Cloud AI Infrastructure (Gartner)

Gartner published its first Magic Quadrant for Cloud AI Infrastructure on 6 July 2026, with positions recorded as of June, and placed six vendors on the Leader rung: Amazon Web Services, Google Cloud, Microsoft, Oracle, Alibaba Cloud and Huawei Cloud.

A category being scored for the first time is worth reading for what it declares rather than for who is on top. Six analysts are credited on this one, which is a larger byline than Gartner normally assembles and a signal about how much ground the report had to cover.

The first Magic Quadrant for Cloud AI Infrastructure, and its six-vendor Leader rung

The report is authored by Dennis Smith, Ed Anderson, Ashish Banerjee, Alessandro Galimberti, Wataru Katsurashima and Carolin Zhou. The Leader rung is corroborated in two directions: a trade report on the quadrant names AWS, Google, Oracle and Microsoft at the top of it, and Huawei Cloud and Alibaba Cloud have each announced their own Leader placements.

First editions tell a buyer where a market's edges are, and this one draws them around a group that is not a peer set in any ordinary sense. Four of the six are general-purpose cloud providers that sell hundreds of services and happen to sell AI capacity among them. Two are Chinese hyperscalers whose placement history in Gartner's cloud coverage has been different, which is the subject of a section below.

The field is wider than the Leader rung. Tencent Cloud, OVHcloud and Vultr appear in the Challenger band. CoreWeave and Nebius are Visionaries, each confirmed in its own announcement. Cloudflare, Crusoe, IBM, Lambda, Nscale and Scaleway also appear in published accounts of the chart, though the boundaries of their placements are not established by the sources that name them.

The Leader rung is four general-purpose clouds plus the two Chinese hyperscalers

The first four names are easy to explain. AWS, Google and Microsoft built the public cloud market and their AI infrastructure is an extension of a business that already existed. Oracle is the interesting one in that group, because it spent a decade as the cloud provider that enterprise buyers did not choose, and the AI build-out gave it a second opening. Its position on the execution axis rests on capacity it can deliver and customers who were already running its databases.

The absence worth naming is IBM, which appears in the chart but not on the Leader rung. IBM was one of the earliest enterprise names in applied AI, and the first quadrant for this category places it outside the top band, which says something about how differently this market scores from the one IBM's reputation was built in. Reputation in AI research and position in AI infrastructure are two different measurements, and only one of them is in this report.

The Challenger band describes a different kind of vendor again. Tencent Cloud is the third of the Chinese hyperscalers, behind the two that made the Leader rung. OVHcloud is European and sells into a market where residency rules create demand that the American clouds serve awkwardly. Vultr is an independent that has been building GPU capacity without a hyperscale business underneath it. The common thread is that all three can deliver AI infrastructure and none of them can claim the breadth of a general-purpose cloud, which is a fair reading of a band that sits below the top one on execution rather than on vision. For a buyer inside one of those footprints, the placement measures the ceiling of the business, not the quality of the capacity.

The vendors that only sell AI infrastructure sit furthest right on vision

CoreWeave and Nebius are Visionaries, and both companies stated the placement themselves. Nebius describes its position as aimed at AI-native startups, researchers, software vendors and enterprises, and lists early access to current-generation GPUs through its standing as an NVIDIA reference platform cloud partner, custom data center and server designs, InfiniBand networking, managed Kubernetes with a Slurm layer on top, an inference service, and consumption-based pricing. CoreWeave's own account emphasizes purpose-built AI infrastructure, its NVIDIA relationship, and growth contracted in advance.

The structural point is not which of them is better. It is that the vision axis rewards a company with no legacy portfolio to protect, and both of these vendors can point an entire roadmap at one workload type because it is the only workload type they serve. The execution axis measures something else, and it rewards regions, reference customers, financing and the ability to deliver capacity on a date. That is why the pure-plays sit right and the hyperscalers sit high, and a buyer reading the chart as a single ranking will misread the shape.

Alibaba Cloud led this quadrant two months before it led the cloud quadrant

Gartner's Magic Quadrant for Strategic Cloud Platform Services, published on 1 September 2026, names five Leaders and describes Alibaba Cloud's placement as its first, as the only China-based and Asia-Pacific vendor in the band.

The AI infrastructure quadrant had already placed Alibaba Cloud on its Leader rung on 6 July, and Huawei Cloud beside it. So the newer, narrower evaluation is where the Chinese hyperscalers reached Gartner's top band first, and the general cloud quadrant followed two months later with one of them. Huawei Cloud's position is the sharper contrast: a Leader in AI infrastructure and a Challenger in the general cloud quadrant published weeks afterward.

