The Week In Cloud: May 26

If you blinked during the week of May 26, 2025, the cloud industry probably launched three new agent platforms, five model updates, and at least one slide deck capable of causing existential dread in an IT budget meeting. This was not a sleepy week for cloud computing. It was the kind of week that made one thing crystal clear: cloud is no longer just where software runs. It is now where AI agents work, where enterprise data gets negotiated into shape, and where infrastructure spending starts to resemble a small nation’s GDP.

The biggest cloud story was not a single product launch. It was the convergence of several trends that have been circling each other for months: hyperscalers racing to become the best home for AI agents, hybrid cloud vendors trying to prove they still matter, and enterprise customers demanding something more practical than “trust us, the chatbot is magical.” In plain English, the market is moving from AI demos to AI operations. The mood has shifted from “look what this model can do” to “show me how this works with my data, my controls, my security, my budget, and my mildly overworked platform team.”

The big theme: cloud has become the operating system for AI agents

The clearest signal this week was that nearly every major player now wants to own the full agent stack. Not just models. Not just compute. The whole stack. That means orchestration, observability, governance, connectors, developer tooling, and enough dashboarding to make everyone feel responsibly supervised.

In other words, the cloud vendors are no longer just selling servers with nicer branding. They are selling “agent factories.” That phrase may sound like it escaped from a product manager’s dream journal, but it captures the market direction surprisingly well. Enterprises do not want one flashy AI assistant. They want fleets of specialized systems that can pull from internal knowledge, trigger workflows, write code, summarize documents, route tickets, and still remain auditable when legal shows up asking difficult questions.

AWS kept doing what AWS does best: ship fast and make it look routine

Amazon Web Services entered the week with the usual leader’s burden: everyone expects it to be enormous, reliable, and slightly less exciting than the newer AI-first narrative coming from rivals. AWS responded in classic AWS fashion by turning practicality into strategy.

Its May 26 weekly roundup highlighted two moves that matter more than the average announcement count would suggest. First, AWS made Anthropic’s Claude Opus 4 and Sonnet 4 generally available in Amazon Bedrock. That is a serious signal for enterprise AI builders who care about coding, reasoning, and agentic workflows but do not want to live in a science experiment. Bedrock’s appeal remains straightforward: access strong models inside a controlled enterprise environment, without forcing customers to build a whole AI platform from spare parts and optimism.

Second, AWS rolled out EKS Dashboard, which centralizes visibility across Kubernetes clusters in different AWS Regions and accounts. That may sound less glamorous than model launches, but it is exactly the kind of thing that wins large customers. Real cloud adoption is often decided by boring-sounding operational tools that quietly save engineering teams from chaos at 2:13 a.m.

The catch for AWS is that investors have become impatient. Earlier in May, Reuters reported AWS revenue growth of 16.9% for the quarter, reaching $29.27 billion, but still missing elevated expectations after Microsoft posted a strong Azure result. So AWS now has two jobs: remain the default enterprise cloud and prove it can look like an AI growth story, not just the dependable adult in the room. That is a harder balancing act than it sounds.

Microsoft turned Build into a full-scale campaign for the agent era

If AWS was the calm operator, Microsoft was the keynote machine with espresso in its bloodstream. Build 2025 made one message impossible to miss: Microsoft wants Azure to be the place where enterprise AI agents are built, observed, governed, and deployed at scale.

The centerpiece was the general availability of Azure AI Foundry Agent Service. Microsoft also pushed broader support for Model Context Protocol, deeper multi-agent orchestration, and a growing toolset for evaluation, monitoring, tracing, and model routing. That matters because enterprise buyers are becoming allergic to raw model hype. They want proof that a system can be tested, measured, and rolled back before it torches a workflow or invents a policy that legal never approved.

Azure AI Foundry’s growing observability layer is especially important. Monitoring quality, safety, cost, and performance in one place is not just a “nice to have.” It is the difference between an AI pilot and an AI program. Microsoft seems to understand that the real competition is no longer just who has the smartest model. It is who can turn model chaos into enterprise process.

Microsoft also expanded the model buffet. The Verge reported that the company added xAI’s Grok 3 models to Azure AI Foundry, which reinforces Microsoft’s increasingly pragmatic stance: Azure wants to host the winners, not just promote one house brand. That strategy makes Azure look less like a single-vendor AI bet and more like a neutral platform for enterprises that want options, even messy ones.

