Apple Will Run Advanced AI Model on Nvidia GPUs via Google Cloud, Ditching Private Infrastructure
Apple said its new frontier AI model, FM Cloud Pro, will run on Nvidia GPUs hosted in Google Cloud for agentic workflows, according to CNBC’s MacKenzie Sigalos citing Craig Federighi. Apple’s private M-series cloud remains for other workloads, but it reportedly couldn’t scale the agentic frontier layer. Apple reported Q2 FY26 revenue of $111.184B and EPS $2.01, and authorized a $100B buyback.
Shifts Apple’s frontier AI compute from its private M-series cloud to Nvidia-on-Google Cloud for agentic workloads, implying scaling constraints and potential margin/CapEx trade-offs.
Apple’s software chief confirmed FM Cloud Pro will run on Nvidia GPUs hosted in Google Cloud, moving frontier agentic workloads off-premise.
Near-term sentiment likely modestly positive for AI competitiveness narrative, but investors may weigh higher third-party compute costs vs. monetization upside.
Background
Apple has reportedly built an M-series-based private AI cloud for privacy-first compute control, but the frontier agentic workload reportedly couldn’t scale there.
Why it matters
The key new fact is the confirmed deployment path for FM Cloud Pro: Nvidia GPUs hosted on Google Cloud for agentic workflows and complex reasoning, while the private cloud remains for other workloads.
Market relevance
A concrete compute-architecture shift for Apple’s frontier AI has direct read-through to AI infrastructure demand (NVDA) and hyperscaler workload growth (GOOGL), with AAPL’s competitive AI positioning as the counterweight.
Market effects
Read-across that even vertically integrated players may outsource frontier/agentic compute to Nvidia + hyperscalers, supporting continued AI infrastructure spend.
No explicit regional demand signal; impact is primarily US-listed AI infrastructure and hyperscaler sentiment.
Reinforces global AI compute supply-chain dependence on Nvidia GPUs and hyperscaler capacity planning.
Alternative perspectives
Apple may be using third-party GPUs only for specific frontier/agentic workloads while keeping most training/inference on-prem, limiting incremental hyperscaler revenue impact.
The article doesn’t address unit economics (cost per token, latency, data transfer/privacy controls) or whether this is a temporary scaling bridge—those could swing the margin and capex implications for AAPL/NVDA/GOOGL.
Key entities
- companyApple
Confirmed FM Cloud Pro frontier/agentic workloads will run on Nvidia GPUs in Google Cloud, departing from its private AI infrastructure scaling approach.
- companyNvidia
Named GPU provider for Apple’s frontier model deployment, reinforcing Nvidia’s AI infrastructure demand thesis.
- companyGoogle Cloud
Named hosting environment for Apple’s frontier model on Nvidia GPUs, supporting hyperscaler backlog/capex narratives.
