newsfilter
← Interviews
Semafor30 min

‘We Have To Do It’: Amazon CEO Andy Jassy On Racing China In The AI Buildout

Andy Jassy, Kate Rooney, Ben Smith

Conference Presentation, Fireside Chat, Interview
  • Capital Expenditure and Growth Projections

    • Amazon is deploying $220 billion in CapEx, with significant capacity reserved for 2027, 2028, and 2029.
    • AI revenue growth has reached over $25 billion annually in the first three years of the generative wave, approximately 440 times faster than AWS's early years (which started at a $57 million run rate).
    • AWS revenue run rate is projected to exceed $1 trillion within the next five to ten years, driven by a 13x expansion from the previous "few hundred billion" expectation.
    • Over 80% of Fortune 100 companies are currently utilizing Bedrock, Amazon's inference service.
    • The AWS backlog stands at approximately $580 billion, with no single customer representing more than a quarter of the total.
    • Amazon expects to shift from near-term cash flow headwinds caused by infrastructure build-out to strong long-term free cash flow and returns once facilities are operational.
  • Market Dynamics and "Bubble" Concerns

    • Andy Jassy rejects the "bubble" characterization, citing diverse demand across AI labs and enterprises focused on cost avoidance and productivity.
    • He anticipates the market will mature from a "barbell" shape (labs vs. cost-saving enterprises) to widespread inference adoption within current production applications.
    • AWS employs a staggered capital deployment strategy: data center construction requires a two-year lead time, while server and networking gear purchases are made with only months of visibility, allowing for demand-based adjustments.
    • Jassy warns that non-diversified data center operators may struggle to build sustainable businesses compared to full-stack cloud providers.
  • Infrastructure and Geopolitics

    • Jassy compares the required AI infrastructure build-out to the U.S. highway system of the 1950s, arguing that grid upgrades and data center expansion are unavoidable national imperatives.
    • He frames the potential U.S. cessation of data center construction as a historic error that could allow China to set global technology norms.
    • Amazon emphasizes community partnership to mitigate local backlash, citing a specific example in St. Joseph County, Indiana, where tax contributions are projected to rise from $1 million to over $3 billion annually.
    • Jassy acknowledges export controls are being strictly followed and that the company will abide by regulations in all operating countries.
  • Strategic Partnerships and "Coopetition"

    • NVIDIA: Described as a long-term partner despite Amazon's internal custom silicon program; Amazon uses both NVIDIA and its own chips (Tranium and Graviton).
    • Custom Silicon: Amazon's chip business (Tranium and Graviton) generated a $25 billion annual revenue run rate as of Q2, with Tranium powering the majority of Bedrock inference.
    • Anthropic: Amazon views its investment and compute partnership as distinct from "circular financing," noting similar historical interdependencies (e.g., Adobe on AWS) and focusing on the mutual benefit of customers accessing best-in-class models.
    • OpenAI: Maintained as a partner to satisfy customer demand for choice, with Bedrock allowing users to access multiple models simultaneously.
    • Model Landscape: Jassy predicts a future with six to eight top-tier models, each excelling in different domains (coding, math, reasoning, visual arts), with Amazon's own models included in this competitive set.
  • AI Agents and Consumer Strategy

    • Amazon will not open its retail platform to external AI agents unless a "fair exchange of value" and a superior customer experience can be guaranteed.
    • The company is prioritizing internal agentic capabilities (e.g., Alexa for shopping) where it can control the context of price comparisons and purchasing logic.
    • Amazon reports over 400 million users interacting with Alexa for shopping in the last 12 months, driving significant downstream revenue.
    • The company anticipates AI agents could be deflationary, aligning with its core strategy of offering the lowest prices, broad selection, and rapid delivery.
  • Enterprise and Economic Outlook

    • Enterprise demand for cloud services remains robust with no observed attenuation in spending, despite high interest rates.
    • Jassy notes a shift from post-pandemic cost optimization back to migration from on-premises infrastructure to the cloud, with 80-85% of global IT spend still on-premises.
    • He projects that the on-premises to cloud migration ratio will flip over the next 10 to 20 years.
    • Consumer behavior remains resilient, characterized by price sensitivity, deal-seeking, and consistent purchasing activity ahead of the holiday season.