The hyperscale big boys came out as "Leaders" of Gartner’s first-ever Magic Quadrant for cloud AI infrastructure, with neocloud heavyweights finding themselves largely finding themselves as either "Challengers" or "Niche Players."
Google Cloud was the highest-placed leader, with the research giant praising the vendor’s custom silicon offerings and integrated AI Hypercomputer architecture. Gartner said Google’s Tensor Processing Units (TPUs) provide enterprises with greater efficiency and scale for foundation model development compared to generalized cloud infrastructure, while also reducing dependencies on “the clogged supply chain of Nvidia.”
Just below Google was Amazon Web Services (AWS), with Gartner labeling its stack as the industry’s “most extensive and mature cloud infrastructure” for AI workloads. AWS’ hardware isolation offering Nitro drew praise for allowing customers to more securely run sensitive or heavily regulated AI workloads, as did its varying levels of managed infrastructure, providing customers with more choice in how they want to manage directly.
Following the hyperscale pair were Alibaba Cloud and Microsoft. Gartner praised the pair for their wide spectrum of cloud AI infrastructure services.
Specific kudos for each vendor saw Microsoft’s linking of its cloud AI infrastructure with its other comprehensive cloud services praised for providing users with simplified deployment means for existing Azure and application users. Alibaba Cloud, meanwhile, won plaudits for its support for open-source technologies, including its own Qwen line of AI models. Its PAI-Lingjun specialized compute platform was also cited as being a solid choice for large-scale training and inference workloads.
The Chinese vendor, however, did draw concerns from Gartner over regulatory and data sovereignty concerns, given that storing data on a Chinese-owned and operated cloud provider potentially provides headaches for some users. Alibaba Cloud was also docked for lacking some of the more advanced technologies available to its western rivals owing to export sanctions.
The final two leaders on Gartner’s cloud AI infra Magic Quadrant were Oracle and Huawei Cloud.
Like Alibaba, Gartner cited potential geopolitical concerns for Huawei, but recognized the vendor’s comprehensive AI life cycle support and strong hybrid and private cloud offerings. Huawei’s custom stack, including its Ascend NPUs and Kunpeng processors, also attracted acclaim from Gartner, with the research firm suggesting control over its own hardware and software stack provides for deeper optimization and resource efficiency, “resulting in faster training times and a better cost-performance ratio for enterprises running large-scale AI workloads.”
Oracle’s strengths, meanwhile, included its OCI Supercluster architecture, providing customers with the means to train foundation models and some of the most demanding AI workloads.
The U.S. giant was cited as having a “smaller ecosystem” of prebuilt, third-party AI integrations and less-mature AI support tooling. Gartner suggests that customers seeking more sophisticated capabilities may need to integrate and manage third-party tools or build custom solutions, which would potentially introduce additional integration and maintenance complexity.
Neocloud disruptors make their mark #
Some of the darlings of the neocloud space found themselves on Gartner’s AI infrastructure Magic Quadrant, though none managed to squeeze their way into the leader category. A total of six of the disruptors made it to the list, with those coming closest to the leadership echelons being CoreWeave and Lambda.
CoreWeave found itself as the top-ranking "Visionary" player, with the pure-play infrastructure provider drawing plaudits for its “purpose-built” platform and native Kubernetes environment.
Lambda topped the niche players segment, while just about eking over the line into the challengers category. Among its strengths cited were cost structure, with a consumption-based pricing found to be “often lower than other vendors in the market,” combined with zero egress fees. Its significant relationship with Nvidia, including a $1.5 billion graphic processing unit (GPU) lease deal, was also noted as a win, providing Lambda with the means to skirt GPU scarcity and ultimately reduce customer project delays typically caused by hardware constraints.
CoreWeave’s own Nvidia partnership was labeled a strength, but Gartner warned that a heavy reliance on the GPU giant for both hardware and financial support introduces both supply chain risk and potential price volatility, as the ability to scale is thereby directly tied to Nvidia’s allocation decisions.
Nscale was the only other neocloud in the niche playes segment, while Nebuis and Crusoe joined CoreWeave in the visionaries category.
Vultr was the only neocloud that found itself in the challengers quadrant, with the Florida-headquartered provider labeled as appealing to enterprises wanting cost-optimized cloud AI infrastructure along with other cloud-native services and flexible pricing.
Its consumption-based pricing was cited as being “often lower than hyperscalers,” while its more diverse compute portfolio, as it provides access to both AMD and Nvidia hardware, provides customer choice. Among the cautions Gartner listed, however, were issues around support response and operational maturity, suggesting some users indicated that customer support times for urgent, complex issues “can occasionally be slow.”
A niche inclusion? #
Arguably, the most intriguing inclusion on this list was Cloudflare. Nestled in the niche players segment, the San Francisco-based firm was cited by Gartner as being “well-positioned” to capitalize on inference use cases as a strategic edge partner alongside centralized cloud providers.
Praised for its zero egress fees and “desirable pricing,” Cloudflare was found to be among the players especially optimized for handling inference workloads, with its GPU-enabled servers capable of delivering responses with latency lower than 50 milliseconds, depending on the model.
Its strength in inference was also its weakness, with Gartner suggesting its lack of support for large-scale AI model training meant potential customers would have to partner with another provider for such services.
Cloudflare’s pricing was another string to its bow. The vendor charges customers based on actual inference usage rather than idle GPU capability, a consumption-based model that may prove tantalizing for more cost-conscious customers.
Definitions and differentiations #
The inaugural Magic Quadrant for cloud AI infrastructure saw Gartner define cloud AI infrastructure providers as those that “focus on delivering infrastructure optimized for AI workloads including AI model training, inference, and servicing.”
To feature on the list, in addition to client relevancy, Gartner determined that providers need to be operational on at least two continents, have at least 50 paying customers, and at least $50 million in revenue.