'Buybacks Don't Lie': Leffler Reads Nvidia's Up Day
Libby Leffler told Bloomberg's 'The Close' that Nvidia's agent-safety system is a step forward and that 'buybacks don't lie,' as NVDA rose 1.68% against a Nasdaq 100 that closed 1.07% lower.
Libby Leffler told Bloomberg's 'The Close' on Sept. 28, 2026 that Nvidia's new system for keeping AI agents from 'going awry' is a step in the right direction and that 'buybacks don't lie,' arguing the session's weakness was macro-driven, tied to oil and Treasury yields, rather than a problem with Nvidia's model; NVDA last traded at 228.86, up 1.68%, while the Nasdaq 100 tracker QQQ closed down 1.07% at $736.53.
Nvidia (NVDA) finished a down day for the broad market on the other side of the tape. The chipmaker's shares last traded at 228.86, up 1.68% from the prior close of 225.07, with a session range of 228.04 to 233.21. The Nasdaq 100 tracking fund (NASDAQ: QQQ) closed at $736.53, down 1.07%. The S&P 500 tracker (NYSEARCA: SPY) ended at $765.61, off 0.74%, and the Dow tracker (NYSEARCA: DIA) at $514.02, down 0.67%. All figures are as of the last trade at 20:00 GMT on Sept. 28, 2026.
That split — one large-cap semiconductor higher while the index it dominates falls — is what Libby Leffler was asked to explain when she joined Romaine Bostick and Emily Graffeo on Bloomberg Technology's "The Close." Her reading: the softness was macro, not Nvidia-specific, driven by oil, Treasury yields and the general mood around AI headlines rather than anything in the company's underlying model.
Why the day's decline looked like a rates story
Leffler's framing matters because it separates two things that often get bundled together. When an AI-heavy index sells off, the reflexive interpretation is that investors have soured on AI demand. Her argument is that the transmission ran the other way around on this session: energy prices and the government bond market moved first, and equity multiples — which are most sensitive to the discount rate applied to distant cash flows — adjusted in response. Long-duration growth names are precisely the cohort that reprices hardest when yields rise, which is why a rates move shows up first in the Nasdaq 100 and only later, if at all, in the fundamentals.
The tape is at least consistent with that. The Dow tracker lost less than the S&P tracker, which lost less than the Nasdaq 100 tracker — the textbook ordering for a duration-driven day rather than a sector-specific scare. And the single most AI-levered mega-cap in the complex closed higher, not lower.
The gap between Nvidia's 1.68% gain and the Nasdaq 100 tracker's 1.07% decline works out to roughly 2.75 percentage points of relative performance in a single session — an illustrative spread calculated from the closing figures above, not a reported statistic. For a stock of Nvidia's weight, outperformance of that scale on a red day is unusual enough to require an explanation, and "the index fell for reasons that are not about this company" is a coherent one.
What "buybacks don't lie" is actually claiming
Leffler's line about buybacks is a statement about credibility, not about arithmetic. Guidance is a forecast. A management commentary is an interpretation. A share repurchase is cash that has irreversibly left the balance sheet, and it is the one capital-allocation decision that cannot be walked back with a footnote. When a board authorizes and then executes buying, it is making a costly signal — costly in the literal sense that the money is gone — that it believes the shares are worth more than the market is paying.
The skeptic's rebuttal is well known and worth stating plainly. Repurchases can be timed badly, and companies have a long history of buying most aggressively at cycle peaks when cash is abundant, which is exactly when valuations are least attractive. Buybacks can also mask dilution from stock compensation rather than genuinely shrinking the share count, so gross repurchase spending and net reduction in shares outstanding are not the same number. And an authorization is permission to buy, not a commitment to buy; the two figures can diverge for quarters at a time.
None of that invalidates Leffler's point, but it does define the test. The question for anyone acting on "buybacks don't lie" is whether the executed spend is tracking the authorization and whether the diluted share count is actually falling — details that sit in the cash flow statement and the share-count line, not in the headline announcement.
