# More Proof, Lower Price

Source: https://www.vistashares.com/ais-pm-commentary-more-proof-lower-price/
Updated: 2026-11-13

[Insights](https://www.vistashares.com/insights)
> Article

### 10 Nov 2026

#
More Proof, Lower Price

- AIS
- AI infrastructure stocks
- AI valuations
- Semiconductors
- AI capex

Written by[David Fetherstonhaugh](https://www.vistashares.com/team-members/david-fetherstonhaugh/)

Chief Investment Officer, VistaShares[Download the PDF](https://www.vistashares.com/docs/ais-pm-commentary-more-proof-lower-price/vistashares-ais-pm-commentary-october-2026/?f=2026%2F11%2Fvistashares-ais-pm-commentary-october-2026.pdf)

**AIS Portfolio Manager Commentary** | VistaShares Artificial Intelligence Supercycle® ETF (AIS) | October 2026

**NYSE: AIS | NAV $53.44 | YTD +37.3% (11/09/2026)**

We are observing a very stark contrast between what the market is willing to give companies credit for and what customers are willing to pay to secure supply. Since our last letter, the AI infrastructure group has absorbed a midterm election, a Federal Reserve rate hike, a hot inflation print and a decline of roughly 24% in AIS. Over the same period, the two largest AI labs reported their fastest revenue growth to date, the largest memory supplier in the United States posted a record quarter, and the chief executive of OpenAI flew to Seoul to secure memory supply through 2029.

AIS closed November 9 at a NAV of $53.44, down roughly 24% from where we wrote in late August and about 40% below the record high of $88.80 set in June, which is the deepest drawdown since the fund launched. We do not want to minimize that, and we do not think the things investors are worried about are imaginary. An election that turned local opposition to data centers into policy, interest rates that are rising again, and a buildout that increasingly depends on outside capital are all real developments, and we spend a good part of this letter on each of them. What we want to do first is describe what has happened to the prices of these businesses, and then what the businesses themselves have reported over the same months, because the two have moved in opposite directions and we think that is the most useful thing we can show you.

## What the Market Is Paying

Semiconductor companies have usually been valued at a premium to the broader market, which makes sense for an industry that has grown faster than the average company. Over the past five years the S&P 500 Semiconductors index has traded at about 1.18 times the forward earnings multiple of the S&P 500, and in June 2024 it reached 1.66 times. As of November 3 it traded at 13.8 times forward earnings against 18.3 times for the S&P 500, a ratio of 0.75, which is lower than the 0.80 reached in May 2020, before AI had appeared in anyone’s earnings report.

**Exhibit 1. Widest discount to the S&P 500 since 2018** (semiconductors forward P/E divided by S&P 500 forward P/E)

Point in time |
Ratio |

May 2020 trough |
0.80x |

Five year average |
1.18x |

June 2024 peak |
1.66x |

November 3, 2026 |
0.75x |

S&P 500 Semiconductors, forward P/E |
13.8x |

S&P 500, forward P/E |
18.3x |

Semis forward P/E, January 2026 peak |
24.9x |

Change from January peak |
-45% |

Versus five year average multiple |
-42% |

Source: Bloomberg, S&P Global, via Citadel Securities, as of November 3, 2026. Parity with the S&P 500 multiple is 1.00x.

A lower multiple can mean one of two things. It can mean that investors expect earnings to fall, in which case the stocks are not really cheaper once the estimates catch up, or it can mean that investors are simply willing to pay less for the same earnings. What stands out to us about the period from June 30 to November 3 is that earnings estimates did not fall. According to Bloomberg data compiled by Citadel Securities, forward earnings estimates for semiconductors rose about 30% over those four months while the forward multiple fell about 35%. The pattern was nearly identical for the broader AI infrastructure group, and it was more pronounced in memory, where estimates rose about 32% and the multiple was cut in half. In each case the multiple compressed by more than the estimates rose, which is why prices fell even as expected earnings improved, and it means the market is now paying less for businesses whose results and visibility have gotten better.

