Who will be the next AWS, but for AI?
In our updated market reports, you will find everything you need
SUMMARY
Microsoft Azure is the most likely company to become the AWS-like platform for AI, although Google Cloud and AWS remain close enough to overturn that answer.
The winner will not be the company with the best model or the most GPUs. It will be the platform that companies keep using after models improve, chips get cheaper and today’s technical advantage fades.
NVIDIA is making the clearest infrastructure profits now, but its role looks more like the universal toll collector of AI than the place where companies store data, manage identities and run complete business systems.
Microsoft has the strongest full route from developer discovery to enterprise deployment. GitHub, Azure, Microsoft 365, Entra and Foundry let it earn at several points inside the same AI project.
Google Cloud is the fastest-moving challenger. It controls Gemini, TPUs, cloud infrastructure and mass-market distribution, and its recent cloud growth shows that those pieces are finally reinforcing one another.
AWS remains the safest alternative answer because it already hosts so much corporate data and software. Its weakness is not infrastructure; it is that fewer developers and employees naturally begin their AI work inside an Amazon product.
OpenAI may own the main interface without owning the underlying cloud. ChatGPT and Codex can become the place where people ask AI to work, while Azure, Google Cloud or AWS keeps the deeper data, security and deployment layer.
CoreWeave and Oracle can become enormous AI-capacity providers without becoming the default AI platform. Their growth is real, but both still need to prove that compute contracts lead to broad, sticky software spending.
Cheap and open models will make the surrounding platform more valuable, not less. When customers can switch intelligence providers, the durable money moves toward data preparation, permissions, monitoring, evaluation and workflow integration.
The market will probably have several winners rather than one AWS-style monopoly. Even so, most companies will still choose a primary control plane, and Azure currently has the clearest claim to that role.
Who will be the next AWS, but for AI?
Why is everyone suddenly looking for an AWS for AI?
The search for an AWS for AI makes sense now because AI infrastructure has become a huge business, while the company that will control it is still unclear.
Synergy Research Group estimated that companies spent $129 billion on cloud infrastructure in the latest reported quarter, $35 billion more than one year earlier. The market had already been expanding for years, but generative AI pushed its growth rate back to levels last seen when cloud computing was much smaller.
The companies benefiting from that growth sell very different things. NVIDIA recently generated $75.2 billion from data-center products in one quarter. Microsoft says its AI business has passed a $37 billion annual revenue run rate. AWS is accelerating again on a much larger cloud base, while CoreWeave has grown from a niche provider into a multibillion-dollar business.
Those numbers explain the excitement, although they do not identify a winner. NVIDIA supplies much of the machinery. CoreWeave rents specialist capacity. OpenAI sells models and applications. Microsoft, Google and Amazon wrap AI around cloud infrastructure, data, security and business software.
AWS became the defining company of the previous cloud era because it turned computing into a place where developers could build almost anything. The AI race now asks which company can create the same habit: when a business starts an AI project, where does it naturally go first, and how much more does it buy once it gets there?
What would the AWS of AI actually control?
The AWS of AI would control the environment where companies build and run AI systems, even when they change models, chips or applications.
AWS became powerful through a simple sequence. A customer rented computing, added storage, connected a database, then brought in identity, monitoring, analytics and security. Each new service made AWS more useful and leaving more painful.
An AI platform needs a similar path. A team might start with a model API, then add inference capacity, company data, evaluation tools, agent workflows, permissions, monitoring and deployment. The leading platform will make those next steps easy enough that customers keep buying from the same provider.
Model choice is central. Few companies want to rebuild an application whenever a better model appears. The platform should let a customer use Claude for one task, Gemini for another, an OpenAI model for coding and a smaller open model for cheap, repetitive work.
The economics matter just as much. Today, a provider can grow quickly by renting scarce GPUs. AWS-like power appears later, when the customer keeps paying for databases, security, data movement and software long after computing capacity becomes easier to find.
So we are looking for four things: a default entry point for developers, a broad set of connected services, growing customer spending across those services and an advantage that survives cheaper models and more abundant chips.
