Interview, Fireside Chat, Conference Presentation
Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN
- CoreWeave projects GPU hardware depreciation over six years with value retention for five-to-six-year contracts, asserting that obsolescence is driven by the repurposing of power and real estate for higher-margin use cases rather than technical failure, and anticipates appreciation in prices for older generation hardware like A100s as new companies emerge to utilize them.
- The company expects relentless demand for AI infrastructure that currently overwhelms global capacity, creating a "disequilibrium" where "time to compute" and constraints on electricity, memory, storage, networking, and optics are more critical than chip availability alone.
- CoreWeave forecasts that supply bottlenecks, including a delayed investment cycle in fab capacity for memory, will eventually trigger a boom-and-bust cycle in capital-intensive industries, which will eliminate weaker competitors and allow strong firms to acquire excess infrastructure at lower prices.
- Financing conditions are expected to improve as lenders gain confidence, with the cost of capital having dropped 600 basis points in two years and projected to reach hyperscaler levels, enabling the company to pay off principal and interest on five-year deals within two and a half years.
- Significant growth constraints are identified as the scarcity of electricity, construction labor, and grid connections, leading to plans to build data centers in remote areas with excess renewable or nuclear energy (such as West Texas), where latency is mitigated by existing underground fiber infrastructure.
- The outlook anticipates the creation of a high-salary "Generation Tool Belt" workforce earning $150,000 to $300,000 to address the shortage of skilled trade workers needed for data center construction, alongside a long-term strategic pivot toward nuclear energy and small modular reactors despite decade-long commissioning timelines.
- CoreWeave expects the industry to adhere to a first-in, first-out allocation model for GPUs, while predicting that custom silicon from hyperscalers will emerge but will not currently challenge NVIDIA's massive ecosystem advantage.
- Perplexity envisions a future computing paradigm where AI serves as the operating system, utilizing a "Personal Computer" hybrid architecture that synchronizes local hardware for private data with server-side execution for complex tasks, aiming to make $10,000 desktop workstations with local AI a consumer commodity.
- Enterprise adoption of Perplexity's "Computer" platform is projected to accelerate faster than consumer usage, with the Enterprise Max tier expected to generate significant cost savings and positive gross margins by replacing manual workflows with autonomous AI agents performing long-duration orchestration tasks.
- The company plans to release features such as "Model Council" for multi-model comparison, integration with third-party platforms like Reddit and LinkedIn via official APIs, and tools enabling non-engineers to deploy code via natural language prompts in messaging interfaces.
- Mistral AI expects to develop frontier open-source models in collaboration with NVIDIA for specialized verticals in finance, engineering, and science, arguing that enterprise AI will rely on localized, verticalized models rather than a single global model to leverage unique intellectual property.
- Mistral AI plans to deploy portable platforms for training and data processing directly on customer infrastructure to ensure data security, maintaining that human expert signals remain essential for training even as synthetic data usage grows for initial model warm-up.
- The outlook from Mistral AI includes a shift toward AI agents requiring a fundamental redesign of IT systems, including the implementation of "context engines" and "deterministic gates" to manage observability, safety, and data access rights without human intervention.
- Perplexity anticipates that the improvement in AI coding tools will exponentially increase the speed of product shipping and enable solo entrepreneurs to build billion-dollar companies, resulting in job displacement but ultimately fostering a future of high-quality work-life balance and individual agency.