For years, “cloud first” was treated as an unquestioned mandate – move everything to the public cloud, and agility, cost savings, and simplified operations would follow. But as enterprises have learned through hard experience with refactoring costs, unpredictable bills, and growing data control concerns, that one-size-fits-all philosophy hasn’t held up. Today’s IT leaders are asking a more pointed question: not “how do we get to the cloud,” but “where should this specific workload actually run?” That shift, from ideology to outcomes, is reshaping how organizations think about virtualization, private cloud, bare metal, and public cloud as complementary tools rather than competing choices.
To unpack what’s driving this shift and what it means for enterprises navigating AI infrastructure demands, networking costs, geopolitical risk, and the ongoing tension between flexibility and lock-in, VMblog spoke with Jeffrey Gregor, General Manager of OVHcloud US. In this Q&A, Gregor shares his perspective on why capacity alone won’t define the next generation of cloud providers, how bare metal fits into a modern hybrid strategy, and why the organizations succeeding today are the ones designing for portability from day one.
++
VMblog: Enterprise cloud strategies seem to be shifting from “cloud first” to “workload first.” What is driving that change, and what should IT leaders be rethinking as a result?
Jeffrey Gregor: For a long time, “cloud first” was treated almost like a philosophy. The assumption was that moving everything to the public cloud would automatically make organizations more agile and help them lower costs and simplify operations. In reality, it’s become more complicated.
Over the past several years, the industry has spent a lot of time debating cloud repatriation. The more interesting story isn’t that organizations are abandoning the cloud. It’s that they’re becoming more selective about which workloads belong in which environments.
Many organizations discovered that moving existing applications to a public cloud required significant refactoring before they could benefit from cloud-native architectures. This happened at the same time that forecasting costs became more difficult, and concerns around data control and operational complexity also started to grow.
All of that has moved the conversation toward business outcomes rather than cloud ideology. Instead of asking, “How do we move everything to the cloud?” organizations are rightly asking, “Where should this workload run to achieve the best outcome?”
For some applications, that’s a public cloud. Others make more sense in a private cloud or should remain on premises for compliance, performance, or operational reasons. Virtualization gives organizations the flexibility to put workloads where they have the right control over their data, where they understand the costs, and where they deliver the best business outcome instead of forcing every application into a single operating model.
Organizations are also paying closer attention to geopolitical risk. We’ve already seen governments restrict access to certain AI services over national security concerns. For instance, the U.S. government restricted the deployment of Anthropic’s Mythos and OpenAI’s frontier models over cybersecurity fears, while nations like China and Russia completely block foreign services like ChatGPT to maintain strict digital sovereignty.
Whether those restrictions are temporary or permanent isn’t really the point. Companies need to understand where their data resides, who controls it, and how quickly they could adapt if circumstances change.
VMblog: As organizations reassess virtualization, private cloud, and public cloud options, what mistakes do you see them making when they evaluate where workloads should live?
Gregor: One of the biggest mistakes is assuming every workload has to become cloud native immediately.
The advantage of virtualization is that workloads can continue to work like they had in the past while organizations modernize at a pace that makes sense for the business. Instead of treating migration as one massive project, they can move existing virtual machines into a private cloud and modernize applications over time.
VMware continues to be the gold standard for virtualization, and there are a lot of organizations that have built years of operational knowledge around it. There are alternatives in the market, but companies sometimes discover months later that they’ve traded familiar capabilities for a solution that requires much more effort to deliver the same results.
Ultimately, the goal isn’t to force every workload into one environment. It’s to give each workload the environment where it performs best while maintaining the flexibility to move it as business requirements, cost, or compliance needs change.
VMblog: Cloud cost conversations have moved beyond compute and storage. Why has data movement become such an important part of the cloud economics discussion?
Gregor: If you’re not moving data around, you’re not doing much useful work. That’s why networking has become such an important part of the cloud economics discussion.
Five years ago, cloud cost optimization was largely about compute utilization. Today, FinOps has become an established discipline because organizations recognize that cloud economics are more complicated than simply purchasing compute and storage.
The challenge is that networking costs are often the hardest to estimate. Applications are constantly moving data between users, services, regions, and systems, and those costs can become a significant part of the monthly bill.
This is particularly true when providers charge for moving data between environments or for taking data back out of their cloud. Egress fees have become a real consideration because they affect both operating costs and long-term flexibility. We’ve also seen providers charge for moving data within their own environments, which many customers don’t anticipate when they’re planning workloads.
