What do you mean this is going to be bad?
NVIDIA are only overvalued by about 4 trillion.
You jest, i'm sure... because if the exponential growth of LLMs and job destruction remains on it's current trajectory Nvidia is significantly under valued.
I want to give you guys a real-world, current example of how the markets do not comprehend what is coming and thus, traditional analysts using traditional 'value ratios' in the current situation is a recipe for missing the boat entirely.
To give a fine, timely example I could suggest Nvidia but I will offer you all a far more intimate example - Coreweave. It's a recent IPO, and has ran 300%+ which has led to analysts saying things like the OP's quote. This will be a fairly long post so I will TLDR at the end. I personally believe it's very important for people to understand the shift we are about to experience.
unlike the companies in the dotcom bubble peak: Coreweave is genuinely an AI Cash Machine That's Printing Money (Not Just GPUs, to line Jensens fuckin wallet
)
The market has overestimated CoreWeave's depreciation and interest expenses, but many people seem to have a fundamental misunderstanding of CoreWeave's business model. They think it's just a small-scale AWS providing GPU instances. However, CRWV's core business logic is different from traditional cloud providers.
Traditional cloud providers usually adopt a self-built data center model, investing heavily in capital expenditures (CapEx) and gradually amortizing costs through economies of scale. Their customer base is broad with high demand elasticity, mostly using on-demand payment models, and customers are typically unwilling to sign long-term fixed contracts. To put it simply, when you launch a new GPU EC2 instance on AWS, it involves demand and capacity forecasting, procurement and supply chain management, as well as data center construction and expansion, which can take six months to a year.
In simple terms, traditional cloud providers' business involves estimating future demand based on current usage, making purchases, and hoping to sell it out. This results in long cycles and capital intensity.
In contrast, CRWV adopts a light-asset capital model centered on customer contract cash flows. Specifically, CRWV doesn't rely entirely on its own funds to build data centers. Instead, it signs 4-year Take-or-Pay contracts with a few large customers, using these guaranteed future cash flows as a credit anchor to obtain construction funds through debt financing. This ensures that every investment is backed by clear contracts, significantly reducing depreciation and financing risks.
In other words, CRWV's business is more like a real estate contractor: they prepare everything needed for GPU powering up, sign contracts with customers, quickly finance and procure the GPU supercomputers themselves, and get them online in about three months.
From the sample contract table shared by CRWV to analysts in Deutsche Bank's 6/16 research report , we can see the essence of this model:
1) A typical contract has a total CapEx of $2.86B, but with a Loan-to-Cost ratio as high as 85%, meaning CRWV only uses 15% of its own funds (about $429M in equity investment) to leverage the entire massive CapEx.
2) This 15% is mainly used for the startup phase, with the rest financed through 9% interest debt, fully repaid within 4 years. By year 5, the debt is cleared, turning into a pure cash flow mode.
3) EBITDA margin is stable at 80%, with leveraged free cash flow recovering the 15% own funds in 2-2.5 years.
4)Customer contracts provide stable cash flows to repay loans and depreciation, reaching up to $688M in later stages, with a cumulative net present value exceeding $1B.
According to CoreWeave's 2025 S-1 prospectus and Q1 financial report, the contract backlog is $29.9B. The 2025 CapEx plan is $20-23B, but most of it is covered through leverage, requiring only a small amount of company-owned or equity funds.
TLDR
In summary, the information asymmetry with CRWV lies in the fact that it is a financial engineer for AI computing power. In the context of large-scale (tens to hundreds of billions of dollars) scarcity of computing power like GB200/300 NVL72, it quickly captures the market and continues to leverage GPU generational differences. The dumbass analysts that cry about dotcom era valuations are missing the forest for the trees. We are going to see companies like Nvidia and Coreweave completely defy all traditional value metrics and rise to astonishing market caps while printing hundreds of billions of real-world dollars, not Dotcom pies in the sky..
EverydayBeast
Is this true?