The agent-safety system and the second AI narrative
The other half of the conversation was Nvidia's new system designed to stop AI agents from "going awry" — the class of software that acts autonomously on a user's behalf rather than simply returning text. Leffler called it a positive step in the right direction.
The commercial logic behind that view is straightforward. Agentic AI is the deployment layer where enterprise budgets either open or stay shut, and the barrier is not model capability but accountability: who is liable when an autonomous process touches a production system, moves money, or changes a record it should not have. Controls that constrain what an agent can do are, in that sense, a sales tool as much as a safety feature. They convert an open-ended risk that a chief information security officer cannot sign off on into a bounded one that can be written into a contract.
For a company whose revenue depends on customers committing to multi-year infrastructure spending, anything that shortens enterprise procurement cycles is a revenue argument. That is the bridge between the safety headline and the buyback signal in Leffler's commentary — both, on her reading, point to a management team comfortable with the durability of demand.
What to check before taking the signal at face value
Three things resolve whether this reading holds. First, the direction of Treasury yields: if the macro explanation is right, a stabilization in rates should let the Nasdaq 100 close the gap with the Dow without any company-specific news. Second, the actual repurchase execution in the next filing — dollars spent against dollars authorized, and diluted share count against the prior period. Third, whether the agent-control system shows up in customer commentary as a reason deployments moved forward, or stays a product announcement without a demand consequence.
Until those land, the position is what the closing prices already say: on a session when the broad market fell and the tech-heavy index fell hardest, the stock at the center of the AI trade went the other way.
Key facts
- NVDA last trade: 228.86, +1.68% (prev close 225.07), as of 20:00 GMT Sept. 28, 2026
- Nasdaq 100 (QQQ): $736.53, -1.07% on the day
- S&P 500 (SPY) / Dow (DIA): $765.61, -0.74% / $514.02, -0.67%
- Commentary: Libby Leffler on Bloomberg's 'The Close' with Romaine Bostick and Emily Graffeo
Frequently asked questions
What did Libby Leffler say about Nvidia's buybacks?
Speaking on Bloomberg's 'The Close' with Romaine Bostick and Emily Graffeo, Libby Leffler said buybacks 'don't lie.' Her argument is that a share repurchase commits cash irreversibly, making it a more credible signal of management's confidence than guidance or commentary, which can be revised later without consequence.
How did Nvidia shares perform on Sept. 28, 2026?
Nvidia last traded at 228.86, up 1.68% from the prior close of 225.07, with a session range of 228.04 to 233.21. The figures are as of the final trade at 20:00 GMT that day. The stock rose while the broad market fell, an unusual divergence for a company of its index weight.
Why did the broader market fall that session?
Leffler attributed the weakness to macro factors — oil prices, Treasury yields and general sentiment around AI news — rather than any problem with Nvidia's underlying business model. The tracking funds bore that out: the Nasdaq 100 fund fell 1.07%, the S&P 500 fund 0.74% and the Dow fund 0.67%.
What is Nvidia's new AI agent system meant to do?
It is designed to keep AI agents from 'going awry' — that is, to constrain what autonomous software can do when acting on a user's behalf. Leffler described it as a positive step in the right direction. Such controls address the accountability questions that typically slow enterprise adoption of agentic AI.
Are buybacks always a reliable bullish signal?
Not automatically. Companies have historically repurchased most heavily near cycle peaks when cash is plentiful and valuations are high. Buybacks can also offset stock-compensation dilution rather than genuinely shrink the share count, and an authorization is permission to buy, not an obligation. The test is executed spend and the diluted share count.
Why do rising Treasury yields hit tech stocks hardest?
Technology and growth companies derive more of their value from cash flows expected far in the future. When bond yields rise, those distant cash flows are discounted more heavily, compressing valuations. That is why a rates-driven selloff typically shows up first in a tech-heavy index like the Nasdaq 100 rather than in the Dow.
Sources
- Libby Leffler on Nvidia Buyback, State of AI Trade — Bloomberg Technology
Photo: Pexels · Pixabay Content License — source
