Michael Burry, The Hyper-Scalers And A $176 Billion ‘Fraud’

12 thoughts on “Michael Burry, The Hyper-Scalers And A $176 Billion ‘Fraud’”

1. Whew.

2. I recall when Zuck announced Meta’s shift to longer term depreciation of servers on their quarterly call a few quarters back, and not a single Wall Streeter blinked. I thought it a yellow flag, because protestations to the contrary he admits that earnings are being managed. Burry’s just using inflammatory language to describe common corporate behavior. I do believe Meta and Oracle bear watching in this respect, as both are investing beyond their means. Blue Owl/Meta’s SPV to offload capex from Meta’s balance sheet is another clear signal.

3. Has anyone estimated the hyperscalers’ AI profits vs their AI capex, for a rough EBIT/capex ratio? I tried this for AMZN and got, with lots of handwaving and licked finger in wind, a guesstimated EBIT/capex of about 3.5%. 1/0.035 = 28.6. So why does it matter if the useful life of an AI GPU is 3 years or 6 years?

4. Seems Burry just following standard short seller procedure – Dump and Pump, no? Maybe he can move the market just a little enough to capture a sizable trading profit…but, really, isn’t this depreciation thing just another can that can be kicked down the road for a long time and never really matter (perhaps until someday it does – but, when? Just like our Federal Government deficits…).

   1. That’s not Burry. Even if he could move these names with a tweet (he can’t), he doesn’t care about making a few million bucks on a short. That’s kinda the whole point of this article. Burry’s chasing the unattainable: An encore to history’s most famous short.

      1. Excellent insight Mr. H. Burry is more interested in cementing his legacy as one the all-time best short sellers. He isn’t interested in a few million from volatility by headlines. If and when the AI bonanza fizzles, the cracks will likely appear first where we see the froth coming out quickly already: the companies with the weakest balance sheets incurring more debt, and/or a nebulous path toward sustained AI profitability. The recent price action of Oracle, Coreweave, Meta, tells that story.

5. Honestly, I’d choose to follow someone, who is all about the money; than someone who is all about the notoriety. At least, I’d know what I was following.

6. Still running the home streaming server on 10 year old plus HDD, depreciate that Burry!

7. Jim Simons. Mathematician. Hedge fund manager. Investor. Philanthropist.
   1976 Oswald Veblen Prize in Geometry. Elected to the National Academy of Sciences in 2014. Code breaker. Renaissance Technologies.

   1. yup, was surprised to not see that name in the roster, but more isn’t always better.

8. Burry is far from the first one to cry foul about AI capex and depreciation – on the contrary, this seems to be fast moving towards the consensus view.

   Also, if we are in a bubble, it is in its very embryonic stages, and likely not contained in the big end of town. They are far from cheap as a basket, and far from truly bubbly valuations too.

9. Sigh, two days late so no one will read this, but I do believe there’s a different dynamic to watch for regarding depreciation. Hyperscalers pay $40K – $55K for current GPUs. When a manufacturer (Apple) defines the “expected” lifetime of a technology (laptop) and there are over 100M customers, the manufacturer prevails in the discussion on depreciation timing, because the market is liquid. When the individual deal size for a product (GPU) is $10B+ (quantity 10K+ of chips), and there are only 10 customers in the industry (that matter), then the customer prevails in the discussion because the market is not liquid. If the hyperscalers say they are going to use those chips for 6 years (following initial buildout), then the market structure dictates that they can have their way, and subsequent tech may be relegated to a high-performance niche for some time. So the risk is out-year under-performance by NVDA, not “fraud” by the hyperscalers.