Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN

Hedge fund origins and risk management

Intrator started CoreWeave in 2017 while running a natural gas algorithmic hedge fund, applying institutional risk management principles to crypto mining before pivoting through CGI rendering and medical research into neural network infrastructure.

Donated GPUs as business tuition

The company purchased early A100 GPUs and donated them to open-source AI researchers to learn parallelized computing, which led to those researchers demanding the same infrastructure when they returned to industry jobs and launched CoreWeave's commercial business.

Specialized layer above silicon

CoreWeave operates exclusively in the purpose-built layer above Nvidia GPUs but below AI models, rejecting the general-purpose cloud model of AWS to focus solely on AI compute integration and software.

Debunking 16-month obsolescence claims

Intrator refutes short seller claims that GPUs become obsolete in 16 months, stating CoreWeave's average client contract is 5 years with infrastructure remaining profitable throughout its entire lifespan.

Appreciating value of older hardware

Contrary to standard depreciation, A100 GPU prices have actually appreciated through the year as new companies emerge to utilize older hardware for different model sizes and inference workloads.

Hardware cascade lifecycle

Bleeding-edge GPUs move through a predictable lifecycle beginning with model training, then experimental work, then long-term inference deployment, similar to how older iPhones find robust secondary markets in developing economies.

Nvidia's first-come-first-served allocation

Despite competition from sovereigns and tech giants willing to pay double, Nvidia allocates GPUs based strictly on order sequence rather than playing favorites, auctioning hardware, or prioritizing specific customers.

Six-year infrastructure economics

CoreWeave utilizes a 6-year depreciation schedule and expects GPUs to remain economically viable until data center power costs exceed the margin of newer hardware, at which point they can be resold or geographically repurposed.

Inference as monetization metric

The shift from research to productization is evidenced by massive inference compute demand, which Intrator views as the true monetization of AI investment where models deliver real-world value.