If you want more recent data on this point, please see our latest GPU cloud market report.
Are NVIDIA, OpenAI and Azure even competing in the same AI market?
NVIDIA, OpenAI and Azure overlap, but they sit in different parts of the AI market.
NVIDIA sits at the computing layer. Its GPUs, networking systems and CUDA software power a large share of model training and inference. CoreWeave, Crusoe, Nebius and other specialist clouds turn that hardware into rentable clusters.
OpenAI, Anthropic, Google, xAI, Meta and several Chinese laboratories compete at the model layer. They supply the intelligence that writes, reasons, sees images or operates tools. Leadership there can change quickly because a new model release can close a gap within weeks.
Microsoft Azure, AWS and Google Cloud are fighting over the operating layer. They connect models with corporate data, databases, identity systems, agent tools, security and application hosting. This layer looks closest to the original cloud market because customers build long-lived systems around it.
A final layer is distribution. ChatGPT, Microsoft 365, GitHub, Google Search, Android, Salesforce and ServiceNow can put AI in front of users before those users ever visit a cloud console.
No company owns every layer. NVIDIA has the leading hardware standard. OpenAI has the largest direct AI audience. Microsoft has the deepest enterprise distribution. Google controls models, chips, cloud infrastructure and mass-market products. AWS still has the largest traditional cloud base.
The likely winner will connect several layers while allowing customers to swap the parts that change fastest.
Who is making the real money from AI infrastructure today?
NVIDIA currently captures the clearest profit pool in AI infrastructure, while the cloud providers are building the more durable customer relationships.
NVIDIA’s latest quarter produced $81.6 billion of total revenue, a gross margin of roughly 75% and $53.5 billion of operating income. Data-center products supplied $75.2 billion of that revenue. No other company is extracting comparable profit from the AI buildout today.
AWS and Google show a different model. AWS produced $14.2 billion of quarterly operating income, while Google Cloud reached $8.8 billion and a 35.6% operating margin.
CoreWeave makes the contrast clearer. The company reported $1.2 billion of adjusted EBITDA, yet it still lost $740 million. Interest expense alone reached $536 million. Rapid demand therefore created a large business without creating clean net profits.
The accounting is different across these companies, so a direct margin ranking has limits. NVIDIA sells expensive systems, AWS and Google recognize usage over time, and CoreWeave finances infrastructure before receiving much of its contracted revenue.
Still, the pattern is clear. Hardware scarcity sends an exceptional share of the money to NVIDIA for now. The cloud platforms can win over a longer period because their customers keep paying for data, software and operational services after the original hardware purchase.
| Company or segment | Latest quarterly revenue | Profitability clue | Main control point |
|---|---|---|---|
| NVIDIA | $81.6 billion | About 75% gross margin | Chips, networking and CUDA |
| AWS | $37.6 billion | $14.2 billion operating income | Broad cloud consumption |
| Google Cloud | $24.8 billion | 35.6% operating margin | Cloud, models, data and TPUs |
| CoreWeave | $2.1 billion | $740 million net loss | Specialist AI capacity |
If you want more recent data on this point, please see our latest GPU cloud market report.
Will developers or CIOs choose the AI platform winner?
Developers will create the early preference, but CIOs and corporate data will decide where the largest AI workloads stay.
AWS first spread because developers could rent infrastructure without waiting months for an internal server. Large companies later approved it because AWS built the security, compliance and reliability they required. The AI winner needs both routes.
Microsoft has the best bridge between them. GitHub now serves more than 180 million developers, and GitHub Copilot is used by 140,000 organizations, nearly three times as many as one year earlier. A developer can begin with Copilot or Visual Studio Code, then move into Azure hosting, databases, model services and security.
OpenAI has a different form of pull. ChatGPT reaches more than 900 million weekly users, while its APIs process more than 15 billion tokens per minute. People already know the product, so OpenAI can enter companies through employees and development teams rather than through a formal cloud migration.
Corporate buyers introduce harder constraints. They care about where customer records live, who can access them, how actions are logged and which vendor already passed procurement. A slightly better model rarely compensates for moving a large data estate or rebuilding identity controls.