**Exhibit 2. Multiples compressed by more than earnings estimates rose** (June 30 to November 3, 2026)

Group |
Change in forward P/E |
Change in forward earnings estimates |

Semiconductors |
-35% |
+30% |

AI infrastructure |
-36% |
+31% |

Memory |
-50% |
+32% |

Source: Bloomberg as compiled by Citadel Securities, Global Market Intelligence.

We want to be careful about how much weight this carries. A forward multiple is only as reliable as the estimate underneath it, and estimates can be revised down as quickly as they were revised up, particularly if the large cloud companies were to slow their spending. The group was also expensive enough in January, at nearly 25 times forward earnings, that some correction was due regardless of what followed. Even so, what we see is earnings that are becoming better supported by reported results and signed contracts, and in many cases accelerating, at the same time as multiples are compressing. Citadel Securities made the same observation in its November note, pointing out that the repricing has been sharpest in the parts of the market where earnings revisions have been strongest, and that every semiconductor company in the S&P 500 now trades at a lower multiple than it did at the end of June. That combination tells us that the decline in these stocks has come from what investors are willing to pay and not from what the companies are expected to earn, which leads naturally to the question of whether those estimates deserve to be trusted. Much of the rest of this letter is our attempt to answer that by looking at what the customers of these companies are actually doing.

## What the Businesses Are Reporting

The money that eventually reaches a memory maker or an optical component supplier starts with someone paying for AI, so the first place we look is the companies that sell it. Anthropic’s amended filing shows third quarter revenue of $16.96 billion, up 47% from the second quarter alone, and it also shows where that revenue is coming from. Customers spending more than $100,000 a year grew from about 200 at the end of 2024 to more than 11,900 at the end of September, and they account for roughly three quarters of revenue. OpenAI told investors that its third quarter revenue rose about 50% from the prior quarter to roughly $10 billion. One estimate we found useful came from David Friedberg on the All-In podcast, who put the number of companies worldwide large enough that $100,000 a year would be less than 1% of their spending at a little over one million, which would mean Anthropic has reached something like 1% of its potential customers. That is an estimate and not a fact, but it is a helpful way of seeing how early this still is.

One step down, the cloud businesses that rent out the computing power reported the same thing from the other side. Google Cloud revenue roughly doubled from a year earlier to $30.7 billion and its backlog of contracted future revenue is near $561 billion, AWS grew 39% to $46.0 billion, which is its fastest rate in nearly five years, and Microsoft said Azure grew 48% and that it had to push some capacity into the first half of 2027 because its suppliers could not deliver quickly enough. These are also profitable businesses, with AWS and Google Cloud earning a combined $28.9 billion of operating income in the quarter, although operating profit at a cloud division is not the same thing as an adequate return on all of the capital that has been invested, and that is a question that will take years to settle.

**Exhibit 3. Revenue at the model companies and the clouds** (most recent reported figures)

Company |
Revenue |
Growth |
Forward signal |

Anthropic |
$17B |
+1,127% y/y |
Up 47% from the second quarter |

OpenAI |
$10B |
+50% q/q |
Enterprise run rate up 107% |

Google Cloud |
$30.7B |
+103% y/y |
Backlog near $561B |

Amazon AWS |
$46.0B |
+39% y/y |
Fastest growth in nearly five years |

Microsoft Azure |
n/d |
+48% y/y |
Capacity pushed into 1H 2027 by supply limits |

Source: company filings and earnings releases; Anthropic Form S-1/A (November 2, 2026). OpenAI figures are unaudited, as reported by The Information. n/d = not disclosed.

The newest evidence, and the least complete, comes from the businesses that are buying AI and trying to put it to work. Surveys published since our last letter suggest that most of them believe it is paying off. In a Wharton and GBK Collective survey of U.S. enterprise leaders, 80% reported a positive return on their AI spending, and Deloitte found that the share of companies reporting revenue gains from AI rose from 20% in January to 28%. Among public companies, Morgan Stanley found that operating margins for the companies it classifies as AI adopters expanded by 310 basis points from 2024 to 2025, about twice the improvement in the MSCI World index.