The bigger takeaway from Build was this: Microsoft is trying to transform Azure from cloud platform into control center. Not merely where workloads run, but where AI systems are managed like a modern software supply chain. That is smart positioning, and it is directly aimed at the concerns CIOs actually have.

Google pushed Vertex AI from model catalog to serious enterprise platform

Google’s May story had the flashy energy of I/O, but the cloud implications were more disciplined than the stage lights suggested. The company used its broader AI momentum to strengthen the enterprise pitch for Google Cloud and Vertex AI.

At I/O 2025, Google highlighted expanded Gemini 2.5 Flash and Pro capabilities for enterprises on Vertex AI, including thought summaries, richer developer controls, and broader support for production use. It also introduced the next wave of generative AI media models on Vertex AI, pushing the platform beyond text and into multimodal content workflows. Translation: Google wants Vertex AI to feel less like an API shelf and more like a complete studio for modern AI applications.

Google also brought Anthropic’s Claude Opus 4 and Sonnet 4 to Vertex AI, with support for Agent Development Kit, Agent Engine, and BigQuery ML integrations. That is important for two reasons. First, it shows Google is willing to win enterprise cloud share even when the model is not its own. Second, it reinforces a broader industry truth: cloud providers increasingly compete on orchestration, security, data access, and workflow integration just as much as they do on foundation models.

Perhaps the most striking signal came from Google’s own scale metrics. By I/O, the company said Gemini usage on Vertex AI had jumped dramatically year over year, while overall token processing across products and APIs had exploded. Those numbers are not just bragging rights. They suggest Google is turning AI demand into cloud consumption, which is exactly the flywheel it needs to close the gap with AWS and Azure.

In practical terms, Google’s pitch is becoming clearer: if your future involves multimodal AI, open tooling, strong developer ergonomics, and deep data integration, Vertex AI is no longer the “interesting alternative.” It is a real contender.

IBM and Oracle reminded everyone that hybrid cloud is not dead, just less photogenic

Not every enterprise is racing to put its whole future into a single hyperscaler’s glossy AI console. That is why IBM and Oracle still matter, and during this cycle both companies leaned hard into the idea that hybrid and multicloud are where the adult decisions happen.

IBM’s case: practical AI needs better plumbing

At IBM Think 2025, the company emphasized hybrid capabilities designed to help businesses build and deploy AI agents with enterprise data. The headline was not pure spectacle. It was integration. IBM introduced webMethods Hybrid Integration to support agent-driven automation across apps, APIs, file transfers, events, and B2B environments. That is deeply unglamorous in the best possible way. Enterprises run on messy connections, and IBM’s strength has always been its willingness to meet that mess where it lives.

IBM also advanced watsonx.data, aiming to unify and govern structured and unstructured information across silos and clouds. That matters because AI projects usually fail less from lack of model intelligence and more from bad data access, weak governance, and disconnected systems. VentureBeat’s framing was apt: enterprise AI is moving from experimentation to implementation, and IBM wants to be the company that helps businesses survive that move without lighting their architecture on fire.

Oracle’s case: spend big, connect everything, become impossible to ignore

Oracle’s cloud strategy in late May looked less like subtle positioning and more like a giant neon sign reading: “We intend to be in this race.” Reuters reported that Oracle was set to spend around $40 billion on Nvidia chips for OpenAI’s new U.S. data center in Texas, a jaw-dropping figure that underscores just how expensive the AI infrastructure war has become.

Even if Oracle is still smaller than the big three in cloud market share, it is using multicloud partnerships, database gravity, and raw infrastructure ambition to punch above its historical weight. Its expanding partnership with IBM to bring watsonx capabilities to Oracle Cloud Infrastructure fits this playbook neatly. Oracle is positioning OCI as a serious venue for enterprise AI and multi-agent workloads, especially for organizations that care more about where their data lives and how it performs than about which keynote got the loudest applause.

The market backdrop: cloud spending is still climbing, but the story has changed

The broader market numbers explain why every cloud vendor suddenly sounds like an AI systems integrator. Cloud spending remains large and growing. Industry reporting on Q1 2025 showed global enterprise spending on cloud infrastructure services reached roughly $94 billion, with the top three vendors controlling about 63% of that market. AWS remained the leader, while Microsoft and Google kept gaining ground.

But the more interesting change is not who sits in first place. It is what customers are buying. The new spending is increasingly tied to AI workloads, model access, inference economics, data pipelines, and governance layers. A few years ago, cloud competition centered on compute, storage, and migration. Now the buying conversation sounds more like this: Which platform helps me deploy AI agents safely, connect to enterprise data, monitor output quality, and avoid becoming permanently trapped in one vendor’s weirdly named ecosystem?

