OpenAI's Chip Strategy: A $100B Funding Gap and Inference Liquidity Crunch

The core financial tension is stark: Nvidia's planned $100 billion investment in OpenAI has stalled. This isn't a minor delay; it represents a massive capital shortfall that directly threatens OpenAI's multi-year data center build-out. With the deal expected to close weeks ago and now dragging on for months, OpenAI faces a liquidity crunch that forces immediate operational decisions.

To avoid bottlenecks, OpenAI is securing inference capacity through alternative partnerships. The company has already struck deals with AMD and Broadcom to deliver tens of billions in revenue, while also exploring chips from startups like Cerebras and Groq. This multi-vendor push is a direct response to a funding gap, not a strategic failure of Nvidia's technology.

The immediate consequence is a scramble for inference hardware. OpenAI's dissatisfaction centers on the speed of Nvidia's chips for specific tasks like software development, where it needs new hardware that could eventually provide about 10% of its inference computing needs. This urgency underscores that the stalled investment creates a tangible pressure to lock down alternative supply now.

The Multi-Pronged Capacity Play: Speed vs. Cost Flow

The immediate financial flow is clear: OpenAI is deploying capital across multiple partners to secure inference capacity now, even as it pays a premium for speed. The company has already struck deals with AMD and others, securing tens of billions in revenue for inference hardware. This multi-vendor push provides near-term liquidity and avoids bottlenecks while the stalled $100 billion Nvidia investment drags on.

Yet this speed comes at a cost. The Broadcom custom chip project, a key long-term cost-control play, has already slipped. OpenAI wanted more power sooner than Broadcom could deliver, pushing the chip's rollout from a planned Q2 2026 to Q3 at the earliest. This delay underscores the trade-off: locking in immediate capacity with existing partners versus waiting for a potentially cheaper, in-house solution that may not arrive on schedule.

The Nvidia partnership remains the anchor for the broader deployment plan. The first gigawatt of Vera Rubin systems is targeted for the second half of 2026, aligning with the company's goal to deploy at least 10 gigawatts. This phased approach, supported by the stalled $100 billion investment, creates a hybrid infrastructure where high-cost, high-performance alternatives are used today, while the long-term, lower-cost Nvidia build-out is scheduled for the latter half of the year.

Catalysts and Risks: The Path to Superintelligence Funding

The primary catalyst for success is the finalization of the $100 billion Nvidia investment deal. This capital is the essential fuel for the 10-gigawatt deployment plan, with the first gigawatt targeted for the second half of 2026. Without this funding, the entire multi-year build-out faces a severe liquidity crunch, forcing OpenAI to rely indefinitely on more expensive inference solutions.

The major structural risk is the continued delay of the Broadcom custom chip. The project, which was to deliver a key cost-control solution, has already slipped from a planned Q2 2026 to Q3 at the earliest. This delay prolongs OpenAI's dependence on higher-cost inference hardware from partners like AMD, undermining the financial efficiency of its hybrid infrastructure strategy.

The ultimate test is whether OpenAI can deploy superintelligence without being bottlenecked by chip supply or cost. The company is walking a tightrope: securing immediate capacity through alternative partnerships while waiting for the delayed Broadcom chip and the long-term Nvidia build-out. Success hinges on the Nvidia deal closing soon enough to fund the 10-gigawatt plan, while the Broadcom chip eventually arrives to provide a cheaper, in-house inference solution.