Amazon revealed on Tuesday that Uber is significantly increasing its reliance on AWS cloud services. This expansion includes utilizing more of Amazon’s proprietary chips for its ride-sharing operations. Specifically, Uber will scale up its deployment of AWS’s Graviton (an energy-efficient, ARM-based server CPU) and begin a pilot program to test Trainium3, AWS’s AI chip designed to compete with Nvidia.
The significance of this agreement lies less in posing a direct, long-term challenge to Nvidia and more in Amazon’s assertive move against its primary cloud rivals, Google and Oracle.
Historically, Uber managed its own data centers. However, in 2023, the ride-hailing firm made headlines by securing substantial, multi-year cloud computing contracts with Oracle and Google. The stated objective was to migrate most of its IT infrastructure from proprietary data centers to these two cloud providers.
As recently as December, Uber reaffirmed this objective, stating in a blog post:
Beginning in February 2023, Uber initiated its transition from on-premise data centers to cloud infrastructure, utilizing OCI and Google Cloud Platform. This process involved the complex task of migrating substantial workloads while also integrating Arm-powered compute instances into an environment previously dominated by x86 architecture.
Notably, in that same post, Uber highlighted its deployment of ARM chips from Ampere within Oracle’s cloud. This detail adds an intriguing layer to the situation.
To grasp the intricate web of connections within Silicon Valley, one might consider the background of Ampere.
Ampere was established by Renee James, a former high-ranking Intel executive, following her unsuccessful bid for the CEO position at the chip manufacturer. Leveraging her influence as an investor at Carlyle, a private equity firm, and her board membership at Oracle, she secured the necessary funding to launch Ampere. Oracle held approximately one-third of the company, which necessitated James relinquishing her role as an independent Oracle director due to the conflict of interest arising from the investment.
(Incidentally, James played a pivotal role on the board in approving Oracle’s $9.3 billion acquisition of NetSuite in 2016, a company where Larry Ellison held significant shares. This transaction subsequently led to an unsuccessful shareholder lawsuit contending that Oracle had overpaid.)
By December, SoftBank, a key competitor, had acquired Ampere, with Oracle divesting its stake for a sizable $2.7 billion pre-tax profit. James departed Oracle’s board at the close of 2024 and is no longer affiliated with Ampere.
Oracle is actively raising capital to construct data centers for projects like OpenAI and Stargate. Larry Ellison explained that Oracle divested Ampere because the company no longer viewed in-house chip design as a competitive edge for its data centers. Instead, Oracle now favors purchasing chips and has entered into significant agreements with Nvidia.
Furthermore, Oracle, SoftBank, and Nvidia are interconnected within OpenAI’s network of reciprocal agreements, which are intended to finance the substantial data center expansion of the AI model developer.
However, AWS has now revealed securing an expanded contract from Uber, a prominent Oracle customer, notably leveraging its proprietary, in-house-designed chips.
Uber joins the ranks of major technology companies such as Anthropic, OpenAI, and Apple, all of whom have either adopted or expanded their use of AWS, driven by these advanced AI chips. In December, Amazon CEO Andy Jassy declared that Trainium had already become a multi-billion dollar enterprise.
(To explore the team and laboratory responsible for these chip designs, refer to our exclusive facility tour.)