% | $11.36T (45.6%) |
| Bitcoin miners turned AI hosts | -29.8% | -35.4% | -27.2% | +28.3% | +23.5% | $45.8B (0.2%) |
| All 20 stocks | -3.3% | -5.4% | +7.7% | n/a | +19.8% | $24.89T (100%) |
Hyperscalers and platforms were the strongest layer in Q3 on both an equal-weight (+12.2%) and cap-weighted (+11.9%) basis. The result was narrow. Microsoft rose 37.5% and Meta 28.7%, Amazon added 4.5%, while Alphabet (-3.7%) and Oracle (-6.3%) declined. Microsoft’s quarter followed its fiscal fourth quarter report on July 29: coverage reported Azure growth of 43% against guidance of 39% to 40%, guidance of 45% for the next quarter, and capital expenditure guidance of more than $50 billion for the September quarter. The shares rose as much as 17% on July 30, which Fortune described as the largest one-day gain since 2008. Year to date the layer is nearly flat on an equal-weight basis (+0.9%) because Oracle remains 29.6% below its 2025 close and Microsoft only recovered its first-half losses in Q3.
AI infrastructure and power returned +4.3% equal-weighted and +1.9% cap-weighted, with the widest internal spread of any layer (standard deviation of 23.8 points). Super Micro (+40.0%) and Arista Networks (+19.8%) gained, Constellation Energy edged up 2.3%, and CoreWeave (-12.5%) and Vertiv (-27.9%) fell. A Nasdaq.com review attributed Vertiv’s July decline to broader caution toward AI stocks during the month and a quarterly report in which revenue growth of 24% (to $3.72 billion) slowed from the prior quarter, even though the company raised its full-year guidance.
Chips and semiconductors were flat on an equal-weight basis (+0.0%), with a median of -4.5%. Nvidia (+14.1%) and AMD (+5.3%) rose; Micron (-7.7%), Broadcom (-7.0%) and TSMC (-4.5%) fell. Nvidia reported fiscal second quarter 2027 revenue of $96.2 billion on August 26, up 106% year on year, with data center revenue of $89.0 billion and an outlook of $108.0 billion (plus or minus 2%) for the following quarter; its shares closed 8.7% higher on August 27. Micron also reported fiscal fourth quarter revenue of $54.23 billion, ahead of expectations, yet the shares closed 2.05% lower on the day of the reaction, according to Finviz coverage, after a year-to-date gain of roughly 273%. The Q3 pullback in memory and foundry names reads, in part, as consolidation after very large first-half gains: Micron and AMD were up 304.4% and 171.3% respectively in the first half alone.
Bitcoin miners turned AI hosts were the clear laggard at -29.8% equal-weighted, with every constituent lower: TeraWulf -40.1%, Core Scientific -37.6%, Cipher Digital (formerly Cipher Mining) -35.4%, Hut 8 -25.4% and IREN -10.6%. These stocks had run hard in the first half (TeraWulf, Cipher and Core Scientific each gained between 66% and 115% from December 31 to June 30), so the Q3 drawdown left all five still higher year to date. Coverage of an August 21 sell-off by 24/7 Wall St. characterised the moves as profit-taking that tracked each stock’s year-to-date gain, noting that a data center ETF fell only 0.6% that day.
Dispersion and concentration
Dispersion within the AI trade was unusually high for a single thematic group. The best Q3 performer (Super Micro, +40.0%) and the worst (TeraWulf, -40.1%) were separated by 80.1 percentage points, and the cross-sectional standard deviation of the 20 quarterly returns was 22.8 points. Dispersion was also visible inside individual layers: in hyperscalers, Microsoft and Oracle were separated by 43.8 points, and in infrastructure, Super Micro and Vertiv by 68.0 points.
Ownership of the basket’s value is highly concentrated. On September 30 the 20 companies had a combined derived market capitalisation of about $24.89 trillion. Nvidia accounted for 22.2%, Alphabet 16.9%, Microsoft 15.3%, Amazon 10.8% and TSMC 7.9%, for a top-five share of 73.1%. The chips and hyperscaler layers together made up 97.7% of the total, the infrastructure and power layer 2.1% and the five miners 0.2%. Because of this, the cap-weighted result is effectively a statement about a handful of megacaps. Measured in dollars, Microsoft added an estimated $1.04 trillion of market value in Q3, Nvidia $0.68 trillion and Meta $0.41 trillion; together this exceeded the basket’s net gain of about $1.78 trillion, with Alphabet ($0.16 trillion), Broadcom ($0.13 trillion), Micron ($0.10 trillion) and TSMC ($0.09 trillion) the largest detractors.
Macro backdrop: earnings, capex and the September 16 hike
Three themes framed the quarter. First, earnings from the largest suppliers and buyers of AI compute were strong in reported terms: Nvidia’s revenue more than doubled year on year and Micron’s fiscal fourth quarter revenue of $54.23 billion beat expectations. Second, hyperscaler capital expenditure guidance stayed elevated. Microsoft guided to more than $50 billion of capex for the September quarter, and Meta narrowed its 2026 capex range to $130 billion to $145 billion when it reported on July 29. Market reactions to capex were not uniform: Microsoft rose sharply after its report, while coverage of Meta’s report described an initial after-hours decline of roughly 8% to 10% tied to spending; Meta nonetheless finished the quarter 28.7% higher. This report does not attribute Meta’s later gains to a single catalyst.
