How Google, Meta, Microsoft, and others balance developing specialized AI chips with their relationship with Nvidia, which has an estimated 70% of AI chip sales
In September, Amazon said it would invest up to $4 billion in Anthropic, a San Francisco start-up working on artificial intelligence.
Context & Ripple Effects
This is an early view of a durable AI-infrastructure tension: the largest cloud and platform companies need Nvidia’s supply and software ecosystem while pursuing more control over the hardware beneath their services. Amazon’s planned Anthropic investment also places capital behind a major AI developer as the compute supply chain is being defined.
Later coverage shows the strategy expanding from internal chip projects into a broader alternative-compute ecosystem: Meta’s planned use of Amazon Graviton chips for AI models and Google’s effort to rent TPUs to Anthropic both turn proprietary silicon into strategic infrastructure.
First-order effects
- Google, Meta, Microsoft and peers must run a dual-track procurement strategy: retain Nvidia capacity for near-term AI deployment while funding specialized chips that can reduce dependence in selected workloads.
- Nvidia’s biggest customers become potential hardware rivals, even as they remain among the company’s most consequential buyers.
Second-order effects
- Chip design alone is insufficient: customers also need software that can run efficiently beyond Nvidia hardware, raising the importance of efforts such as OpenAI’s Triton work across AI chips.
- Cloud providers gain an incentive to turn in-house processors into externally usable capacity, creating alternatives for AI developers rather than treating custom silicon solely as an internal cost-control tool.
Third-order effects
- If these programs scale, AI computing is likely to become more heterogeneous: Nvidia may remain central, but workload-specific chips and cloud-operated alternatives can dilute the assumption of one default hardware platform.
- The competitive boundary shifts from selling processors alone to controlling the integrated stack—chips, developer tools, cloud capacity and startup relationships—making partnerships with AI model builders strategically important.
The trend: This is one data point in the shift from Nvidia-centered AI buildouts toward heterogeneous, cloud-controlled compute stacks.