For a buyer, that gap is informative rather than contradictory. It means the AI build-out is the part of these vendors' businesses that Gartner has been able to evaluate favorably, and that the wider cloud portfolio has not caught up to it. It also means the two documents should be read together, since a shortlist drawn from one of them will be missing vendors that the other scores well.

Gartner's definition covers the stack, not the accelerators

The market is defined as infrastructure optimized for AI workloads including training, inference and servicing. The criteria spread across accelerator clusters, high-performance storage, data preparation, integration with AI and machine learning libraries, model training platforms, inference engines, model API services, and operational governance.

Read that as a purchase list rather than as a definition, because each item is bought differently. Accelerator clusters are a capacity contract with a delivery date. High-performance storage is a throughput requirement that most buyers discover late. Data preparation is where the project actually spends its time. Inference engines and model API services are the runtime. Governance is the part that decides whether any of it can be audited.

A vendor that scores well on accelerators and weakly on governance is selling the first third of that list. The first edition of a quadrant tends to compress these differences into a single placement, which is the reason the report is worth reading past the chart.

Forrester scored the same ground eight months earlier with thirteen vendors

The cross-analyst counterpart is The Forrester Wave: AI Infrastructure Solutions, Q4 2025, published on 16 December 2025 and scoring thirteen vendors, with an emphasis on compute, networking and storage rather than on platforms or applications. AWS, Google Cloud and Alibaba Cloud are the Leaders confirmed by the vendors' own announcements, and all three are Leaders at Gartner as well.

The three that can be checked are therefore Leaders at both firms, which is the strongest agreement between two analysts that a first edition can produce, and it is a small sample rather than a verdict on the whole field.

The criteria sets explain where the two documents will disagree. Forrester describes its evaluation as centered on compute, networking and storage, which is the hardware and fabric layer, and it states that it deliberately left platforms and applications out of scope. Gartner's definition reaches further up: data preparation, model training platforms, inference engines and model API services are all inside it, and so is operational governance. A vendor that is excellent at delivering capacity and weak at the software layer above it can therefore score well in one document and poorly in the other without either firm being wrong. That is the single most useful thing to understand before comparing a Forrester placement with a Gartner one in this market, because the two are not competing rankings of the same list.

Forrester's own framing is the more useful part. It states that the Q4 2025 edition took a different approach from its 2023 predecessor, doubling down on infrastructure differentiation, and warns that comparing the two editions may be particularly ill advised. An analyst firm telling buyers not to read its two reports as a trend line is worth taking at its word, and it applies to the wider habit of treating any placement movement as a signal. Gartner's first edition, published seven months after Forrester's, is a starting point rather than a comparison.

What to ask before you commit to an AI infrastructure platform

Ask what the capacity commitment actually guarantees. An accelerator contract has a quantity, a delivery date and a term, and a placement on a quadrant has none of those. The question that decides the project is when the capacity arrives, and the chart does not answer it.

Ask where the data has to live before asking which vendor is best. Two of the six Leaders are subject to data residency and export rules that make them unusable for some buyers and the only viable option for others, and a shortlist that ignores that produces a comparison nobody can act on.

Ask how the platform handles inference separately from training. Training is a scheduled capacity problem and inference is a latency problem with a cost curve attached, and a vendor that is strong on one is not automatically strong on the other, which is why the pure-play vendors sit where they do on the vision axis.

Ask what the governance and metering layer looks like on day one. Token consumption, model routing, per-team attribution and audit trails are the parts of this market that a buyer keeps paying for after the capacity is delivered.

Ask which of the two analyst documents you are relying on. Forrester scored thirteen vendors in December 2025 against compute, networking and storage. Gartner scored a wider field in July 2026 against a definition that includes model API services and operational governance. They are not measuring the same list, and a vendor absent from one of them has not necessarily been judged badly by it.

Ask what the first edition cannot tell you. There is no prior report to compare against, no movement to interpret, and no way to distinguish a placement that reflects a durable position from one that reflects a market still forming. That is not a reason to ignore the chart. It is a reason to read the rest of the document before acting on it.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner's research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.

The layer above this market is covered separately on this site. The Forrester Wave: AI Platforms, Q3 2026 scores fifteen vendors against nineteen criteria, and it is the document that decides what runs on the capacity a buyer provisions here. AI Platforms covers model routing, agent frameworks and the governance tooling that this quadrant folds into its infrastructure definition instead. Forrester evaluates frontier model providers outside both, in a Landscape due in the fourth quarter of 2026 and a Wave in the first quarter of 2027.

Not every AI workload lands in a public cloud. Distributed Hybrid Infrastructure covers the systems that put the same accelerators on a buyer's own floor, and the tradeoff is visible in this quadrant's own vendor list rather than only in the architecture: every vendor on the Leader rung also sells a deployed or sovereign variant of its platform, which a list of clouds on its own does not convey.