From: allin
CoreWeave, a “neocloud” company specializing in cloud computing, has recently undergone a significant initial public offering (IPO) [09:40:50]. This event highlights evolving dynamics in the cloud computing market, particularly concerning the demand for high-performance GPUs and the competitive landscape of cloud providers [03:21:00].
CoreWeave IPO Details
CoreWeave aimed to raise 2 billion [10:02:00]. The company achieved a $23 billion valuation [10:06:00]. Key financial details include:
- Revenue: $2 billion last year, representing an 8x increase [10:10:59].
- Debt: Approximately $8 billion [10:14:49].
- Customers: Microsoft accounts for over 60% of CoreWeave’s revenue, while Nvidia contributes 15% [10:17:58]. Nvidia also holds a 6% stake in CoreWeave [10:26:00].
Market Perception and Differentiators
Despite some negativity on platforms like X (formerly Twitter), where investors view CoreWeave as an undifferentiated, commodity business with high debt and concentrated customers [12:49:00], there’s a strong counter-argument regarding its operational complexity. Running large GPU training clusters is “really hard” due to challenges like synchronizing thousands of GPUs, preventing melting, and managing cable disconnections [13:42:00]. Experts suggest CoreWeave manages these “as well as anyone” [14:45:00].
The company’s ability to operate efficiently in this complex environment is considered a key differentiator, drawing parallels to the early skepticism faced by AWS, which was initially perceived as a “terrible business” [14:14:00]. CoreWeave further differentiates itself through acquisitions, such as that of Weights & Biases, which aims to “decommodify” their offerings [15:51:00].
Ultimately, CoreWeave is seen as a “good company” and a “great business” that brings valuable competition to the existing cloud providers [15:13:00].
Cloud Computing Market Dynamics and Nvidia’s Role
The cloud computing market was historically dominated by a “stable three-player oligopoly” consisting of Amazon, Google, and Microsoft [03:21:00]. Nvidia strategically invested in “neoclouds” like CoreWeave to diversify its customer base [02:55:00]. This move reduced Nvidia’s reliance on the major hyperscalers, who often sought custom versions of Nvidia’s servers, thereby giving Nvidia more leverage [03:28:00]. CoreWeave, by adopting Nvidia’s standard reference design, could deploy GPUs to the market much faster [03:49:00].
Demand for GPUs remains exceptionally high. The market is currently “not overbuilt,” with China actively acquiring every available GPU, leading to warnings of shortages and daily increases in memory prices [11:08:00]. OpenAI has even gated its new image generation service due to insufficient GPU availability [11:36:00]. Nvidia’s new Blackwell GPU generation is anticipated to be “very successful” [11:44:00]. This high demand creates a “prisoners dilemma” where companies like Meta and Google are compelled to invest in Blackwell to avoid falling behind competitors in AI capabilities [11:47:00].
Nvidia’s Financials and Product Transition
Nvidia has experienced a significant product transition with its Hopper to Blackwell GPUs, an event described as the “biggest product transition in history” for semiconductors, comparable only to the iPhone’s evolution [04:32:00]. This transition is cited as a primary reason for a substantial increase in Nvidia’s accounts receivable, jumping from 5.5 billion year-over-year [04:18:00]. While an increase in accounts receivable is generally not ideal, it is deemed understandable given the complexity of the Blackwell rollout [05:22:00].
The Blackwell server is significantly more complex than its predecessor, Hopper:
- Weight: 3,000 lbs (three times Hopper’s 1,000 lbs) [05:50:00].
- Power Consumption: 120 KW (twice Hopper’s 60 KW) [06:00:00].
- Cooling: Liquid-cooled, unlike Hopper’s air-cooling [06:09:00].
This complexity means customers face significant infrastructure upgrades (like new generators and cooling systems) to implement Blackwell, making the deployment process more involved than a simple upgrade [06:22:00]. The delay in payment (accounts receivable) is attributed to customers awaiting full installation and operational readiness of these new, highly complex systems [07:44:00]. This strategy of diversifying its buyer base and promoting its reference architecture mirrors Intel’s “Intel Inside” program from the 80s and 90s [08:10:00].