CoreWeave

Jing Zhao

The IPO market in 2025 has witnessed a selective rebound following the prolonged underperformance of the 2021 cohort. This paper examines the valuation dynamics and market reception of recent IPOs, with a focus on SaaS and technology companies, and compares them to the 2021 experience. Through case studies of firms that chose to go public, such as Figma and Coreweave, as well as companies that remained private, including Databricks and Stripe, and those that underperformed, like Chime, the study identifies key factors influencing public market outcomes. Findings indicate that investor preferences have shifted from growth-at-all-costs to sustainable, margin-aware, and AI-leveraged growth. The paper also proposes categorization of AI tickers, and introduces Revenue Weighted Metric (RWM), to better represent relationship between growth, profitability and 2025 public market valuations. The paper concludes with implications for founders and investors regarding IPO timing, strategic positioning, and a new AI valuation playbook in face of public market reactions to AI value creation in 2025.

Olayimika Oyebanji

The rapid growth and computational demands of sophisticated AI models have made Graphics Processing Units (GPUs) essential, yet ongoing shortages have turned securing sufficient compute capacity into a major operational challenge. Global AI infrastructure spending is projected to surge, potentially reaching $758 billion by 2029, with compute resources already accounting for 60–70% of large model expenses. This immense demand, coupled with supply constraints, is driving a critical market shift: enterprises and AI labs are increasingly bypassing traditional hyperscalers to secure massive, dedicated capacity agreements with specialized "neocloud" providers. While these large capacity deals, such as CoreWeave's $35 billion-plus agreement with Meta and Argentum AI's $1.74 billion in capacity agreements, illustrate organizational adaptation to current constraints, persistent challenges—including power availability, capital intensity, HBM supply limitations, and low average GPU utilization (cited as low as 5% in some deployments)—continue to shape the market. The trajectory of future AI development remains fundamentally dependent on the industry's ability to effectively resolve these foundational infrastructure and efficiency hurdles.

Wes Brown