Our Take
The OpenAI relationship creates unusual conflicts where the customer is also creditor, potential shareholder, and has executives with personal stakes in the vendor.
Why it matters
This structure previews how AI supply chains will blur traditional vendor-customer boundaries as compute demand outstrips supply. Banking interest at $10B suggests strong institutional appetite for AI infrastructure plays.
Do this week
Procurement teams: audit your AI hardware vendor relationships for similar customer-creditor conflicts before budget planning so you can avoid dependency traps.
Cerebras files for $3.5B IPO with OpenAI as customer and creditor
Cerebras Systems filed to sell 28 million shares at $115-125 per share, targeting a $26.6 billion valuation (per SEC filing). The AI chipmaker makes the Wafer-Scale Engine 3, positioned as a faster, lower-power alternative to GPUs for inference workloads.
OpenAI provided Cerebras a $1 billion loan in December, secured by warrants for over 33 million shares. The companies also signed a multi-year agreement worth more than $10 billion (company-reported). OpenAI executives including Sam Altman, Greg Brockman, and Ilya Sutskever invested as angels, though their stakes fall below disclosure thresholds.
Banks report $10 billion in orders for the $3.5 billion offering (per Bloomberg), suggesting the company will price above the announced range. Major institutional backers include Benchmark, Eclipse, Fidelity, and Foundation Capital with 5%+ stakes each.
Unusual vendor-customer-creditor triangle emerges
The OpenAI relationship creates multiple dependency layers: customer, creditor, and potential major shareholder through warrant conversion. Legal filings from Elon Musk's lawsuit revealed OpenAI once considered acquiring Cerebras entirely, with Musk's attorneys claiming he was unaware of OpenAI executives' personal investments.
Cerebras delayed its 2024 IPO attempt due to federal review of investments from Abu Dhabi's G42, which remains a major customer. The company raised $1.1 billion in September 2024 at an $8.1 billion valuation, then $1 billion more in February 2025 at $23 billion before filing at $26.6 billion.
Supply chain conflicts multiply in AI infrastructure
The customer-as-creditor structure signals how AI compute scarcity forces non-traditional financing arrangements. Organizations building on specialized AI hardware should map these relationships to avoid vendor lock-in scenarios where switching costs include loan obligations.
For procurement teams, the Cerebras-OpenAI structure shows how personal investments by executives can create undisclosed conflicts in vendor selection. Standard conflict-of-interest reviews may miss angel investments below disclosure thresholds.