open source AI

Google told Meta it cannot provide all the Gemini computing capacity Meta wants. Google Cloud's backlog nearly doubled, GPU costs surged 150%, and 7 GW of planned U.S. data center capacity has been delayed. The AI compute crunch is no longer a hyperscaler problem — it's an enterprise architecture problem that demands immediate diversification.

DeepSeek V4 ships a 1.6T-parameter open-weight model with 1M-token context at $1.74/M tokens and Huawei chip optimization. What enterprise AI leaders do next.