That is why Microsoft, AWS and Google have an advantage over a pure model provider. Developers may choose the first tool, but the company’s data, security rules and existing contracts usually shape the permanent platform.
Is NVIDIA already the AWS of AI?
NVIDIA is currently the most powerful company in AI infrastructure, although its position resembles a universal technology standard more than a full cloud platform.
CUDA is the core advantage. Developers have spent years building libraries, tools and optimized code around NVIDIA hardware. Clouds can offer competing chips, yet customers often ask for NVIDIA because the software already works and skilled engineers already know it.
The company is also taking a larger share of the complete data center. In its latest quarter, computing revenue reached $60.4 billion and networking revenue reached $14.8 billion. Networking grew 199% year over year, showing that NVIDIA is selling the connections between machines as well as the processors inside them.
That reach gives NVIDIA an enviable position. AWS, Azure, Google Cloud, Oracle and CoreWeave all compete to sell NVIDIA capacity. NVIDIA can benefit whichever cloud wins the customer.
The AWS comparison weakens when we look at what happens after the chip is installed. Most companies do not store corporate records, manage employee identities, operate production databases or run complete business applications through NVIDIA. Those workloads remain with a cloud or software platform.
NVIDIA also earns heavily during infrastructure buildouts and upgrade cycles. A cloud platform earns every day that an application runs, then collects more revenue when the customer adds storage, databases or security.
NVIDIA is the most valuable toll collector of the AI era. That may prove more profitable than owning the leading AI cloud, but it is a different job.
If you want more recent data on this point, please see our latest GPU cloud market report.
Can CoreWeave become the AWS for AI after the GPU shortage ends?
CoreWeave has built a major AI cloud, but the company has not yet shown that customers will stay for its software once specialist GPU capacity becomes easier to buy.
Its recent growth is exceptional. Quarterly revenue more than doubled year over year to $2.1 billion, while contracted backlog reached $99.4 billion. New agreements included a $21 billion commitment from Meta and a multiyear deal with Anthropic.
CoreWeave is also moving beyond raw GPU rentals. It added flexible reservations, spot capacity, dedicated inference, workload evaluation and the Weights & Biases development platform. Those products can give customers reasons to remain even when another provider offers the same chip.
The financial structure remains the hard part. During the same quarter, CoreWeave spent heavily on new infrastructure, paid $536 million in net interest and reported a $740 million net loss. It also secured an $8.5 billion delayed-draw loan facility as it expanded beyond one gigawatt of active power.
Large contracts make this financing possible, but they also concentrate the business around a small number of buyers. The two largest customers accounted for about 65% of quarterly revenue in CoreWeave’s filing. AWS grew into a utility for millions of customers; CoreWeave looks closer to a specialist supplier for a handful of enormous AI builders.
A startup cannot win this race by borrowing more than Microsoft, Google or Amazon. CoreWeave needs better cluster performance, easier software and higher utilization to compensate for its higher financing risk.
For now, CoreWeave has proved that an AI-native cloud can reach billions of dollars in revenue. It still has to prove that its platform remains special when GPUs are widely available.
| CoreWeave measure | Latest result | What we learn |
|---|---|---|
| Revenue growth | 112% year over year | Demand is already large |
| Contracted backlog | $99.4 billion | Buyers are reserving years of capacity |
| Top two customers | About 65% of revenue | Customer concentration remains high |
| Net interest expense | $536 million | Financing takes a large share of sales |
| Net loss | $740 million | Fast growth has not produced net profit |
Is Microsoft Azure the clearest AWS-for-AI candidate right now?
Microsoft Azure is currently the best single candidate because it can reach developers, employees, corporate data and IT buyers through products they already use.
Microsoft says its AI business has passed a $37 billion annual revenue run rate, up 123% from one year earlier. Azure and other cloud services grew 40% in the latest reported quarter, while Microsoft Cloud revenue reached $54.5 billion.
