From: allin
CoreWeave is identified as a new type of infrastructure provider known as “neocloud” [00:53:42]. This term refers to companies that use GPUs to build data centers [00:53:50].
Company Profile and Business Model
CoreWeave got an early start on acquiring GPUs, securing a significant number of NVIDIA GPUs [00:53:55]. The company operates 32 data centers, housing 250,000 NVIDIA GPUs [00:54:04]. To put this scale in perspective, Elon Musk’s largest data center, Colossus, has 100,000 GPUs [00:54:11].
The founders of CoreWeave were originally commodities traders, which influenced their strategy to lock down the full supply of data centers, power, and chips [00:57:39]. Their technical decision to not use hypervisors, allowing users to write directly to the bare metal, provided a “very native approach” that gave them significant traction and enabled them to compete with larger cloud providers like AWS, GCP, and Azure [00:55:27]. This approach allows for very low latency throughput, beneficial for efficient transaction processing [00:58:08].
Financials and IPO
CoreWeave is anticipating an IPO, with analysts estimating they will raise at least 30 billion [00:54:21]. A secondary market valuation in November 2024 (this date seems to refer to the future from the recording date) valued the company at $23 billion [00:54:29].
The company has shown incredible revenue growth, reaching 1 billion in losses in 2024 [00:54:48]. A significant portion of these losses are attributed to interest payments on their substantial debt load, which is nearly $8 billion, incurred to acquire the GPUs [00:54:55].
Risks and Challenges
GPU Useful Life and Amortization
A major question for CoreWeave’s business model is the period of amortization and the useful life of the NVIDIA GPUs [00:56:11]. The useful lifespan for GPUs is typically considered three to five years before a new generation becomes significantly more powerful and energy-efficient [00:57:06]. This contrasts with older servers using CPUs for tasks like running Facebook, which can last five to seven years [00:57:19]. If CoreWeave’s calculation for the useful life of these GPUs is incorrect, the business could be “deeply underwater” [00:56:36].
Single Client Dependency
Another significant risk is CoreWeave’s high dependency on a single client, Microsoft, which accounts for 60% of their revenue [00:58:20]. This relationship with Microsoft may be tied to servicing the OpenAI deal or for Azure, though the exact purpose remains unclear [00:58:28].
Transitory Business Model
There is a concern that CoreWeave’s business model could be transitory, similar to “speed doublers” in the early internet era that provided an arbitrage between dial-up and broadband internet [00:59:32]. As major players engage in massive capital expenditure to create their own infrastructure, these “neocloud” services could be effectively replaced or bundled into existing cloud offerings like GCP or AWS [01:00:36].