That’s why people are often surprised when they get their cloud bills. Organizations need to understand not only their compute and storage costs, but also how their applications communicate, how much data they move, and whether those networking costs will remain predictable over time.
VMblog: Many companies are trying to avoid lock-in without adding more operational complexity. What does real cloud flexibility look like in practice?
Gregor: Real flexibility means having choices without creating more work for your operations team.
Most organizations have already realized they don’t want to put all their eggs in one basket. Different workloads have different requirements, and that’s why hybrid and multi-cloud strategies have become the norm rather than the exception.
Virtualization plays an important role because it makes workloads portable. When workloads can move between on premises infrastructure, private cloud, and public cloud, organizations have much more freedom to respond to changing business requirements, performance needs, or economics.
Lock-in doesn’t just come from technology. It can also come from contracts, pricing models, and data movement costs. If moving your workloads or your data becomes prohibitively expensive, which it absolutely can, your options become much more limited.
The organizations that are succeeding are designing for portability from the beginning rather than assuming they’ll stay in the same environment forever.
VMblog: AI has created enormous infrastructure demand, but the economics are still uncertain. How should enterprises approach AI infrastructure without overcommitting too early?
Gregor: AI is moving so quickly that it’s difficult to justify building everything on premises.
Every major hyperscaler is investing tens of billions of dollars in AI infrastructure, yet most enterprises don’t have the scale or the business case to make those same bets. That doesn’t mean they should try to. It means they need an infrastructure strategy that gives them the flexibility to take advantage of new technologies as they become available.
GPU technology continues to evolve rapidly, hardware remains expensive, and demand often exceeds supply. Making large capital investments doesn’t necessarily mean you’re buying the right infrastructure for what your business will need in the future.
Cloud gives organizations much more flexibility. They can take advantage of new hardware as it becomes available without having to refresh their own infrastructure every time a new generation is introduced. For most organizations, cloud is the place to run AI workloads because it gives them the flexibility to adapt as both the technology and the economics continue to evolve.
At the same time, organizations still need to maintain control over sensitive data. They need to know where their data is, who controls it, and how it’s being protected. That’s why many are looking at private cloud environments where they can run their own models, protect their intellectual property, and take advantage of cloud-scale infrastructure without giving up visibility or control.
VMblog: Many organizations are discovering that not every workload belongs on shared public cloud infrastructure. How should IT leaders think about the role of bare metal as part of a modern hybrid cloud strategy?
Gregor: Bare metal gives organizations a practical way to start taking advantage of the cloud without having to redesign every application first. This is really the lift-and-shift opportunity. You can take what you’re already running on premises and move it to bare metal in the cloud without worrying about maintaining hardware, refreshing servers, or expanding your own data center.
From there, organizations can modernize at their own pace. They can build virtualization on top of bare metal, migrate workloads over time, and expand capacity as the business grows instead of trying to do everything at once.
We’ve seen this approach work particularly well for organizations with high-performance workloads, including blockchain, online gaming, ad tech, and healthcare. Those industries often require dedicated performance and greater control over infrastructure, but the broader principle applies across many sectors. Bare metal gives organizations another deployment option so they can choose the right infrastructure for each workload instead of forcing every application into a shared environment.
VMblog: Looking ahead, what will separate cloud providers that are truly prepared for the next phase of enterprise infrastructure from those simply adding more capacity?
Gregor: The simple answer is that capacity alone won’t be enough.
The more nuanced answer that organizations are looking for providers that give them flexibility, predictable economics, and confidence that they’ll remain in control of their data as infrastructure requirements continue to evolve.
Data sovereignty and data residency will become increasingly important as organizations navigate changing regulatory requirements and geopolitical uncertainty. AI capabilities will also matter, but so will giving customers the freedom to decide where and how those AI workloads run.
The strongest providers will support a mix of public cloud, private cloud, bare metal, and virtualization rather than assuming one model fits every application. They’ll also give customers options across different performance and pricing tiers so organizations aren’t paying for cutting-edge hardware when an existing workload doesn’t require it.
The most successful providers won’t necessarily be the ones with the biggest data centers. They’ll be those that make it easiest for customers to change their set up as their needs change. Organizations need the freedom to move workloads, adopt new technologies, and adjust their strategies as business needs evolve. They also need the flexibility to choose the right infrastructure for each workload because not every workload needs to be running on the latest and greatest hardware. That’s the kind of flexibility enterprises will increasingly expect from their cloud providers.
##