The distribution behind those figures is hard to copy. GitHub and Visual Studio reach developers. Microsoft 365 reaches office workers. Dynamics and Power Platform reach business teams. Entra, Defender and Azure reach the people who control identity, security and infrastructure.
Microsoft can therefore earn from several steps of the same project. A developer writes code with Copilot, selects a model through Foundry, connects company data, deploys the application on Azure and gives employees access through Microsoft software.
The model catalog also reduces dependence on any single laboratory. Foundry offers OpenAI models alongside products from Anthropic, Meta, Mistral, Cohere, DeepSeek and others. A customer can change the model while leaving its data and operations on Azure.
There is a real cost to this growth. Microsoft Cloud’s gross margin has slipped to 66% as the company builds data centers and absorbs more AI usage. Azure will only deserve the AWS comparison if AI customers expand into profitable data, security and software services.
Today, no rival matches Microsoft’s full route from developer discovery to corporate deployment. That breadth puts Azure slightly ahead.
Can Microsoft Azure win without leaning on OpenAI?
Microsoft Azure can now win even when a customer chooses another model, which makes the platform less dependent on OpenAI than it once was.
OpenAI gave Microsoft its early lead. Azure supplied infrastructure, Microsoft gained access to frontier models, and ChatGPT created urgency inside large companies. The partnership still matters because OpenAI usage drives cloud demand and because Microsoft has invested heavily in the company.
The relationship also contains friction. OpenAI sells directly to businesses, builds coding products, develops agent tools and wants to become the main interface for work. Those ambitions overlap with Copilot, GitHub and Microsoft 365.
Microsoft has spent the past two years making the surrounding platform more important than the model. Foundry supports a wide range of proprietary and open models. Azure provides the databases, identity controls, hosting, evaluation and monitoring that stay in place when a customer switches models.
A reseller loses when its supplier changes. A platform can keep the workload because the expensive parts of the application sit around the model.
Azure still benefits from OpenAI’s growth, and a serious break between the companies would create disruption. Yet Microsoft no longer needs every AI project to use an OpenAI model. That is enough to treat Azure as a platform in its own right.
Is Google Cloud catching Microsoft faster than expected?
Google Cloud is currently the fastest-improving AWS-for-AI candidate, and its latest results put it much closer to Microsoft than the old cloud rankings suggest.
Google Cloud revenue rose 82% year over year to $24.8 billion in the latest quarter. Operating income more than tripled to $8.8 billion, while the operating margin reached 35.6%. Backlog climbed by more than $50 billion in three months to $514 billion.
That mix of speed and profitability stands out. Cloud providers usually suffer lower margins when they add capacity quickly because new data centers begin depreciating before they are fully used. Google managed to accelerate growth while producing a margin close to AWS.
Google also controls more of the technical stack than Microsoft. It designs TPUs, builds Gemini models, operates the cloud and owns products such as Search, Workspace, Android, Chrome and YouTube. Lately, it has started selling TPU systems for use inside customer data centers, extending its reach beyond Google Cloud.
The latest results also include a warning. Alphabet raised its annual capital-spending plan to between $195 billion and $205 billion and produced negative free cash flow during the quarter. Part of the cloud acceleration came from TPU system sales, which resemble hardware revenue more than recurring software consumption.
Google’s remaining weakness is habit. Many companies still treat AWS or Azure as the natural home for core enterprise applications. Google must turn Gemini interest and TPU demand into long-lived database, security and application workloads.
If Google repeats anything close to its current cloud performance for several more quarters, our answer could change. Right now, Google is the main challenger and the one gaining ground fastest.
If you want more recent data on this point, please see our latest GPU cloud market report.
Could AWS simply win the AI cloud race again?
AWS can still become the leading AI platform, and its existing customer base gives it a better chance than the recent AI narrative implies.
AWS remains the largest traditional cloud business. Revenue grew 28% to $37.6 billion in its latest quarter, the fastest growth in 15 quarters, while operating income reached $14.2 billion.
Its advantage appears whenever AI touches existing systems. A company with data in S3, databases on AWS and security policies built around Amazon has a practical reason to use Bedrock, SageMaker and AgentCore. Moving the data can cost more than changing the model.
