AllianceBernstein's Qiu: AI Has Left the Hardware Phase

Lei Qiu, AllianceBernstein's Chief Investment Officer for Thematic Innovation Equities, said on Bloomberg that the latest earnings reports show strong and persistent AI demand, with adoption broadening beyond hardware into select software and the infrastructure layer and no sign of a slowdown.

Lei Qiu, Chief Investment Officer for Thematic Innovation Equities at AllianceBernstein, told Bloomberg Television that the argument over whether artificial intelligence spending translates into real business activity is largely settled. Reading across the latest round of corporate earnings, Qiu said demand for AI is strong and persistent, and that the evidence now points to adoption broadening beyond hardware into select software names and the infrastructure layer that sits between the two.

That is a specific claim, and a more demanding one than the familiar bullish line about chip orders. Qiu's point is that AI has stopped being a capital expenditure story confined to a handful of suppliers and has started showing up inside the way companies actually run. In her framing, AI is "solidly moving into the adoption phase," with no sign of a slowdown.

Why the hardware-to-software handoff is the test

For most of the current cycle, the cleanest way to measure AI enthusiasm has been to count the money flowing to the companies that build the physical layer: accelerators, networking gear, power and cooling, and the data centre shells that house all of it. That spending is easy to verify because it appears as capital expenditure on the buyer's cash flow statement and as revenue on the seller's income statement in the same quarter.

The problem is that capital expenditure by itself proves conviction, not payback. A hyperscaler can spend heavily on compute for several years on the expectation of future demand. What it cannot do indefinitely is spend without eventual evidence that customers are paying for the output. That is why the software and infrastructure layers matter to the argument Qiu is making: those are the places where AI shows up as recurring revenue from end users rather than as a one-off equipment purchase.

"Infrastructure layer" in this context means the software and services that sit between raw silicon and a finished application — data platforms, orchestration, security, the plumbing that lets a company point a model at its own information without rebuilding its systems. It is unglamorous and, historically, sticky. Spending that lands there tends to be budgeted rather than opportunistic, which is what Qiu appears to mean by AI becoming "more integrated into business models."

What the tape looked like as she spoke

Qiu's remarks landed on a firm session for US equities. As of the last trade at 19:54:50 GMT on 3 September 2026, the SPDR S&P 500 ETF Trust (NYSEARCA: SPY) was at $773.13, up 1.04% from its previous close of $765.16, having traded between $766.83 and $774.03 on the day. The Invesco QQQ Trust (NASDAQ: QQQ), which tracks the tech-heavy Nasdaq 100 and is the closest liquid proxy for the AI complex, stood at $717.23, up 1.13%, against a previous close of $709.24 and a day range of $709.69 to $718.91.

The Dow, notably, was leading. The SPDR Dow Jones Industrial Average ETF Trust (NYSEARCA: DIA) traded at $536.88, up 1.18% from $530.62, inside a range of $532.94 to $537.62. A session in which the industrial-weighted index outpaces the Nasdaq proxy is not the signature of a narrow, chips-only rally. It is at least consistent with the broadening Qiu described, though a single day proves nothing on its own.

How to check the claim rather than take it

Qiu is a thematic innovation investor talking about the theme she is paid to invest in, so the appropriate response is not agreement but a checklist. Several things would confirm or undercut the adoption thesis over the next few quarters, and all of them are observable in filings rather than in commentary.

• Software revenue mix. Whether AI-attributed product lines are disclosed separately and whether they are growing off a base large enough to move consolidated numbers, rather than being described qualitatively in a shareholder letter.
• Net revenue retention. Adoption implies existing customers spending more year over year. Retention rates are the least gameable evidence that AI features are being paid for rather than bundled.
• Backlog and remaining performance obligations. Contracted future revenue is the bridge between a demand narrative and cash. Growth here, converting on schedule, is harder to dismiss than bookings commentary.
• Gross margin under load. Serving models costs money. If AI revenue arrives with materially worse gross margin, the adoption is real but the economics are not yet.
• Capex guidance revisions. Qiu says there is no sign of a slowdown. The first sign would be a hyperscaler trimming or deferring its capital plan, or shifting the mix from training to cheaper inference capacity.