We do not want to overstate this. In the same Wharton survey only 28% described their return as strong, in McKinsey’s annual survey only 37% of respondents could attribute any operating profit to AI, which is unchanged from a year ago, and J.P. Morgan Asset Management points out that only 2% of S&P 500 companies have put a number on its effect on their earnings. Most of the revenue so far comes from a fairly narrow set of uses, chiefly writing and building software, and the products aimed at people who do not write code, such as Meta’s Muse and OpenAI’s Dots, launched only in the past three months and are still at the stage where usage comes before revenue. Total AI revenue, at roughly $175 billion a year, is also still small next to the roughly $800 billion that Goldman Sachs estimates the largest cloud companies will spend on capital projects this year. Our reading is that these limits describe a technology that is early in being adopted and not one that is failing to find customers, but it is a reading, and what would change our mind is a sharp slowdown in enterprise revenue at the model companies or a decline in the cloud backlogs, neither of which has happened so far.

## What Customers Are Committing To

Revenue growth at the AI labs matters to AIS because of what those companies do with it, which is to buy computing capacity, and computing capacity is built from the chips, memory, networking and components that our holdings make. The clearest sign of how much capacity they want is that they are committing to it years in advance. Anthropic, which set the price range for its initial public offering this morning, has earmarked roughly $85 billion of the proceeds for compute and infrastructure, and Microsoft disclosed $10 billion of revenue from its OpenAI arrangements in the September quarter alone.

Memory is where this shows up most directly. Micron reported fiscal fourth quarter revenue of $54.2 billion, above the high end of its own guidance, and said that more than 75% of its fiscal 2027 shipments are already under allocation or contract, which is an unusual amount of visibility for a business that has historically sold much of its output at whatever the price happened to be that quarter. SK hynix has said that customer requests for its high bandwidth memory exceed what it can supply by more than two to one, and on the first weekend of November Sam Altman flew to South Korea unannounced to meet executives at Samsung and SK Group in order to secure more of it.

CoreWeave, which reported after the close today, gave a similar picture for computing capacity itself. Its backlog of contracted revenue rose to $132.5 billion from $104 billion at the end of June, management said the company is largely sold out through 2027 with far more qualified demand than it can serve, and it raised prices twice during the quarter, which is not something a supplier can do when its customers have alternatives.

None of this removes the risk that things go wrong between a signed contract and a delivered product. Contracts can be renegotiated, construction can run late, and a full order book at a supplier tells us about demand and not about the returns its shareholders will ultimately earn. CoreWeave itself carries close to $47 billion of debt and reported a net loss of $709 million in the same quarter. What the commitments do tell us is that the customers closest to the end demand are behaving as though they expect it to keep growing, and that much of what our holdings will produce in 2027 has already been spoken for.

## The Concerns Behind the Selloff

With that as background, we want to take the concerns behind the selloff one at a time, because each of them is real and we think the useful question in each case is how much of the spending already underway it can actually reach.

The largest single driver of the drawdown was the midterm election on November 3. Democrats took both houses of Congress along with a large number of governorships and state legislatures, Texas elected a governor who had campaigned on tougher limits for data center development in a state that holds roughly a quarter of the planned capacity in the country, and voters in ten states decided 41 local measures on data centers, nearly all of which went against new development. Two days later technology stocks had their sharpest selloff since 2020, with Nvidia down roughly 9% and Micron down 17.5% in a single session, as analysts estimated that 10% to 20% of planned U.S. AI data centers could be cancelled or face a much longer path to approval.

The distinction we think matters here is between projects that have been announced and projects that are already being built, because local restrictions work by stopping new approvals and have very little say over a site that has already cleared zoning and has steel in the ground. Construction data as of November 5 shows 86.7 gigawatts of U.S. data center capacity in progress, of which 50 gigawatts has reached vertical construction and another 22.5 gigawatts is more than three quarters complete. The counties where binding restrictions passed hold less than 5% of that total, and most of the capacity in those counties is already standing.

