Capital & Industrial Strategy

Anthropic's revenue run rate crossed $30 billion — more than tripling since the end of 2025 — and the company confirmed a long-term agreement with Google and Broadcom to supply custom TPU chips and computing capacity worth hundreds of billions of dollars, according to reports from Bloomberg, the Financial Times, The Wall Street Journal, and CNBC. Broadcom will design future generations of Google's Tensor Processing Units, while Anthropic gains assured access to the resulting infrastructure. Reuters confirmed the multi-year chip development agreement between Broadcom and Google.

The structure of the deal — vertical integration from chip design through infrastructure provisioning to end-customer access — is notable. Anthropic is effectively pre-committing capacity years in advance, reducing its dependence on spot market availability and locking out competitors from the same supply. Google, meanwhile, deepens its role as both infrastructure provider and indirect investor (having previously put billions into Anthropic). Broadcom gains a strategic foothold in custom AI silicon, moving beyond its traditional networking business.

Samsung Electronics reported an eight-fold year-over-year profit increase, beating analyst estimates, driven by robust AI memory chip sales despite markets disrupted by Middle East conflict, according to Bloomberg. Separately, data center builder Firmus Technologies raised $505 million in a round led by Coatue Management, with backing from Nvidia, as part of a global push to finance AI infrastructure, per another Bloomberg report.

The Samsung result is significant because memory — HBM in particular — is a supply-constrained input for AI accelerators. Strong sales indicate that hyperscalers and AI labs are still ordering aggressively, undeterred by macro uncertainty. The Firmus round, meanwhile, shows that infrastructure builders can still attract growth-stage capital even as broader venture markets tighten. Nvidia's participation signals strategic interest in ensuring data center capacity keeps pace with its chip shipments.

Nvidia's acquisition of SchedMD, the company behind the widely used Slurm workload manager for high-performance computing and AI clusters, has sparked concern among AI specialists about potential restrictions on software access, according to Reuters. Slurm is critical infrastructure for scheduling jobs across GPU clusters, and many academic and enterprise users fear Nvidia may use control over the software to steer customers toward its hardware or limit interoperability with competing chips.

This follows a pattern of vertical integration by compute providers. Nvidia already dominates GPU hardware and has built a software moat with CUDA. Acquiring the job scheduler layer gives it influence over how workloads are allocated and optimised — potentially making it harder for AMD, Intel, or custom silicon providers to compete on equal footing in multi-vendor clusters.

OpenAI, Anthropic, and Google have begun working together to prevent Chinese competitors from extracting results from their models to gain an edge in the global AI race, according to Bloomberg. The collaboration marks the first time these rivals have coordinated on defensive measures against intellectual property theft. Chinese labs have been using techniques such as distillation — querying a frontier model repeatedly to train a smaller, cheaper copy — to replicate capabilities without incurring the full training cost.

The move suggests that US labs now view IP leakage as a greater threat than domestic competition. It also indicates that API-based access, once seen as a revenue stream, is increasingly treated as a security surface. The coordination likely involves shared rate-limiting strategies, query fingerprinting, and potentially coordinated refusals to serve certain customers or regions.