$74,000 Just to Download Your Own Data?!

Private cloud is back, but this is not a return to the old enterprise data center model of slow provisioning, ticket queues, static infrastructure, and underused hardware sitting in corporate facilities. In this episode of Cloud Computing Insider, David Linthicum looks at why enterprises are reconsidering private cloud, hybrid cloud, and cloud-like infrastructure inside the business. The drivers are very different from the private cloud conversation of 10 or 15 years ago. VMware uncertainty after the Broadcom acquisition has forced many enterprises to rethink their virtualization and private infrastructure strategies. OpenStack is getting renewed attention as companies look for more open and flexible private cloud options. AI workloads are pushing infrastructure closer to enterprise data because of latency, cost, GPU availability, data sovereignty, and compliance concerns. Edge computing is also forcing companies to run cloud-like services outside centralized public cloud regions. This does not mean public cloud is going away. It means the old "cloud-first for everything" assumption is being replaced by a more mature workload-placement model. The new private cloud is not about nostalgia. It is about API-driven infrastructure, automation, self-service, cloud-native operations, Kubernetes, platform engineering, on-prem AI, distributed edge environments, and better control over economics and risk. The mistake would be thinking private cloud is just the old data center with a new label. The opportunity is building internal infrastructure that behaves like cloud where it matters, while avoiding the runaway cost, lock-in, latency, sovereignty, and operational tradeoffs that come with putting every workload in public cloud.  

I'm reacting to reports that major cloud providers have tried to keep Insider's Guide to Cloud Computing out of internal libraries, partner events, and industry conversations. If true, that reaction says more about the industry's discomfort with scrutiny than it says about my book. This video is not an attack on cloud computing. I've spent decades building, teaching, and advising enterprises on cloud architecture. It is an attack on the propaganda that says cloud is always cheaper, automatically more secure, endlessly reliable, and right for every workload.   I break down the costs vendors too often minimize: egress fees, API consumption, managed-service sprawl, overprovisioning, and lock-in that can erode negotiating power. I also address the security and reliability risks that do not disappear when a workload moves to a hyperscaler. The goal is simple: give enterprise leaders the facts they need to make cloud decisions based on business value, architecture, resilience, and transparent economics—not keynote promises. If you are responsible for technology strategy, cloud budgets, security, or modernization, this conversation matters. Read the book, challenge the assumptions, and never let the company sending the invoice be your only source of architecture advice. The stakes are too high for blind trust alone.  

Nvidia GPUs have become the default engine of the AI boom, but AWS is making a very different bet: if customers are building and running large AI models inside Amazon's cloud, Amazon wants more of that workload running on Amazon silicon. In this episode of Cloud Computing Insider, David Linthicum examines AWS Trainium, Amazon's purpose-built AI accelerator for model training and inference, and why it matters for enterprise cloud strategy. AWS says Trainium is designed to deliver strong economics for high-performance AI training and inference at scale, while Trn2 instances offer 30–40% better price performance than comparable GPU-based EC2 P5e and P5en instances. That does not mean Nvidia is going away. AWS still offers a large portfolio of Nvidia-powered instances, and Nvidia remains deeply embedded in AI development through CUDA, GPUs, networking, software, and developer mindshare. The real issue is control. If AWS can move more AI workloads onto Trainium, it can reduce dependence on scarce Nvidia GPUs, improve cloud margins, offer differentiated pricing, and keep customers deeper inside the AWS ecosystem. For enterprises, Trainium is not just a chip story. It is a cloud architecture, cost management, vendor lock-in, performance portability, and AI infrastructure strategy story.

I've spent decades watching the cloud industry mature, and here's the uncomfortable truth: the biggest cloud conferences are not neutral industry events. They are controlled stages owned by the same hyperscalers that dominate the market: AWS, Microsoft, and Google. These companies decide who gets invited, who gets approved to speak, who gets to demo, and which messages can reach thousands of customers, partners, analysts, and enterprise decision-makers.   That matters because these events shape how the market thinks about cloud strategy, AI, modernization, security, and cost. When the agenda is controlled by the vendor, the story naturally favors the vendor. You hear the success stories, the polished demos, the customer wins, and the product launches. What you hear far less about are the cost overruns, failed migrations, lock-in, complexity, skills gaps, and architectural mistakes that real enterprises deal with every day. I'm not saying hyperscalers should not promote their platforms. Of course they will. But we should stop pretending these blockbuster conferences are open forums for independent debate. They are curated narratives, and if your perspective challenges that narrative, you may never get near the microphone.  

Not every cloud product changed enterprise computing. Many were useful. Some were overhyped. A few were absorbed into larger platforms and forgotten. But a small number of cloud technologies fundamentally changed how enterprises buy, build, run, scale, and govern systems. In this episode of Cloud Computing Insider, David Linthicum breaks down five cloud products and technology categories that actually mattered: Microsoft Azure, cloud data and analytics platforms, serverless computing, Kubernetes and managed container platforms, and the original AWS building blocks, EC2 and S3. These technologies changed enterprise computing because they reduced real constraints. They helped companies avoid long hardware procurement cycles, standardize platforms, reduce server management, break down data silos, and modernize legacy integration patterns. This is not a list of the newest or flashiest cloud services. It is a list of the cloud technologies that changed enterprise behavior. They reshaped IT operating models, architecture decisions, application deployment, data strategy, governance, and the economics of infrastructure. The cloud products that mattered most were not just products. They became patterns that every major enterprise technology organization had to respond to.