**Exhibit 4. Most of the pipeline is past the point a ban can reach** (U.S. data center capacity under construction)

Under construction |
86.7 GW |

Vertical construction or beyond, 58% |
50.0 GW |

More than 75% complete, 26% |
22.5 GW |

Source: Aterio construction tracker, as of November 5, 2026. Vertical construction begins at first steel, roughly 20% completion.

Texas is the case everyone is watching, and the governor-elect said after the election that projects already permitted will be allowed to finish, which covers nearly all of the state’s expected deliveries for 2027. What voters mostly asked for is that data centers pay for the power and grid upgrades they require, which the largest developers had already begun committing to, and this week the Senate advanced a bipartisan permitting bill along the same lines by a vote of 73 to 25. We expect new rules and not broad bans, and independent estimates put the reduction in national capacity at 2% to 4%. There is a real cost in this, since projects that are still only announcements will take longer and some capacity may be built overseas instead, and construction that is underway can still be delayed by labor, equipment or a slow utility connection. For AIS the more relevant point is that our holdings mostly sell chips, memory, optics and components into that capacity wherever it ends up being built, while the companies most exposed to a slowdown in Texas are the ones that own land and power there, which is a small part of the portfolio.

The second concern is financing, and here we think the worry is well placed. In the third quarter capital spending consumed 68% of Microsoft’s operating cash flow, 106% of Meta’s, 121% of Alphabet’s and 182% of Amazon’s, which means that three of the four largest buyers are now spending more than their businesses generate and are making up the difference with outside capital. Anthropic reported an operating loss of $567 million in the quarter as its compute costs rose, and OpenAI raised $11 billion in its latest round against a reported target of $20 billion. A buildout funded this way depends on capital markets staying open, and if they were to tighten we think the orders most at risk would be the ones for 2028 and beyond that have not yet been placed, and not the supply already under contract for next year.

The third is interest rates. This morning’s October inflation report came in above expectations, with core prices now rising at a 3.7% annualized pace over the past three months, and a rate increase in December is largely priced in. Higher rates reduce what investors will pay for any company whose earnings lie mostly in the future, and that has been a large part of the multiple compression described at the start of this letter. Rates do not change the orders that are already in place, but they can keep valuations low for longer than the underlying results would suggest, and we expect them to remain the main source of short term volatility in the fund.

## Portfolio

We have made no changes to the portfolio since our last trades on September 23. The declines over this period followed politics, rates and sentiment, and because we did not see a change in orders or earnings among our holdings we did not see a reason to sell, while the unsettled outlook for rates and policy meant we also did not see a reason to add.

The most significant single detractor in the period was Marvell, following its announced acquisition of Lumentum. We continue to believe in Marvell’s position in custom silicon and connectivity, and we see real strategic logic in owning indium phosphide supply, since those lasers are the scarcest input in the optical chain. What we question is the structure of the deal, meaning the price paid and the dilution from an all-stock transaction of roughly $115 billion, and the stock fell meaningfully after the announcement.

## Performance Context

AIS closed November 9 at a NAV of $53.44 with $864 million in assets, down from $75.94 and just over $1 billion at the end of September. Year to date through November 9, AIS is up approximately 37.3% on a NAV basis. As of October 31, the most recent month end, AIS was up 59.12% year to date on a NAV basis, 50.10% over one year and 140.26% since inception.

We began this letter with the contrast between what the market is paying for these companies and what their customers are paying to secure supply, and after going through the evidence that contrast is still where we end up. Since June the stocks have fallen sharply and the multiples investors will pay have fallen further, for reasons that are understandable and in some cases may persist. Over the same months earnings estimates have risen, the companies selling AI have reported faster growth, their customers have committed to supply further into the future, and most of the capacity that was under construction is still being built. Among our largest holdings we have not seen cancelled orders or cuts to capital spending guidance. We cannot say when sentiment will improve, and estimates can change, but on the evidence we have today we believe the earnings outlook for the businesses in the portfolio is stronger than it was when we wrote in August.

