The short-seller who known as the Enron fraud has a pessimistic tackle 2025’s hottest IPO.
The unreal intelligence (AI) growth is in full swing, which implies heaps and many comparisons to the late-’90s dot-com growth and bust. One such investor who was round at the moment is short-seller Jim Chanos, who turned well-known by shorting Enron earlier than it crashed in 2001.
Chanos nonetheless sells quick via his funding advisory Kynikos Associates, and not too long ago took to X, previously generally known as Twitter, to make a bearish case for certainly one of this yr’s hottest IPOs, AI “neocloud” CoreWeave (CRWV -3.17%).
So, ought to CoreWeave’s shareholders be apprehensive? The bear case is value noting, however CoreWeave administration additionally has a severe argument to refute it.
What’s the helpful lifetime of a GPU?
Final week on X, Chanos questioned whether or not CoreWeave was appropriately accounting for the depreciation of its AI GPUs, that are just about fully Nvidia (NVDA 0.86%) chips. He wrote:
$CRWV’s capital employed averaged $19.2B within the 2Q. D&A plus lease amortization was $637M within the Q, or $2,548M annualized. That is 7.5 yr life, for predominantly GPU’s. Adjusted EBITDA was $3.0B, annualized. Draw your personal conclusions.
— James Chanos (@RealJimChanos) August 12, 2025
Chanos is mainly saying that CoreWeave could also be giving a “beneficiant” helpful life to its GPU chips, plus after all the ancillary infrastructure of buildings, networking, and reminiscence related to them.
As an example, if the helpful lives of its GPUs have been solely half what CoreWeave estimates at, say, 3.75 years, that may double depreciation expense. And if depreciation expense have been doubled, CoreWeave would not have made an adjusted (non-GAAP) working earnings of $200 million final quarter — it will have reasonably had a $400 million working loss.
In different phrases, if CoreWeave is overestimating the helpful lives of GPUs, its present progress could solely find yourself resulting in larger and larger losses, not earnings. It might akin to a financial institution rising its mortgage e book quite a bit, solely to have these loans go bitter a couple of years out. One other instance can be if an insurance coverage firm underestimates losses, it might present robust progress and earnings for some time, till these losses got here again to chunk it afterward.
Thus, it is maybe not stunning that Chanos really calls CoreWeave not a know-how firm however reasonably a “monetary firm” with “short-lived property that really depreciate.”
Is CoreWeave being too beneficiant with helpful lives?
There is a good case to be made that CoreWeave could, actually, be too beneficiant estimating the helpful lives of its GPUs. In any case, Nvidia not too long ago moved from introducing a brand new chip structure each two years to yearly. Meaning every year, Nvidia can have a more recent, quicker GPU structure. The extra fast introduction might actually trigger older generations to grow to be out of date extra rapidly.
Moreover, CoreWeave admits in its SEC filings that overestimating the helpful lifetime of its chips is an enormous threat to its enterprise mannequin. Within the “dangers” part of its current quarterly report, CoreWeave says:
We regularly work to improve and improve our platform, options, and companies in response to buyer demand and to maintain up with technological modifications. A part of this course of entails biking out outdated elements of our infrastructure and changing them with the newest know-how obtainable. This requires us to make sure estimates with respect to the helpful lifetime of the elements of our infrastructure and to maximise the worth of the elements of our infrastructure, together with our GPUs, to the fullest extent attainable. We can’t assure that our estimates will likely be correct or that our makes an attempt at maximizing worth will likely be profitable. Any modifications to the numerous assumptions underlying our estimates or to the estimates of our elements’ helpful lives, or any incapacity to redeploy elements of our current infrastructure to increase previous their contracted life might considerably have an effect on our enterprise, working outcomes, monetary situation, and prospects.
Again earlier than its March IPO, CoreWeave acknowledged in its S-1 registration assertion it believes the helpful lifetime of its infrastructure exceeded that of its typical contract, which vary from two to 5 years. CoreWeave additionally acknowledged it has structured its contracts with a median typical “payback interval” of two.5 years. Meaning if CoreWeave spends $1 billion on an information heart, CoreWeave believes it would earn $1 billion in EBITDA from that knowledge heart over two and a half years.
Thus, profitability past that 2.5 yr payback interval is determined by the “helpful life” of the infrastructure. As an example, if the helpful lifetime of the infrastructure is 2.5 years as an alternative of seven.5 years, then CoreWeave does not make any cash, primarily. And since CoreWeave makes use of numerous debt to fund that infrastructure buildout, it has extra curiosity prices to bear as properly on high of that.
Picture supply: Getty Pictures.
However CoreWeave has a very good protection — for now, not less than
In its S-1, CoreWeave maintained that it has efficiently rerented older Nvidia A100 GPUs in new contracts with both the identical or new counterparties after their preliminary contracts have been up. Administration additionally famous that older GPUs could also be enticing for inference-related workloads or lower-intensity coaching for smaller fashions sooner or later.
On its current second-quarter convention name, administration continued to assert it has been capable of proceed renting A100s and H100s for inference and different lower-intensity jobs, saying particularly that older chips falling off-contract “are nonetheless being recontracted out.”
The quick case does not seem like at hand… not less than not but
Whereas coaching the newest and biggest massive AI fashions will usually require the newest and biggest chips, plainly inference and lower-cost coaching workloads supply CoreWeave loads of alternative to recontract its older chips at this level. Subsequently, it does not appear that Chanos’ principle holds up — not less than not but.
Nonetheless, if Nvidia’s extra fast tempo of chip introduction means older chips exit of favor extra rapidly, it might doubtlessly be an issue. And naturally if there have been to be extra competitors and oversupply throughout clouds and neoclouds, perhaps with competitors within the inference market, that would put pricing strain on renewal contracts, which might ultimately grow to be an issue.
So CoreWeave’s close to future seems vibrant and intact amid AI hypergrowth; nevertheless buyers needs to be conscious the corporate must keep or discover new use instances for older chips because it grows, and might want to proceed to do that in perpetuity.
