Amazon's AWS pays Synopsys over $1 billion for chip blueprints to build past Nvidia
Amazon's cloud signed a multi-year deal worth over $1 billion for Synopsys chip-design blueprints, weeks after a reported plan to sell its own AI silicon outside AWS, and days after its finance chief claimed a $25 billion AI run rate with server paybacks under three years.
Vincent Jiang · 2 min read
On September 30, Synopsys−3.35% — Synopsys, down 3.35 percent today, the chip-design software firm, and Amazon Web Services−2.33% — Amazon Web Services, down 2.33 percent today announced a multi-year deal worth more than $1 billion: AWS licenses ready-made chip blueprints, a Synopsys line that earned $1.75 billion in its most recent fiscal year and competes with Arm−7.10% — Arm, down 7.10 percent today Holdings, the traditional leader in licensed designs 12. AWS becomes the lead customer for the application-optimized part of that business, with work tailored to Trainium, the AI training chip AWS designs in-house, and Graviton, its server processor 3. Payments move to a license-plus-royalty model, so Synopsys earns more as Amazon's chip volumes rise 3.
The supplier that buys its cloud back
The contract runs both ways: Synopsys agreed to adopt AWS compute, storage and the Amazon Bedrock AI platform, which makes the design vendor a cloud customer too 1. Two weeks later, on October 12, Chip Dispatch relayed a Bloomberg report that AWS is in talks to sell Trainium, its training chip, and Inferentia, its inference chip, both built at TSMC−2.33% — TSMC, down 2.33 percent today, directly to buyers outside its own cloud 4. No US hyperscaler sells merchant AI silicon: Google has rented out its TPU only as a cloud service since 2018, and Microsoft−2.35% — Microsoft, down 2.35 percent today keeps Maia inside Azure 4. Amazon is separately working to move roughly $8 billion of Nvidia accelerators off its balance sheet 5.
The capex bill that has to pay it back
Amazon's quarterly capex has gone from $21 billion to $53 billion in two years
Data
| Capital spending | |
|---|---|
| Q3 '24 | $21.28B |
| Q4 '24 | $26.05B |
| Q1 '25 | $24.26B |
| Q2 '25 | $31.37B |
| Q3 '25 | $34.23B |
| Q4 '25 | $38.47B |
| Q1 '26 | $43.23B |
| Q2 '26 | $53.08B |
Amazon's quarterly capital spending has climbed from $21.3 billion in Q3 2024 to $53.1 billion in Q2 2026 6. AWS chief executive Matt Garman put 2026 spending near $220 billion on the a16z podcast, and said Trainium capacity is sold out a year ahead 7. AWS finance chief John Felton told Axios on October 16 that "the demand is real": an AI revenue run rate above $25 billion growing triple digits year over year, AI servers paying back in under three years, and an order backlog near $500 billion at Q2 89. Amazon chief executive Andy Jassy has said Trainium could save tens of billions in annual capital spending and add several hundred basis points of operating margin 3.
No prices, no customers, no date yet
The merchant plan carries no price list, no committed customer and no launch date 4, and Jassy insists AWS remains a major Nvidia−1.59% — Nvidia, down 1.59 percent today partner 3. Gartner−0.95% — Gartner, down 0.95 percent today, for the bear side, warns that more than $3.5 trillion of AI infrastructure spending could face impairment by 2030 8. Amazon discusses third-quarter results on October 29 10; the lines to read are capex and whatever it says about the Nvidia position 5. The tenant is buying the field.
Deepdive
AI-generated from this story and its cited sources. Not investment advice.