Bedrock now serves more than 100,000 organizations and offers hundreds of models. Amazon is also attacking the largest AI cost through its own chips. Its Graviton, Trainium and Nitro business has passed a $20 billion annual revenue run rate and is growing at a triple-digit rate.
The scale behind that chip effort has become difficult to ignore. Amazon says it obtained more than 2.1 million AI chips over the previous 12 months, with Trainium making up more than half. OpenAI has committed to use about two gigawatts of Trainium capacity, while Anthropic has agreed to secure up to five gigawatts across current and future generations.
AWS entered the generative AI race without the attention created by ChatGPT or Gemini. That made the company look late even while its customers, infrastructure and profits remained enormous.
The gap is mainly in developer excitement and everyday distribution. Microsoft owns GitHub and Microsoft 365. Google owns Search and Workspace. AWS needs Bedrock and its agent tools to become the place developers begin, rather than simply the easiest choice for customers already on AWS.
The next AWS for AI may still be AWS. We rank it behind Microsoft today because Microsoft has the stronger path into both developers and employees.
Is Oracle becoming more than a giant GPU landlord?
Oracle is now a serious AI infrastructure provider, but its platform still depends too heavily on a few huge contracts to look like the next AWS.
Oracle Cloud Infrastructure revenue rose 93% to $5.8 billion in its latest quarter. Remaining performance obligations reached $638 billion, up from $138 billion one year earlier. That jump shows how aggressively AI laboratories and large companies are reserving future capacity.
The composition deserves a closer look. Oracle disclosed that most of the recent backlog increase came from large AI contracts. Customers had prepaid for GPUs or supplied the hardware themselves, and those arrangements accounted for $75 billion. This lowers Oracle’s financing burden, although it also shows that much of the demand begins with infrastructure rather than Oracle software.
Oracle still has useful assets around the compute. Its databases hold critical corporate information, while Fusion, NetSuite and Oracle Health sit inside finance, operations and healthcare workflows. Those products give Oracle a route into enterprise AI that a specialist cloud lacks.
The company is spending heavily to serve the contracts. Free cash flow was negative $23.7 billion for its latest fiscal year. Oracle raised $48 billion through debt and equity during that period and expects roughly $40 billion more in the following year.
Oracle can become one of the largest AI capacity providers. Becoming the default platform requires a broader developer ecosystem and clear evidence that GPU contracts lead customers into databases, agents, security and applications.
At the moment, Oracle looks like an increasingly important landlord with valuable software attached, rather than the place most teams naturally start an AI project.
Can OpenAI become the AI platform without owning the cloud?
OpenAI can become the main interface for AI work without owning the leading cloud, giving it a role closer to an operating system than to AWS.
The company already has unmatched direct distribution in AI. More than 50 million people pay for ChatGPT, and more than one million businesses use OpenAI products. Enterprise activity represents over 40% of revenue.
OpenAI is also moving deeper into work. Codex has reached three million weekly active users, while the company is building tools for agents, enterprise controls, application connections and shared work. A developer or employee may increasingly begin inside ChatGPT or Codex and let OpenAI choose the infrastructure underneath.
That route can be extremely valuable. Windows became powerful without Microsoft manufacturing every computer, and Google Search grew without Google owning every network that carried a query.
OpenAI’s difficulty is trust from builders. The company sells models through APIs while launching products that can compete with the companies using those APIs. A startup may hesitate to build its whole business on a provider that can enter its category.
Physical infrastructure is another constraint. OpenAI relies on large commitments with Microsoft, AWS, Oracle, CoreWeave and other partners. Spreading demand across suppliers reduces dependence, but the underlying cost and delivery still sit partly outside OpenAI.
We expect OpenAI to own a major platform layer. It could become the dominant place where people ask AI to perform work, while Microsoft, Google or AWS operates the deeper environment where company data and applications live.
Will cheap and open AI models stop one company from dominating?
Cheap and open AI models will weaken model lock-in, but they will strengthen platforms that handle everything around the model.