## IMPORTANT INFORMATION: FOR INVESTMENT PROFESSIONAL USE ONLY

For standardized performance, current holdings, and the fund prospectus, please visit [vistashares.com/etf/ais](https://www.vistashares.com/etf/ais/).

*Performance data quoted represents past performance; past performance does not guarantee future results. The investment return and principal value of an investment will fluctuate so that an investor’s shares, when redeemed, may be worth more or less than their original cost. Current performance of the fund may be lower or higher than the performance quoted. Performance current to the most recent month-end can be obtained by calling (844) 875-2288.*

**Investors should consider the investment objectives, risks, charges and expenses carefully before investing. For a prospectus or summary prospectus with this and other information about the Fund, please call (844) 875-2288. Read the prospectus or summary prospectus carefully before investing.**

Investing involves risk, including possible loss of principal.

Holdings are subject to change and should not be considered a recommendation to buy or sell any security. References to specific securities and their issuers are for illustrative purposes only.

Third party data is cited for context. VistaShares is not affiliated with Aterio, S&P Dow Jones Indices, Bloomberg, Citadel Securities, Morgan Stanley, Goldman Sachs, J.P. Morgan Asset Management, McKinsey, Wharton and GBK Collective, Deloitte, All-In Podcast or any company or official referenced and has not independently verified their figures. Construction data reflects Aterio’s tracker as of November 5, 2026. Valuation data reflects S&P 500 Semiconductors and S&P 500 forward price to earnings as of November 3, 2026. Capacity delivery figures for 2026 through 2028 are VistaShares estimates based on the construction pipeline and are inherently uncertain; actual results may differ materially. Statements attributed to public officials and executives are paraphrased from public remarks.

**Artificial Intelligence Risk.** Issuers engaged in artificial intelligence typically have high research and capital expenditures and, as a result, their profitability can vary widely, if they are profitable at all. The space in which they are engaged is highly competitive and issuers’ products and services may become obsolete very quickly. These companies are heavily dependent on intellectual property rights and may be adversely affected by loss or impairment of those rights.

**Consumer Discretionary Sector Risk.** The success of electrical consumer product manufacturers and retailers is tied closely to the performance of the overall domestic and global economy, interest rates, competition and consumer confidence. Success depends heavily on disposable household income and consumer spending, especially when it comes to green or clean renewable energy solutions.

**Equity Market Risk.** Common stocks are generally exposed to greater risk than other types of securities, such as preferred stock and debt obligations, because common stockholders generally have inferior rights to receive payment from specific issuers.

**Technology Sector Risks.** The Fund will invest substantially in companies in the technology sector, and therefore the performance of the Fund could be negatively impacted by events affecting this sector. Market or economic factors impacting technology companies and companies that rely heavily on technological advances could have a significant effect on the value of the Fund’s investments.

**Semiconductor Industry Risk.** Companies in the semiconductor industry may be affected by intense competition, cyclical demand, rapid product obsolescence, dependence on a limited number of customers and suppliers, and significant capital requirements.

**Foreign Securities Risk.** Investments in securities or other instruments of non-U.S. issuers involve certain risks not involved in domestic investments and may experience more rapid and extreme changes in value than investments in securities of U.S. companies.

**Swap Agreements Risk.** Swap agreements are entered into primarily with major global financial institutions for a specified period which may range from one day to more than one year. In a standard swap transaction, two parties agree to exchange the return (or differentials in rates of return) earned or realized on particular predetermined reference or underlying securities or instruments.

**Index Strategy Risk.** The Fund’s strategy is linked to an Index maintained by the Index Provider that exercises complete control over the Index.

**New Fund Risk.** The Fund is a recently organized management investment company with no operating history. As a result, prospective investors do not have an extensive track record or history on which to base their investment decisions.

**Foreside Fund Services, LLC, distributor.**