We scan new podcasts and send you the top 5 insights daily.
Evan Conrad says SF Compute performs cloud-like work—data-center site visits, bill-of-materials design, cluster architecture, rack rollout, and backplane management of PDUs, switches, DPUs and BMCs—but refuses to own hardware. Ownership would let it manipulate the forwards market, so it stays independent like Switzerland.
Evan Conrad says SF Compute shipped the ability for customers to subdivide their InfiniBand fabric without asking the operator. That lets platform-as-a-service companies redraw cluster boundaries on the fly and scale up or down into the market. He calls it virtualizing the data center and making it programmable.
The decision to own GPUs extends far beyond the purchase price. It requires navigating a complex supply chain with international vendors, securing high-value insurance for transit, and investing in specialized infrastructure like liquid cooling systems, which most data centers lack.
Instead of building data centers, Modal runs a software layer across 17 cloud and bare-metal providers. This allows them to focus on software innovation and build a reliability layer that can leverage less-reliable but available 'neo-cloud' capacity.
Investing tens of billions into proprietary compute clusters carries less downside risk when an organization owns the physical hardware. If internal consumer models underperform, valuable data center capacity can be redeployed internally or leased to compute-hungry third parties. Owning depreciable infrastructure also provides structural cost advantages over renting cloud compute.
Evan Conrad says two VMs from one operator can differ by 40% in kernel-launch performance, and storage, WAN transit, preemption and chip generation also change value. Rather than wait for standardized contracts, SF Compute lets buyers and sellers attach multiple properties to orders, accepting wider spreads and arbitrage for flexible buyers.
SF Compute argues that for a true compute market to exist, it needs an independent operator that handles physical settlement. This means running the data centers to standardize quality and earn Wall Street's trust, much like an electrical grid operator.
Oracle is mitigating the immense capital expenditure of its AI cloud buildout by allowing customers to provide their own hardware. This 'BYOH' model, while still a small part of its business, reassures investors by allowing Oracle to expand capacity without footing the entire bill for expensive GPUs.
HydroHost's strategy is built on the thesis that data centers are moving beyond being mere cost centers for public clouds. It provides software for them to become "Neo Clouds," serving AI companies directly. This model gives data centers more control and upside, mimicking how crypto miners bypassed clouds for better hardware access.
Engineers are building home data centers with open-weight models to escape cloud token costs and vendor lock-in. Running compute locally provides complete control over intellectual property and mitigates fears that cloud providers might train models on proprietary data.
Instead of building a vertically integrated cloud, AMP acts as a neutral "Independent System Operator" (ISO) for compute. This model, borrowed from the power grid, focuses on pooling supply and demand across multiple clouds and silicon providers without owning the assets, aiming to make "flops flow like megawatts."