Where the thesis is most exposed

The vulnerability in a broadening argument is the word "select." Qiu was careful to say adoption is spreading into select software, not software as a category. That qualifier does a lot of work. It concedes that a large share of the sector may be a loser rather than a winner from AI — vendors whose pricing is per-seat and whose seats are the very roles automation touches, or whose products can be reconstructed by a capable model.

The infrastructure layer carries its own risk, which is that the hyperscalers who buy the hardware also build competing middleware and can bundle it. Every layer of the stack is exposed to compression from the layer above or below.

There is also a timing distinction worth holding onto. Adoption phase does not mean returns phase. Companies can integrate AI into their business models and still see the productivity gains competed away into lower prices for their customers rather than retained as margin. Whether AI is a shareholder-value story or a consumer-surplus story is a separate question from whether it is being adopted, and Qiu's comments, as reported by Bloomberg Markets, address the second.

What to watch next

The near-term signals are earnings-driven. Watch for AI revenue disclosure moving from adjectives to line items; for the ratio of inference to training in disclosed data centre workloads, which tells you whether models are being used or merely built; and for whether the market's leadership continues to widen out from semiconductors, as the relative moves in the QQQ and DIA on this session did.

If Qiu is right, the next few reporting cycles should show AI revenue appearing in more places and in more ordinary forms — inside renewals, inside price increases, inside contracted backlog. If she is early, the hardware numbers will keep growing for a while and the rest of the stack will stay a story about pilots. The distinction will be visible in the filings well before it is visible in the commentary.

Frequently asked questions

What exactly did Lei Qiu say about AI?

Qiu, AllianceBernstein's Chief Investment Officer for Thematic Innovation Equities, told Bloomberg that the latest earnings reports point to strong and persistent AI demand. She said there is clear evidence adoption is broadening beyond hardware into select software and the infrastructure layer, that AI is becoming more integrated into business models, and that the technology is solidly in an adoption phase with no sign of a slowdown.

What is meant by the AI 'infrastructure layer'?

It refers to the software and services sitting between raw computing hardware and finished applications: data platforms, model orchestration, security tooling and integration plumbing. It lets a company apply AI models to its own data without rebuilding its systems. Spending there tends to be budgeted and recurring rather than one-off, which is why investors read it as evidence of genuine adoption rather than experimentation.

How were markets trading when the comments were made?

As of the last trade at 19:54:50 GMT on 3 September 2026, the S&P 500 proxy SPY was at $773.13, up 1.04%. The Nasdaq 100 proxy QQQ stood at $717.23, up 1.13%, and the Dow proxy DIA was at $536.88, up 1.18%. All three major US benchmarks were higher on the session, with the Dow leading.

Why does the shift from hardware to software matter?

Hardware spending shows conviction by buyers but not payback. Software and infrastructure revenue is where AI shows up as recurring payments from end customers rather than one-off equipment purchases. If AI-linked revenue appears in subscription lines, renewals and contracted backlog, it suggests the technology is being paid for in ordinary business budgets rather than funded from experimental spend.

What would disprove the adoption thesis?

Several observable signals would undercut it: a major cloud operator trimming or deferring capital expenditure guidance, AI-attributed software revenue staying qualitative rather than appearing as disclosed line items, weakening net revenue retention among software vendors, or contracted backlog failing to convert on schedule. Deteriorating gross margins on AI revenue would show adoption is real but the economics are not.

Does adoption automatically mean higher profits?

No. Companies can integrate AI into operations and still see the resulting productivity gains competed away into lower prices for customers rather than retained as margin. Adoption phase and returns phase are distinct. Whether AI ultimately accrues to shareholders or to consumers is a separate question from whether businesses are using it, which is what Qiu's comments address.

Sources

• Lei Qiu Says AI Is Moving Into Adoption Phase — Bloomberg Markets

Photo: Archie Binamira · Pexels Licence — source