TLDR
- OpenAI secured 16GW of compute from AMD and Broadcom through 2029.
- Nvidia holds a 7% stake in CoreWeave, which has $22B in OpenAI deals.
- U.S. data centers may use 14% of national power by 2030, says McKinsey.
- OpenAI’s contracts include equity warrants tied to AMD performance.
A new wave of tech-driven finance is unfolding as OpenAI commits to a $300 billion hardware expansion, forming a closed-loop system of suppliers, financiers, and energy partners. With chip deliveries tied to equity gains and infrastructure backed by long-term contracts, the firm is shaping a new financial model. This structure is raising questions across capital markets about sustainability, risk, and the future of AI-linked finance.
Vendor Ties Connect Chip Supply to Financial Returns
OpenAI has signed long-term chip supply deals with AMD and Broadcom to support its planned infrastructure buildout from 2026 through 2029. AMD is set to deliver six gigawatts of GPU compute using its Instinct line, while Broadcom will co-design and build ten gigawatts of custom accelerators and systems.
Both contracts contain financial incentives beyond product delivery. AMD granted OpenAI equity warrants linked to performance milestones, allowing OpenAI to benefit if AMD’s share price rises. This ties OpenAI’s expansion to the financial health of its supplier. Meanwhile, Broadcom’s deal includes shared hardware design and integration over several years, strengthening the operational link between the two companies.
These arrangements support OpenAI’s broader compute growth strategy. The hardware will underpin the company’s five new U.S.-based Stargate sites, developed in partnership with Oracle and SoftBank. Combined, these projects form one of the largest private infrastructure efforts in the tech sector.
Financing Loops Blur the Line Between Supplier and Customer
OpenAI’s approach relies on vendor-financing and equity-linked contracts that connect financial outcomes across companies in its supply chain. Earlier this year, Nvidia disclosed a 7 percent stake in CoreWeave. That same infrastructure provider expanded its agreements with OpenAI by $6.5 billion, bringing its total contract value to over $22 billion.
These cross-holdings form a feedback loop. Nvidia, as a chip vendor, benefits from CoreWeave’s success, which in turn depends on OpenAI’s demand. Bloomberg reported that Nvidia’s financing may reach $100 billion, linking supply to demand in a single funding structure. If OpenAI slows purchases, it affects both vendor returns and infrastructure usage.
Bank of America’s October 2025 survey found that 54 percent of fund managers view the current AI boom as a bubble. Cash balances remain near 3.8 percent, which could drive volatility if capacity outpaces demand in the short term. The S&P 500’s exposure to major AI stocks also raises risks for passive investors.
Energy Contracts Shape Expansion Timeline
Electricity is central to OpenAI’s infrastructure plan. The announced compute would require up to 16 gigawatts of power. This matches or exceeds the energy consumption of several smaller countries.
Goldman Sachs expects global data center power use to grow 165 percent by 2030 compared to 2023 levels. In the U.S., McKinsey estimates compound growth of 25 percent per year, pushing data centers toward 14 percent of national electricity demand by the end of the decade.
These trends are forcing operators to secure long-term power contracts and consider new locations where grid capacity is higher. If permitting and interconnection do not keep pace with hardware deployment, delays could create delivery mismatches and strain cash flows.
Timeline to 2029 Defines Financial and Technical Risk
OpenAI’s hardware and energy commitments align with a delivery timeline from late 2026 through the end of 2029. Broadcom’s rollout of custom silicon is expected to finish by 2029, with AMD’s GPU deliveries starting in the second half of 2026.
Custom chip performance will play a key role in the economics of OpenAI’s model. Better energy efficiency and cost per unit of compute could help make the infrastructure self-funding. Broadcom’s design includes accelerators, networking, and rack systems, though performance will depend on memory and packaging.
Each vendor’s execution will determine how quickly OpenAI can convert capacity into revenue. Usage-backed enterprise contracts and energy agreements will also be key. These metrics will help measure whether the model shifts from a feedback loop to a sustainable system.