Model leadership already moves quickly. Stanford’s latest AI Index found that the gap between leading US and Chinese models had nearly closed, with the top American model ahead by only 2.7% on its aggregate measure. A company cannot safely assume that today’s best model will remain best.
Open standards are spreading as well. Anthropic contributed the Model Context Protocol to the Linux Foundation’s Agentic AI Foundation, alongside projects from OpenAI and Block. MCP gives agents a common way to connect with tools and data, reducing the need to build every integration for one vendor.
This kind of portability lowers the value of exclusivity. Customers can route easy tasks to a cheap model, sensitive tasks to a locally hosted model and difficult reasoning to a frontier model. Most serious businesses will use several.
The hard work then moves into the surrounding system: preparing data, controlling permissions, evaluating answers, logging actions, monitoring costs and connecting agents to real software. Companies do not want to rebuild that machinery every time they change a model.
Open models therefore favor Microsoft, AWS and Google more than a laboratory that depends mainly on model premiums. The broad platforms can earn money whichever model the customer selects.
Does multi-cloud AI mean there cannot be one winner?
Multi-cloud AI will prevent one company from owning the whole market, but most businesses will still choose a primary platform.
Frontier laboratories already spread their workloads. OpenAI buys capacity from Microsoft, AWS, Oracle and CoreWeave. Anthropic works across AWS, Google and specialist providers. They need more chips than one supplier can reliably deliver, and they want leverage in price negotiations.
Large companies will also mix products. Employees may use ChatGPT, developers may prefer Claude, marketing teams may work in Gemini and production applications may run beside existing databases on AWS or Azure.
That variety sounds more fragmented than it usually is. A company still needs one main identity system, one set of security policies, one place to monitor agents and a small number of approved data platforms. Moving those systems is much harder than changing a model endpoint.
The market will resemble corporate software today. Businesses buy from many vendors, yet Microsoft, AWS, Google, Salesforce or ServiceNow often owns the central relationship for a particular workflow.
We do not expect an AWS-style monopoly. A leading AI platform can still capture a much larger share than rivals by becoming the default control plane for data, permissions and deployment.
Which company has the strongest AI platform flywheel today?
Microsoft currently has the strongest complete AI platform flywheel, while Google has the fastest one and NVIDIA has the most profitable one.
Microsoft starts with GitHub, Microsoft 365 and a huge base of Azure customers. Those products create AI demand. Foundry supplies models, Azure supplies computing and databases, Entra controls access, and Microsoft 365 puts the finished tools in front of employees. One customer can pay Microsoft at several points in the same workflow.
Google connects a different set of assets. Gemini drives demand for TPUs and cloud capacity. Lower infrastructure costs can improve Gemini products. Search, Workspace and Android distribute them. Google Cloud then sells the same models and infrastructure to companies. Its latest cloud growth shows that the loop is now producing serious revenue.
AWS begins with the deepest traditional cloud base. Bedrock and AgentCore can pull customers into Trainium, storage, databases and security. Its custom chips may also improve margins as inference becomes the larger workload.
NVIDIA’s loop remains narrower but extremely strong. More CUDA developers attract more software, which creates more demand for NVIDIA systems, which funds better chips and networking. Every cloud then competes to carry the platform.
CoreWeave and Oracle have contract-driven loops. Large bookings support infrastructure expansion, which creates capacity for more bookings. These businesses need software and a broader customer base before that cycle becomes as durable as the hyperscalers’ ecosystems.
| Candidate | Main advantage today | Main weakness | Our current view |
|---|---|---|---|
| Microsoft Azure | Developers, enterprise software and cloud in one system | AI spending is pressuring cloud margins | Best overall candidate |
| Google Cloud | Models, TPUs, cloud and consumer distribution | Still changing enterprise habits | Fastest challenger |
| AWS | Largest existing cloud base and strong custom chips | Weaker direct AI distribution | Fully capable of winning |
| NVIDIA | Universal hardware and software standard | Does not own most business workflows | Most powerful toll collector |
| OpenAI | Largest direct AI audience | Depends on partners and competes with builders | Likely interface leader |
| CoreWeave | Specialist performance and secured capacity | Debt and customer concentration | Major supplier, early platform |
| Oracle | Huge contracts and valuable databases | Limited bottom-up developer pull | Strong infrastructure contender |
Who will be the next AWS for AI?
