Nvidia’s $100B OpenAI Bet: 3-Year Payback and the Big Question — Who Will Broadcom & AMD Back?
Nvidia just pledged up to $100B to OpenAI — and could get every dollar back within 3–4 years. The real question now: What will Broadcom and AMD back next? Anthropic? xAI? A national AI build-out?
What’s new (and why it matters)
OpenAI is building 10 GW of AI data centers running on Nvidia gear.
Nvidia’s investment isn’t charity — it ensures priority supplier status and a huge share of OpenAI’s capex.
Jensen Huang hinted at shipping 4–5M GPUs in 2025 — roughly double last year’s shipments — showing demand is accelerating.
This creates a flywheel: Nvidia invests → OpenAI builds → OpenAI buys Nvidia hardware → Nvidia profits and reinvests.
The payback math (at a glance)
Key numbers:
Cost per GW: $50–60B
Nvidia’s share per GW: $35B
Gross margin: ~75%
Break-even point: 3.8 GW of deployments → roughly mid-2028 under a realistic ramp.
Even using a cautious ramp (1 GW in 2026, 1.5 GW in 2027, 2 GW in 2028), Nvidia appears on track to recoup its $100B investment by mid-2028.
Why the Loop Favors Nvidia
OpenAI is effectively pre-paying Nvidia for hardware by taking its capital.
Nvidia locks in priority supply and pricing power.
Every GW beyond payback drops most of the $26B+ of gross profit straight to the bottom line.
Market Map: Winners & Pressure Points
Winners: Nvidia, TSMC, HBM suppliers, hyperscalers with access to this capacity (Microsoft first in line).
Losers or pressured: Broadcom (ASIC push), AMD (needs MI400+ execution), anyone trying to compete on price-only silicon.
Now the Big Question: Who Will Broadcom & AMD Back?
Nvidia just wrote a $100B check — forcing competitors to respond. Broadcom and AMD can’t just sit on the sidelines and hope OpenAI diversifies; they need to sponsor their own builds:
Anthropic or xAI: A showcase deployment proving their silicon can scale at GPT-4.5/5 class workloads.
Neutral AI Clouds: Fund a multi-vendor compute platform for enterprises seeking Nvidia alternatives.
Microsoft’s Next Wave: Pre-finance MI-series or ASIC capacity to get guaranteed Azure deployment share.
Government-Backed Clusters: Partner with national AI initiatives (US, EU, Japan) to diversify supply chains.
Now What: Nvidia just set the bar — the next headline might read, “Anthropic Secures $20B From AMD and Broadcom to Build MI400-First Superclusters.” 😂