Amazon.com Inc. just landed a key hire from its biggest cloud rival. Steve Molloy, a chip industry veteran who spent the past seven years at Google, recently joined Amazon in a newly created position focused on AI chips. The move signals Amazon’s determination to close the gap with Google’s tensor processing units, or TPUs, as demand for custom silicon surges across the AI landscape. The Information broke the news first, highlighting how Amazon aims to tackle CEO Andy Jassy’s push for more efficient inference workloads.
Molloy’s arrival comes at a pivotal moment. Amazon’s in-house silicon efforts, led by Trainium and Inferentia chips, now generate over $20 billion in annual run rate revenue, growing at triple-digit percentages year over year. Jassy revealed in his recent shareholder letter that if the chips business operated independently—selling to AWS customers and third parties alike—its run rate would hit $50 billion. “There’s so much demand for our chips that it’s quite possible we’ll sell racks of them to third parties in the future,” he wrote. Trainium powers most inference on Amazon Bedrock, AWS’s fast-growing service, saving tens of billions in capital expenditures annually compared to rivals’ hardware. East Bay Times covered Jassy’s comments, noting the unit’s explosive trajectory.
Google’s TPUs set the benchmark. Molloy, during his tenure there, contributed to designs that dominate Alphabet’s AI infrastructure. Now at Amazon—reporting into a structure under Peter DeSantis, the AWS veteran tapped last December to oversee AI models, custom silicon, and quantum efforts—he’ll help refine Trainium generations. Trainium3 servers are four times faster and more energy-efficient than predecessors, Amazon announced late last year. Trainium4, designed for compatibility with Nvidia’s NVLink Fusion networking, is already in development. And Trainium2? Sold out. Yahoo Finance detailed DeSantis’s role, while AOL Finance reported on Trainium3 benchmarks.
But competition heats up. Everyone wants independence from Nvidia Corp.’s dominance. Google builds TPUs. Meta Platforms Inc. deploys MTIA chips. Even Anthropic PBC, backed by up to $25 billion more from Amazon atop $8 billion prior, trains models on Trainium—committing over $100 billion to AWS compute across five gigawatts. Intel Corp. is in talks with Amazon and Google for advanced packaging on custom AI silicon like Trainium and TPUs, potentially shifting assembly from TSMC. Amazon partners with Marvell Technology Inc. for Trainium designs under a five-year deal. IDN Financials and TechSpot outlined Intel’s overtures; Motley Fool spotlighted Marvell’s role.
Custom Silicon Fuels AWS’s Edge
Two large AWS customers already want all of Amazon’s 2026 Graviton CPU capacity—Amazon’s custom Arm-based processors—but can’t have it due to competing demand. AWS plans $200 billion in 2026 capex, much already committed by customers like OpenAI and unnamed others, monetizing into 2027-2028. Jassy calls lacking custom silicon a “structural disadvantage” for big inference businesses chasing margins. Amazon ramping Nvidia GPUs too, per Nvidia CEO Jensen Huang, but in-house chips deliver hundreds of basis points in operating margin gains.
Molloy’s expertise matters here. Poaching from Google underscores the talent war. Arm Holdings Plc. grabbed Amazon AI chip director Rami Sinno last year to build its own silicon. Amazon rehires laid-off engineers for AI validation amid coding mishaps. Yet AWS leads: 40% of its AI compute from custom chips, outpacing most except Google Cloud. Anthropic’s models? All trained on Trainium, per AWS CEO Matt Garman.
Inference shift favors Amazon. Training demands peak GPUs; running models at scale needs efficiency. Trainium excels there. Jassy projects capex savings and margin expansion as workloads migrate. With Molloy aboard, next-gen Trainium could widen the lead. Short punchy. Amazon’s silicon bet pays off.
Risks linger. CPU shortages loom. Supply chains strain under hyperscaler demands. Intel’s packaging pitch might diversify manufacturing. But Amazon’s vertical integration—from Annapurna Labs acquisition to massive data centers like Project Rainier—positions it strongly. Jassy’s letter flags unannounced deals signaling more upside.
Steve Molloy. One hire. Massive implications. As AI compute explodes, Amazon’s chip arsenal grows deadlier.
WebProNews is an iEntry Publication