While public attention is fixed on AI models from companies like OpenAI and Anthropic, I believe a critical new infrastructure layer is quietly taking shape. A recent announcement from Akamai Technologies caught my eye. On March 3, 2026, they revealed plans to deploy thousands of NVIDIA's Blackwell GPUs to build a distributed AI inference platform, signaling a significant strategic pivot.
Akamai is not a typical AI company. Founded in 1998 out of MIT, it pioneered the Content Delivery Network (CDN), distributing content from servers close to users instead of from centralized data centers. Now, it's applying that same architectural logic to AI. Instead of competing with hyperscalers like AWS or Google Cloud on large-scale model training, Akamai is leveraging its vast global network—hundreds of thousands of servers across more than 130 countries—for distributed AI inference.
This edge-first approach is optimized for a different set of challenges than traditional cloud platforms. The key advantages are:
- Processing AI tasks closer to end users
- Achieving lower latency for real-time applications
- Reducing expensive data egress costs
- Enabling localized model fine-tuning to comply with data regulations
This strategy seems tailor-made for the rise of agentic AI. As systems interact more with the physical world—from surgical robotics and transportation to smart power grids—the need for low-latency, local processing becomes paramount. Akamai’s move isn’t about competing directly with the hyperscalers, but rather building a complementary infrastructure layer for a future where AI is deeply and physically integrated into our environment.