Nvidia is partnering with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to stand up independent compute financing platforms aimed at mobilizing more than $500B in third-party capital for AI infrastructure — funding frontier labs, enterprises, and AI clouds without any of it landing on Nvidia's own balance sheet. The pitch, in Jensen Huang's words, is that Nvidia compute is "broadly adopted, flexible across models and workloads, fungible and transferable across customers and operators" — the same collateral quality that lets institutional lenders underwrite an airplane or a power plant, now being extended to GPUs at systemic scale. It's the standing infrastructure for the kind of one-off anchor-tenant deals surfacing lab by lab this week (more below), built so banks and asset managers don't have to reinvent the underwriting model each time. (Nvidia)
Private Companies
Anthropicpartnership
Anthropic, Macquarie Asset Management, and GIC are launching Theseus Infrastructure, a platform that develops data centers purpose-built for Anthropic and leases them back under long-term agreements, with Macquarie and GIC funding the majority of each project's equity while Anthropic serves as sole anchor tenant. It's a single-customer, lease-back variant of the equity-and-debt stack Firmus used for its own buildout last week — investors take the site and construction risk, Anthropic locks in compute without owning the real estate. Anthropic also pledged to cover any local electricity-price increases the facilities cause. (Bloomberg)
Firebirdlaunch
Firebird switched on the CIS region's largest AI factory in Hrazdan, Armenia — an Nvidia DSX facility scaling toward 70,000-plus Blackwell and Vera Rubin GPUs — and Nvidia said it intends to invest directly in the company, following an earlier stake from CoreWeave. The buildout extends to a second market, with 125 megawatts secured in Kazakhstan as part of a roughly 2-gigawatt roadmap targeted for 2028. The bet is geographic: chase power in frontier markets Western developers have mostly skipped, with Nvidia's stake as the signal the GPUs actually arrive. (NVIDIA)
Public Markets
Intel priced an upsized $20B stock sale at $95/share Monday night — its first follow-on offering since 1971 — earmarking $19.7B in net proceeds for capex, advanced packaging, and external foundry capacity. Shares fell as investors weighed dilution against unproven foundry traction. (CNBC)
Emerging
Memory economics: A new paper, Memory Scarcity, Open Models, and the Restructuring of the AI Industry, models how memory-chip price surges, increasingly capable open-weight models, and hyperscalers reselling their own compute will reshape industry structure through 2030 — and its core finding cuts against the infrastructure-bull case: the cost gap between incumbents and new entrants never closes, because depreciation schedules favor whoever bought hardware first. The buildout the industry has already committed to only pencils out if token demand keeps compounding near 2x annually while pricing power holds, a narrower path than most public buildout narratives assume. (arXiv)
Custom silicon: Microsoft is in talks with TSMC to secure capacity for more than 300,000 next-generation Maia 300 accelerators for 2027 delivery, with an eventual target above 1 million units, as it works to shift a larger share of Azure's AI inference onto chips it controls rather than externally purchased GPUs. A public unveiling is planned for as soon as September, and an order at this scale would make Microsoft a foundry customer large enough to matter to TSMC's advanced-packaging roadmap independent of Nvidia's own volumes. (Bloomberg)
❝
Join us at the NYC Tech Summit on September 16. Learn from the founders of Etched, Cloudflare, and Runway. Connect with the leaders shaping what's next in tech. Applications now open at primarysummit.vc.