Tepper Buys, Loeb Exits, Soros Buys Puts on NVIDIA

Three billionaire fund managers took opposing positions on NVIDIA in the same quarter: David Tepper added shares, Dan Loeb sold his entire stake, and George Soros bought put options, according to 24/7 Wall St.

The quarterly disclosure season rarely produces a cleaner disagreement than this one. Three of the most closely watched money managers in the United States looked at the same semiconductor stock over the same three months and reached three incompatible conclusions. David Tepper bought more of it. Dan Loeb sold every share he owned. George Soros bought put options — a contract that gains value when the underlying stock falls.

NVIDIA (NASDAQ: NVDA) last traded at 225.16, down 0.06% on the day from a prior close of 225.30, with an intraday range of 224.50 to 227.49, as of the close on Friday, 14 August 2026. That is a stock going almost nowhere on the day — which is a fair visual for the standoff the filings describe. The broad market was similarly still: the S&P 500 tracker (NYSEARCA: SPY) closed at $776.34, off 0.20%, and the Nasdaq 100 tracker (NASDAQ: QQQ) at $731.07, down 0.14%.

What each manager actually did

The three trades, as reported by 24/7 Wall St, are three different kinds of statement.

• Tepper added. Increasing an existing position is the mildest of the three signals. It says the manager was already comfortable and became more so — or that the price came to him.
• Loeb exited completely. Selling every share is categorical. It is not trimming for risk management or rebalancing a position that outgrew its sleeve. It removes the name from the book.
• Soros bought puts. This is the only one of the three that is an outright bet against the price, and it is the most ambiguous. Puts can express a short thesis, but institutions also buy them as portfolio insurance against a concentrated technology exposure held elsewhere. Without knowing what sits on the other side of the trade, the intent is unreadable from the filing alone.

That last point deserves emphasis, because put purchases are routinely reported as bearish calls when they are frequently hedges. A large fund with heavy AI-adjacent exposure across chips, power and data-center real estate might buy downside protection on the single most liquid proxy in that complex precisely so it can keep the rest of the book intact.

The disagreement is about capex, not chips

Nobody in this argument is disputing that NVIDIA sells the accelerators that AI model training requires. The dispute is about the durability of the spending that buys them. NVIDIA's revenue depends on the capital budgets of a small number of very large customers — hyperscalers, sovereign programs, and a lengthening tail of neoclouds. Those budgets are decisions, not physics, and they can be revised.

A bull case in Tepper's direction rests on the spending being structural: multi-year build-outs, committed power contracts, and inference demand that grows after training capacity is installed rather than shrinking. On that view, a stock trading near its recent range is not expensive against the earnings the build-out delivers.

A bear case in the Loeb or Soros direction does not require AI to fail. It only requires the rate of change in capex to slow, or customers to shift a meaningful share of workloads onto their own silicon. Peak growth rates and peak stock multiples tend to arrive at the same time, and a company this large has to grow off a base that keeps getting harder to lap.

Why 13F filings flatter the reader

Position disclosures are a rear-view mirror with a delay. They describe what a fund held on a quarter-end date, not what it holds now, and they say nothing about the price paid, the reason, or whether the trade has already been reversed. A manager can be out of a stock on the reporting date and back in it before the filing is public.

They also strip out context that changes the meaning entirely. A complete exit could reflect a valuation judgment — or a redemption, a mandate change, or the need to fund a higher-conviction idea somewhere else. An addition could be conviction or could be a systematic top-up. And derivative disclosures in particular are notoriously partial: the filings capture certain option positions without capturing the offsetting exposure that gives them purpose.

The useful takeaway is not to copy any of the three. It is that the smartest available money does not agree on this name, which means the range of plausible outcomes is genuinely wide.

What the tape is saying alongside the filings

The market data offers a modest counterpoint to the drama. NVIDIA's last session was a 0.06% drift lower inside a narrow band, in a tape where all three major benchmark trackers finished slightly negative. There is no evidence in that day's pricing of a stampede in either direction. The stock is being argued over, not repriced.

For investors, the practical questions are the ones the filings cannot answer:

• Does the next round of hyperscaler capital-spending guidance point up, flat or down?
• How much of the customer base is moving toward internal accelerators, and on what timeline?
• Is demand shifting from training clusters to inference capacity, and does that mix change pricing power?
• Do the same managers reverse course in the following quarter's disclosures — which is common enough that it should be assumed possible?

Reading a three-way split without picking a side

A stock with a unanimous institutional view is usually a stock with a crowded trade and a thin margin of safety. NVIDIA does not have that problem. What it has is a valuation that requires a forecast about the spending plans of a handful of companies, and reasonable people making opposite forecasts.

Anyone taking a position here should be explicit about which forecast they are making, and size accordingly. Tepper's addition, Loeb's exit and Soros's puts are all defensible expressions of a view. They are not, however, three signals pointing at one answer — and treating any single filing as a verdict is how the disclosure season most often misleads.

The next set of filings will show whether the divide held. In the meantime, the more informative data will come from customers rather than from fund managers: capital-expenditure commentary, power and site announcements, and whatever the largest buyers say about their own chip roadmaps.

Key facts

• NVDA last close: 225.16, -0.06%, as of 14 Aug 2026 20:00 GMT
• The three trades: Tepper added shares; Loeb sold his entire stake; Soros bought puts
• Same period: All three positions changed in the same quarter
• Market backdrop: SPY $776.34 (-0.20%), QQQ $731.07 (-0.14%), DIA $536.80 (-0.21%)

Frequently asked questions

What did David Tepper, Dan Loeb and George Soros each do with NVIDIA?

According to 24/7 Wall St, David Tepper bought additional NVIDIA shares, Dan Loeb sold every share he held, and George Soros bought put options on the stock. All three changes occurred during the same quarter, making it an unusually direct disagreement among three prominent fund managers over the same holding.

Does buying puts always mean a manager is bearish?

No. A put option gains value when the underlying stock falls, so it can express a bearish view. But large institutions also buy puts as insurance on a concentrated technology exposure held elsewhere in the portfolio. Position filings often disclose the option without the offsetting exposure, so intent cannot be read from the filing alone.

Where did NVIDIA stock last trade?

NVIDIA last traded at 225.16, down 0.06% from a prior close of 225.30, with a session range of 224.50 to 227.49, as of the market close on Friday, 14 August 2026. That was a narrow, quiet session rather than any sharp repricing in either direction.

Why are quarterly position disclosures a limited guide?

They report holdings as of a quarter-end date and become public with a delay, so a fund may already have reversed the trade. They also omit the price paid, the reason for the change, and any offsetting positions. A complete exit could reflect valuation, a redemption, or funding a different idea entirely.

What is the core disagreement over NVIDIA?

Not whether AI chips are in demand, but how durable the capital spending that buys them is. Bulls see multi-year, committed data-center build-outs with inference demand following training. Bears need only a slowdown in the rate of capex growth, or customers shifting workloads to their own in-house silicon, to compress the multiple.

What should investors watch next on NVIDIA?

Capital-expenditure guidance from the largest cloud buyers, announcements on power and data-center sites, the pace at which big customers adopt internally designed accelerators, and the mix shift from training to inference. The next round of fund disclosures will also show whether any of the three managers reversed course.

Sources

• Billionaires Battle on NVIDIA: David Tepper Bought More. Dan Loeb Sold Every Share. George Soros Bought Puts. — 24/7 Wall St

Photo: Alesia Kozik · Pexels Licence — source