That shift has consequences. It means Kubernetes visibility tools can matter as much as model launches. It means observability is becoming a selling point, not an afterthought. It means hybrid cloud vendors still have leverage. And it means the next phase of cloud competition may be won not by the loudest AI marketing, but by the provider that makes AI systems boring enough to trust.

What business leaders should take away from this week

First, multicloud is no longer just a procurement slogan. It is becoming a practical design choice for AI-era architectures. Models are spreading across platforms, and enterprises increasingly want flexibility without rewriting their whole stack every quarter.

Second, the winners will be the platforms that combine model access with governance, observability, and data integration. Smart outputs are nice. Auditable smart outputs are what get funded.

Third, the cloud market is entering a more mature AI phase. The novelty is fading. The hard questions are arriving. How do you evaluate agents? How do you control cost? How do you connect systems without increasing risk? How do you stop “innovation” from becoming a synonym for “we’ll fix it after production”? The vendors that answer those questions well will shape the next year of enterprise cloud spending.

Conclusion: the cloud week that felt like a preview of the next two years

The week of May 26 did not rewrite cloud computing in one dramatic moment. It did something more important. It showed how the market is being reassembled in real time. AWS doubled down on practical enterprise AI delivery. Microsoft pushed hard on agent platforms and operational controls. Google sharpened Vertex AI into a stronger enterprise bet. IBM made the case that hybrid data realities still rule the corporate world. Oracle reminded everyone that infrastructure scale and multicloud ambition can still change the conversation.

Put it all together, and the message is clear: cloud is entering its agent-first era. The next battle will not be won by whoever shouts “AI” the most times per keynote. It will be won by whoever makes AI useful, governable, connected, and profitable. Which is less glamorous than science fiction, sure. But for actual businesses, it is much better.

Experience from the field: what a week like this feels like for real cloud teams

If you work close to cloud strategy, platform engineering, DevOps, security, FinOps, or enterprise architecture, a week like May 26 feels both exciting and slightly ridiculous. Exciting because the tools are getting better. Ridiculous because every vendor seems convinced your team has unlimited time, unlimited budget, and a deep desire to learn six new acronyms before lunch.

In practice, weeks like this create a very familiar rhythm inside real companies. The executive team sees the headlines and asks whether the organization now has an “AI agent strategy.” The product team wants to test new capabilities immediately. The engineering team opens the documentation, squints at the pricing page, and quietly wonders who exactly will maintain all of this in production. Security asks whether the shiny new feature can be audited. Compliance asks where the data flows. Finance asks why every innovation seems to come with a token bill. The answer, naturally, is “it depends,” which nobody enjoys hearing.

There is also a genuine emotional shift happening in cloud teams right now. A couple of years ago, cloud roadmaps were dominated by migration, modernization, containerization, and cost optimization. Those things still matter, but now they are joined by an entirely new layer of questions: Which models should we support? How do we compare outputs? What counts as acceptable latency for an agent? When does retrieval help, and when does it just give the AI more opportunities to be confidently strange?

The lived experience of this moment is not one giant breakthrough. It is constant translation. Teams are translating executive urgency into technical reality. They are translating vendor claims into proof-of-concept work. They are translating old governance frameworks into new AI-specific guardrails. And they are translating between clouds, because almost nobody operates in a neat, one-platform fairy tale.

Another real-world truth is that operational maturity is becoming a competitive advantage inside companies. The teams that already know how to manage Kubernetes visibility, identity controls, observability pipelines, logging, data lineage, and change management are in a much better position to adopt AI responsibly. The teams that skipped those basics are discovering that “agentic transformation” is less magical when nobody can explain what a system touched, why it made a decision, or how much the last experiment cost.

Still, there is reason for optimism. The tools are slowly becoming more practical. Multi-agent orchestration is becoming less theoretical. Observability is getting better. Hybrid integration is improving. Model choice is expanding. The best part of a week like this is not the hype cycle. It is the realization that cloud platforms are finally being forced to solve the gritty problems that enterprises have been yelling about for years.

So the experience of “The Week In Cloud: May 26” is this: equal parts acceleration and reality check. You leave the week impressed by the innovation, cautious about the complexity, and very aware that the future of cloud will not be decided by who has the prettiest demo. It will be decided by which platform helps real teams build systems that work on ordinary Tuesdays, under normal budgets, with human beings who still need sleep.