The cloud was supposed to be infinite — but AI may have finally exposed its limits. In this video, we break down why Amazon Web Services is growing at a massive pace, yet still struggling to keep up with demand from AI companies and enterprise customers. AWS revenue is surging, backlog is exploding, and Amazon is planning enormous infrastructure spending, but even that may not be enough to satisfy demand through 2026 and 2027.  This story is bigger than one earnings report. It shows how artificial intelligence is changing the economics of cloud computing. The new bottlenecks are not just software or algorithms — they are power grids, data centers, GPUs, memory, cooling systems, networking gear, and construction timelines. Cloud capacity is becoming a scarce resource, and companies that reserve compute early may gain a serious advantage over competitors who wait too long.  We'll look at why AWS is still winning financially, why that success creates new pressure, and why the future of AI may depend less on smarter models and more on who can build enough physical infrastructure fast enough.

For two decades, enterprise architects fought tooth and nail against proprietary APIs, databases, and multi-year cloud contracts to keep their systems portable. We thought containers, microservices, and Kubernetes had finally solved the vendor lock-in problem for good. But while everyone was celebrating software portability, the major cloud providers quietly pulled off the ultimate architectural bait-and-switch: they moved the lock-in down into the physical silicon. In this video, I break down why custom cloud chips—like AWS Graviton and Trainium, Google TPUs and Axion, and Microsoft Azure's Maia and Cobalt—are rewriting the rules of enterprise IT. While these purpose-built processors offer unbeatable price-to-performance and massive runtime cost cuts on paper, they come with a hidden operational tax. Once your core AI models, compilation pipelines, and runtime frameworks are hyper-tuned to execute on a specific vendor's proprietary instruction set, you can't just containerize your workload and shift it elsewhere. The hardware moat is real, and it's turning standard cloud migrations into multi-million-dollar re-engineering nightmares. Watch this breakdown before you migrate your mission-critical applications to custom cloud hardware, and learn how to navigate performance gains without falling into the silicon trap.

Technology has never been more powerful, more available, or more affordable to deploy. Cloud computing, SaaS, DevOps, automation, cybersecurity platforms, data tools, and now AI were all supposed to make IT faster, cheaper, and more valuable to the business. But the numbers tell a more uncomfortable story: organizations are spending more on technology than ever, while productivity gains and business outcomes are not keeping pace.  In this conversation, David Linthicum breaks down why IT may actually be getting worse at its core job: turning technology dollars into measurable business value. The issue is not that the tools are failing. The issue is that IT organizations have become buried in complexity, technical debt, poor architecture, bloated processes, cloud waste, and metrics that reward activity instead of outcomes.  If global IT spending keeps climbing while productivity growth slows, then leaders need to ask a harder question: are we really becoming more efficient, or are we just buying more technology to work around the problems we refuse to fix?  This episode challenges the conventional wisdom around digital transformation and asks whether modern IT has lost the plot.    

Moving a petabyte out of the public cloud isn't a technical problem — it's a money problem. Uploading is free on every major provider, but downloading that same data back to your own servers triggers egress fees that can run into the tens of thousands of dollars. In this video, I break down exactly what it costs to pull one petabyte from AWS, Azure, and Google Cloud back to on-premises systems, comparing the open internet against private circuits like Direct Connect, ExpressRoute, and Cloud Interconnect. I walk through the real math, the hidden fees nobody mentions, the time each path takes, and the trade-offs around security, governance, and bandwidth. If you're planning a cloud migration, a repatriation, or just want to understand why the cloud keeps your data sticky, this is the breakdown you need before you sign anything — or before you hit "download" on a petabyte.

AI data centers are getting so massive and power-hungry that some companies are looking beyond Earth for the next cloud computing frontier. In this video, we look at Starcloud, a startup trying to build data centers in space using satellites packed with AI hardware, solar panels, radiators, and high-speed communications. The idea sounds like science fiction, but Starcloud has already launched an Nvidia H100 GPU into orbit and raised major funding to scale its vision. The pitch is simple: Earth-based data centers are running into serious limits, including power shortages, land constraints, cooling costs, grid delays, water usage, and local permitting battles. Space could offer constant solar energy, natural cooling through radiation, and room to build compute infrastructure at a scale that is difficult on the ground. But the challenges are enormous. Launch costs, radiation, orbital debris, latency, regulation, hardware reliability, and connecting thousands of GPUs in orbit could make this one of the hardest infrastructure bets in tech. If it works, the future of cloud computing may not be in Virginia, Texas, or Silicon Valley — it may be in low Earth orbit.