OpenAI's path to self-developed chips, free from NVIDIA's constraints, is facing a severe funding struggle. The "Nexus" chip project, once seen as a strong collaboration, has recently hit a deadlock due to funding conditions. As a key partner, chip giant Broadcom has proposed a rather strict condition: only if Microsoft commits to purchasing 40% of the initial production capacity, will Broadcom invest real money.

This ambitious project, dubbed "Nexus," aims to build a 10-gigawatt computing cluster by 2030, with power consumption equivalent to five Hoover Dams. The project's purpose is clear—to reduce server costs by 20% to 30% through self-developed chips and alleviate over $200 billion in operational funding pressures before 2029. However, the first phase alone, which requires 1.3 gigawatts, would cost about $18 billion, and the total chip production costs for the entire project are expected to soar to $180 billion.

Broadcom's caution is not unfounded. Although Microsoft has reserved data center space, it has not yet signed a formal purchase agreement. Broadcom believes that only with Microsoft's top-tier creditworthiness and operational experience can this astronomical investment see any hope of repayment. For Broadcom, binding Microsoft to the project is a "safety valve," but for OpenAI, this adds a layer of complex uncertainty.

Internal memos show that the partnership agreement already includes a risk contingency plan: if Microsoft fails to meet the purchase target, OpenAI must find its own buyers. More worrying is that OpenAI has had similar "hot air then cold" precedents before, such as a previously reported $100 billion data center collaboration with NVIDIA that ultimately failed to materialize.

In the current complex competitive landscape, Microsoft no longer pays revenue shares to OpenAI and has lost exclusive technology rights. Meanwhile, OpenAI has begun integrating its models into Amazon AWS to seek diversification. In this delicate "alliance game," whether Microsoft is willing to bear the huge procurement risk for OpenAI's self-developed chip dream will directly determine whether the "Nexus" project takes off or becomes another expensive paper plan.