Microsoft Azure is the most likely AWS-like platform for AI today, although Google and AWS are close enough that the race is far from settled.
Microsoft has the best combination of developer access, enterprise distribution, cloud infrastructure, data products, security and model choice. A customer can discover AI through GitHub or Microsoft 365, build through Foundry, connect data in Azure and keep the finished application inside Microsoft’s operating environment.
Google has become the main threat. Its cloud business is growing much faster, its margins are already strong and its control of Gemini, TPUs and mass-market products could eventually produce the better full stack. Another few quarters near its current trajectory would force us to reconsider the ranking.
AWS remains the safest alternative answer. It already owns the largest cloud installed base, its chip business is growing quickly and Bedrock gives customers broad model choice. Amazon needs to make its existing platform the natural home for AI.
NVIDIA will probably collect the most obvious infrastructure profits for some time. Its position spans every major cloud, although the company controls less of the customer’s data, security and everyday work. CoreWeave and Oracle can become huge capacity providers without becoming the default development platform. OpenAI may own the user interface while somebody else owns the underlying enterprise system.
The conclusion is sharper than “too early to tell.” Microsoft leads because its products cover the whole journey from the first line of code to a company-wide deployment. Google could overtake it if its latest cloud acceleration proves durable. AWS could retake the lead through its installed base and cheaper custom chips.
As of now, Microsoft Azure is our answer. The eventual market will have several powerful winners, but Azure has the clearest path to becoming the platform that companies keep using even as models, chips and AI applications change.
If you want more recent data on this point, please see our latest GPU cloud market report.
OUR METHODOLOGY
This analysis tests which company is most likely to become the AWS-like platform for AI. We compare the leading candidates across the dimensions that made AWS powerful: where developers begin, how broadly a provider can support an AI project, whether customers expand their spending over time, how deeply the platform connects with corporate data and operations, and whether its advantage can survive changing models, cheaper inference and more abundant computing capacity.
We kept different forms of leadership separate. Current revenue, profitability, developer adoption, infrastructure scale, direct user distribution and long-term platform control do not necessarily belong to the same company. NVIDIA can dominate the hardware profit pool without owning the main enterprise relationship, while OpenAI can lead the user interface without operating the underlying cloud.
No single figure determined the ranking. We prioritized evidence of actual scale, usage, customer commitments, financial performance, model choice, product integration and access to developers or corporate buyers. We then looked for several independent indicators supporting the same conclusion.
Recent quarterly figures are used as current evidence, not as a permanent forecast. Cloud growth, capital spending, margins, backlog and AI adoption can move quickly, so the final ranking reflects the market as it stands now rather than a mechanical score or a prediction based on one quarter alone.
We also tested whether each advantage would remain valuable after the GPU shortage eases and model prices fall. This is important because AWS-like power appears when customers keep buying databases, identity, security, monitoring, data services and workflow software after raw computing capacity becomes easier to obtain.
Key sources used for this analysis include: Synergy Research Group on Q1 2026 cloud-infrastructure spending and growth, NVIDIA’s first-quarter fiscal 2027 results, Microsoft’s fiscal 2026 third-quarter earnings call, Microsoft’s fiscal 2026 third-quarter financial results, the Azure AI Foundry model catalog, GitHub’s Octoverse research, OpenAI’s adoption figures for ChatGPT and business users, and OpenAI’s Codex adoption update.
Additional sources include: CoreWeave’s first-quarter 2026 results, Alphabet’s cloud and full-year results, Amazon’s first-quarter 2026 results, AWS documentation on Bedrock model choice, Oracle’s fourth-quarter and fiscal-year 2026 results, Stanford’s 2026 AI Index technical-performance findings, the Linux Foundation’s Agentic AI Foundation announcement, and Anthropic’s Model Context Protocol documentation.