Third, monetary policy tightened. On September 16, 2026 the Federal Open Market Committee raised the federal funds target range by 25 basis points to 3.75% to 4.00%, citing inflation that remained well above its 2% target. The hike came late in the quarter, and the closing data in this scorecard do not isolate its effect on individual stocks. Over the full quarter, the broad market proxies were modestly positive (SPY +2.1%, QQQ +0.5%), and the higher-beta, smaller-capitalisation parts of the AI trade (the miners, CoreWeave and Vertiv) recorded the largest declines.
Year-to-date context
Through September 30, the cap-weighted basket was up 19.8% year to date, compared with 11.8% for SPY and 20.4% for QQQ. On an equal-weight basis, chips led (+106.6%), followed by the miners (+28.3%), infrastructure and power (+27.6%) and hyperscalers (+0.9%). The two largest year-to-date gainers were Micron (+273.2%) and AMD (+185.7%); the two largest decliners were Oracle (-29.6%) and Constellation Energy (-28.1%). Q3 therefore partly reversed the first-half leadership: the layers and stocks with the biggest first-half gains (memory, foundry, miners, Vertiv) gave back ground, while hyperscalers that had lagged in the first half recovered.
Methodology
Universe. 20 US-listed stocks selected for direct AI exposure, five per layer. TSMC is represented by its New York-listed ADR (TSM). Alphabet is represented by class A shares (GOOGL). IREN and Hut 8 are included as US-listed AI hosting names; Cipher Mining now trades as Cipher Digital Inc. under the same ticker. Selection is illustrative, not an index, and excludes other relevant names.
Prices and returns. Daily closing prices for December 31, 2025, June 30, 2026 and September 30, 2026 were taken from stockanalysis.com price history pages and its price history data feed. Each close was cross-checked against the prior trading day’s close and the reported daily percentage change. Returns are simple price returns computed in Python as (end close / start close) minus 1, using unadjusted closes; dividends are excluded, which slightly understates total return for dividend payers. Q3 is June 30 to September 30, 2026; YTD is December 31, 2025 to September 30, 2026.
Benchmarks. The S&P 500 and Nasdaq-100 are represented by the SPY and QQQ ETFs, whose price returns closely track, but do not exactly equal, the indices. Index levels themselves were not used.
Market capitalisation. stockanalysis.com reports market cap and last close as of October 5, 2026. September 30 market cap is derived as the October 5 market cap multiplied by the ratio of the September 30 close to the October 5 close, which assumes an unchanged share count over those three trading days. Reported market caps are rounded to three or four significant figures, so derived values carry similar rounding. Alphabet’s figure reflects all share classes.
Aggregates. Equal-weight figures are simple averages of the stock returns. Cap-weighted returns compare total September 30 market cap with an implied starting market cap (September 30 cap divided by one plus the stock’s return), which assumes constant share counts over the period. Dollar contributions use the same implied starting values. Dispersion is the population standard deviation of the 20 Q3 returns.
Limitations. Attribution of moves to earnings, capex or policy is based on contemporaneous media coverage and is descriptive, not causal. This report is not investment advice.
Sources
- Stock Analysis, daily price history pages for all 20 stocks (NVDA shown) (as of 2026-10-05)
- Stock Analysis, company overview pages: market cap and last close (NVDA shown) (as of 2026-10-05)
- Stock Analysis, SPY and QQQ price history (as of 2026-10-05)
- NVIDIA Q2 fiscal 2027 results press release, SEC Form 8-K exhibit (as of 2026-08-26)
- Fortune, Microsoft’s biggest one-day gain since 2008 (as of 2026-07-30)
- Digital Applied, Meta Q2 2026 results and capex guidance (as of 2026-07-29)
- Finviz, Micron fiscal Q4 2026 results and share reaction (accessed 2026-10-06)
- Nasdaq.com, Why Vertiv Holdings plunged in July (as of 2026-08-11)
- 24/7 Wall St., AI miner sell-off of August 21, 2026 (as of 2026-08-21)
- Admirals, Fed raises rates to 3.75% to 4.00% (as of 2026-09-18)
New reports, trackers and scorecards on markets, AI and digital assets, sent when they publish.
Subscribe to the newsletterQuestions on this report
How did AI stocks perform in Q3 2026?
A 20-stock basket of US-listed AI-exposed equities returned +7.7% cap-weighted from June 30 to September 30, 2026, versus +2.1% for SPY and +0.5% for QQQ. Equal-weighted, the basket fell 3.3%, and 12 of the 20 stocks declined.
Which part of the AI stack led in Q3 2026?
Hyperscalers and platforms led with a +12.2% equal-weighted return, driven by Microsoft (+37.5%) and Meta (+28.7%). Bitcoin miners turned AI hosts lagged at -29.8%, with all five names lower.
How concentrated is the AI equity trade?
On September 30, 2026 the five largest companies (Nvidia, Alphabet, Microsoft, Amazon, TSMC) were 73.1% of the basket's roughly $24.89 trillion market cap. Nvidia alone was 22.2%; the five miners combined were 0.2%.
BlockWest Research reports are built from public filings, company disclosures and third-party data providers named in each report. Figures are point-in-time and are not updated after publication unless noted. This is research and commentary, not investment advice. Spotted an error? Email the desk via the contact page.