Nvidia – Short Attack

A lot of the mutterings don’t hold water.

• Following what was an excellent set of results, questions have arisen with regard to whether Nvidia’s accounts are reflecting that demand is softer than we are being led to believe.
• RFM puts its financial hat on and concludes that while there are some issues, the rest look like sloppy financial analysis dressed up to support a short position in Nvidia’s stock.
  • First, Inventories: where FQ3 26 inventories increased by 32% QoQ which struck many as odd as Jensen stated that demand was “insane” and that there were some supply constraints.
  • At a high level, one would have expected inventories to fall as when demand is high and supply is constrained, inventory is sold and can’t be replaced.
  • This is where the published analysis is a bit sloppy as the situation at Nvidia is nuanced because the quantity of components from 3^rd parties that Nvidia is reselling is increasing very quickly.
  • This is caused by the migration of Hopper to Blackwell where Hopper is sold as a tray with GPUs on it that goes into a cabinet while Blackwell is sold as a whole cabinet.
  • This means that the number of 3^rd party components being sold per system has gone from 10s of thousands to over a million.
  • Hence, Nvidia is now buying in way more 3^rd party components than the other semiconductor companies that it is being compared against, meaning that its inventories are going to be larger than those that just sell chips.
  • This combined with the strong demand that Nvidia has forecast for FQ4 26 meaning it has to buy in more, is why the inventories have increased as opposed to weak demand causing unsold inventory to pile up.
  • Second, Gross margin: which I think is a sign of Nvidia’s strength as opposed to others who view FQ3 26 as a sign of weakness.
  • As the components of 3^rd parties increases in the revenue mix  pressure in increases on gross margins because one has to pay the margin of the supplier.
  • The fact that Nvidia can see a mix shift of this scale and maintain gross margins of 73.4% is a sign of the company’s ability to charge what it likes.
  • It is a signal that demand for its products is so strong that it is able to mark up 3^rd party products to maintain gross margins and clients are still willing to pay.
  • As the mix continues to move, I would expect gross margin to continue to drift downwards but not by a very large amount.
  • This is a sign of market power not weakness.
  • Third, Share buybacks: where Mr Burry makes a very valid point.
  • A share buyback has not returned the money to shareholders until the shares that have been repurchased are cancelled and the diluted share count declines.
  • A very large proportion of companies seem to think that buying back shares, holding them in treasury and then issuing them back to employees counts as returning cash to shareholders.
  • In fact, it is a way to hide staff costs by moving the expense from salaries to share-based compensation which everyone is encouraged to ignore when looking at the economic performance of the company.
  • Looking at the last two full fiscal years, I give Nvidia a C minus grade.
  • In fiscal 2024 and 2025, Nvidia spent $43.2bn on share buybacks which should have resulted in around 450m shares being bought.
  • However, during that period, the share count fell by only 266m, meaning that not all of the value was given back to shareholders.
  • All sorts of arguments around share-based compensation and acquisitions with shares can be made, but the reality is that the cash that was promised was not fully returned.
  • Nvidia has fared better than many of its peers, but not close to what I would call top-tier performance.
  • Fourth, Cash flow: where one of the top critic’s commentary that has been picked up and recycled (see here) states that cash flow from operations was just $14.5bn in FQ3 26.
  • This is inaccurate as cash flow from operations, which is a true reflection of how the business is doing, was $23.8bn which was $6bn lower due to the increases in working capital (inventories) I have addressed above.
  • If I reverse out the investments being made in growing the business by building inventory, then cash conversion was 83% which I think is pretty good.
  • Fifth, Circular deals: which are a cause for long-term concern, but are not about to trigger a collapse in Nvidia’s financial performance in the short-term.
  • With all of the parties being increasingly inter-related, there is a risk that a line of dominoes is being set up where a problem in one place causes problems elsewhere resulting in a systemic collapse.
  • However, these large deals all have an element of flexibility to them such that if demand weakens or there is not enough electricity to power a data centre, the deals can be delayed without penalties.
  • This is normal practice where long-term capital expenditure type deals are concerned and so the idea that OpenAI or Anthropic will be in a position where they are forced to buy more capacity triggering a collapse is not realistic.
  • Sixth, Smart money exit: which is the idea that because some well-known and historically successful investors have sold their positions, then that must signal the top.
  • Any investor who says that they can reliably pick the bottom and the top of a share price movement is not telling the truth.
  • The reality is that the easy money in Nvidia has been made and those who have a contrarian or value-based mindset will be inclined to move on other investments.
  • It does not mean that the top has been made or that a collapse is imminent, but it does mean that the meteoric rise is over and the shares are no longer cheap.
  • This is not new news to anyone who has a pulse.
• The net result, is that while Nvidia could certainly do a better job of returning cash to shareholders and there is growing risk of inter-related deals all hitting one another, I don’t see very much wrong with Nvidia’s financial statements.
• In fact, compared to many of the other AI companies that are burning money like crazy or trying to figure out how to make a return on compute that costs $40bn per GW to build, Nvidia looks pretty clean and transparent.
• There is certainly a correction of some size coming, which will hit Nvidia when it comes but for the next few quarters, the outlook remains pretty good.
• I suspect the shorts will get burned on this one unless, like Mr Burry, they can endure pain for an extended period.