The stock of neocloud CoreWeave has taken a beating recently, but volatility is nothing particularly new for the company, which went public in March of last year. On 40% of its trading days, it’s moved by more than 5%. From the Oct. 6 close through Tuesday, the stock declined by 32%, including two days in which the shares fell 15% and 12%. Shares have traded as high as $187 and as low as $34.
It has been the front-line battleground stock of the AI trade since day one.
CoreWeave invented the neocloud model, which is more of a rapid growth strategy for the artificial-intelligence cloud business than anything else. CoreWeave rents out Nvidia GPU servers over the internet in the same way Amazon Web Services does. AWS invented the modern cloud and rents out many different types of servers in addition to Nvidia GPUs, but its accelerated growth in the past year is largely because of AI demand. Wall Street analysts expect AWS sales to grow by 35% this year to $174 billion.
Meanwhile, CoreWeave’s growth is all about AI demand. In 2022 it had only $16 million in sales, but it was in the right place at the right time when ChatGPT made its debut in November of that year. CoreWeave began a very close relationship with Nvidia, which now holds 9% of its shares, is its customer of last resort, and also its most important supplier. CoreWeave has typically been one of the first clouds to make new Nvidia gear available to customers, including the latest Vera Rubin servers launched in September.
In an environment of undersupply—which has existed for almost four years now—this puts CoreWeave in an enviable position: “If you need GPUs (and you do), we can get ’em.”
The neocloud growth model begins with simple availability. CoreWeave uses that leverage to get customers like Microsoft, Meta Platforms, and OpenAI into long-term take-or-pay contracts that protect CoreWeave’s downside. In mid-August when the company reported its second quarter, it had nearly $130 billion in its backlog.
CoreWeave takes those contracts and uses them to give lenders confidence to finance the many loan obligations it has. When the company went public in March of last year, its debt was fetching double-digit interest rates, but those have since come down. In an $8.5 billion March loan tied to the neocloud’s Meta contract, the rate was around only 6%—investment grade territory—but subsequent loans have been at much higher rates, including unsecured 2032 senior notes at 9.6% in June.
Because of the loss of confidence in the sector, those senior notes are trading at 79 cents on the dollar, and have a yield to maturity of 15%.
The proceeds of that debt, along with convertible debt and equity sales, buys the Nvidia equipment that goes into the data center buildings which the company leases from Core Scientific and others. With one notable exception about a year ago, CoreWeave has been able to get these new data centers launched on time, and that has fueled its triple-digit sales growth. This year analysts expect revenue to expand by 151% to $13 billion followed by another doubling next year.
Despite its mounting depreciation expenses from all the Capex, CoreWeave is able to eke out a small adjusted operating profit, but that gets demolished by interest expenses. The company’s key expense lever is bringing down its cost of capital—which it has—but rising long-term interest rates and falling confidence in data center debt may reverse that.
It’s the rapid cycle of debt, capital expenditures and sales growth that makes this train run. But a locomotive barreling down the tracks can be a disaster if it derails, which is why CoreWeave stock has become so contentious since its IPO. If you are bullish on AI, you can look at the triple-digit sales growth and the close relationship with Nvidia. Bears only see the mounting debt, the Capex, and little hope for free cash flow anytime soon.
Whenever AI sentiment shifts, like it has recently, CoreWeave stock is on the front-line because it’s a highly leveraged bet on AI. It hasn’t helped that company insiders have sold tens of millions of shares since CoreWeave’s lockup expired about 14 months ago.
In the end, like the entire AI trade, it comes down to whether AI computing demand growth continues to outpace supply, and 2027 is shaping up as the real test. AI models got much better in 2026, and risk-averse enterprises are beginning to implement AI into their business processes. Demand exploded this year and that has expected to continue into next year.
But so far, it has been hard for these companies to find a measurable return on investment, and if that persists into next year, it could stall demand growth. At the same time, 2027 will be a year when supply explodes. Research firm SemiAnalysis estimates that a total of up to 51 gigawatts of new data centers may come online globally next year, which would be equivalent to the new capacity of the last three years combined.
Unless enterprises can realize returns on their AI investments in 2027, it could finally be the moment where supply outpaces demand and the bubble bursts.
Comments