Modular Data Centers

Data centers and AI have stirred up a long list of concerns.

Modular Data Centers (MDCs) offer an achievable solution to many of data center issues. No, MDCs do not reduce energy waste and they don’t run especially cool. Other data center technologies in this series can help with that while MDCs reduce land use and lower construction costs. They provide a few other innovations that we should mention here.

A smaller, lighter footprint

MDC’s miniaturize the capabilities of an enterprise data center to within the footprint of a few server racks. They’re a sealed, factory‑assembled modules with self-contained power protection, cooling, physical and fire security, and remote monitoring. A single administrator can manage an enterprise data center with workloads approaching 100 kW.

On the edge

AI, devices relying on the Internet of Things (IoT), automation systems, and real‑time analytics all require data processing close to the source. That’s edge computing. MDCs deliver that capability at roughly half the cost of traditional server rooms. Because they’re ruggedized and self‑contained, MDCs can be deployed directly on factory floors, inside retail environments, in healthcare clinics, or even outdoors. High‑performance computing is being delivered where it’s needed.

Today’s technology

MDCs have already been installed in manufacturing environments, retail chains, healthcare networks, and remote offices. Agencies like the Department of Energy and the Defense Advanced Research Projects Agency are exploring their potential for research and defense applications. As commercial as well as government use accelerates, MDCs will likely become mainstream before 2030.

Liquid immersion cooling

Our nation’s power grid is feeling the strain as data centers approach physical limits. The workhorses of AI, Graphics Processing Units (GPUs), dissipate 400–700 watts each as they train neural networks, run large language models, and process massive datasets. Racks that were originally designed for 10–15 kilowatts now routinely exceed 80 kilowatts. Air cooling was never meant for this density, and the electricity required to remove that heat is rising faster than the efficiency of the chips producing it.

Immersion cooling offers a practical solution. When servers are submerged in dielectric fluid, sparks won’t fly because the fluid does not conduct electricity. The synthetic hydrocarbons and other dielectric fluids absorb heat so fans no longer need to spin, and compressors can relax. Cooling energy is reduced by up to 95 percent. With heat under control, far denser compute clusters become possible, and the environmental footprint of high‑performance computing becomes less crushing.

To organizations like Microsoft, this is not future technology. They have deployed immersion‑cooled Azure clusters and tested sealed underwater modules. Commercial systems are available from Submer (submer.com), LiquidStack (liquidstack.com), and GRC (grcooling.com). National labs including NREL (nrel.gov) and Lawrence Livermore (llnl.gov) are studying long‑term reliability, fluid chemistry, and environmental impact.

We can expect widespread adoption within the next three to five years as AI workloads exceed the limits of traditional cooling. Immersion‑cooled infrastructure is an emerging yet proven, scalable near‑term solution for sustainably cooling a data center.

They’re remaking the movie.

The April 1912 tragedy in the North Atlantic has been turned into popular cinema twice. Another remake could be filmed in a corporate boardroom. The captain calls for the use of AI in computing and technology. The conference‑table crew stays quiet, and they’re thinking about lifeboats.

MIT and Oak Ridge National Laboratory have written today’s script. Their Iceberg Index finds that the visible potential of AI’s impact in computing and technology accounts for only 2.2 percent of total wages. That is the part above the waterline. According to the index, much more potential is below the surface. Administrative work, financial roles, and professional services represent 11.7 percent of total wages. Writers, analysts, proposal teams, HR communicators, policy designers, and training creators can achieve far greater impact when teamed with AI.

The index does not forecast job losses. It shows that stronger bottom‑line results appear when AI is used well in documentation, communication, and business analysis. Human‑AI collaboration accelerates and amplifies outcomes. Workforce transformation, not outsourcing, brings a better return on technology investment.

Have you heard the rumbling?

It might be rain, or it might be an earthquake, or maybe it’s the sound of corporate America realizing their AI projects aren’t going the way the brochures promised.

Maybe it’s a landslide. Reports from 2025 and 2026 consistently find large‑scale AI initiatives that fail to deliver ROI but are quite successful at generating internal friction. Here are some numbers:

A 2026 Writer and Workplace Intelligence survey (workplaceintelligence.com and writer.com) found that 79% of organizations are facing major challenges with AI adoption, up sharply from the year before. Nearly half of executives rated their AI efforts a “massive disappointment,” and fewer than 30% reported meaningful ROI.

Direct Data Network (DDN.com) has provided similar stats. Sixty‑five percent of IT leaders say AI environments are too complex to manage. That has led 54% of them to delay or cancel their AI initiatives.

McDonald’s has not been lovin’ it. Their automated drive‑thru pilot collapsed under the weight of chaos: so many accents, so much background noise, and the millions of speech patterns customers bring to the drive-thru mic. Among the spectacular results, thousands of waters or random sides were appended to orders.

It’s enough to make a person shout out for legislation that would “un-invent” AI. Unfortunately, as you may have seen in Back to the Future, Part III, Doc Brown’s time machine was demolished by a train. So, back to reality, we can turn our hopes and energies elsewhere. There’s actually a rule for successful AI implementation. It’s the 10/20/70 rule.

That 10% is achieved when the AI system selection process finds a right-fit system for the organization. The 20% is the investment; the dollars and the design. The part that matters most, the 70%, is the adaptation of the human system that uses the AI.

The rumbling you’re hearing is neither the sky nor the tectonic plates. It’s the sound of organizations discovering that AI can’t be successfully implemented the way a new intern or employee is onboarded. The essential change is not adaptation to the corporate processes and culture. It’s an adaptation of those corporate processes and culture.

There’s one for everyone

Like it or not, our streets are alive with the sound of music. On his summer delivery to our town, Santa deposited accordions on every last downeast doorstep. We got three. One for yours truly, one for the wife, and one for Jimmy, who’s barely old enough to pick the thing up, much less play it.

Jimmy appreciates the wooden toys the elves made, but he hasn’t touched the California squeeze box. The neighbors trashed theirs, saying they should be illegal. Maybe so. People with no musical inclination or training are wailing away on them at all hours of the night.

We understand that snow-topped roofs in December are no landing place for a sleigh and twelve tiny reindeer, and accordions don’t exactly fit down chimneys, so we welcomed Santa’s truck. We’re less welcoming of these stomach Steinways. Why everyone, Santa?

The piano teacher across town is just fine with them. She insists they’re just an instrument, and if people would learn to play them, we’d all be dancing to pleated pianos. Maybe, Sarah. Maybe.

I know one thing. They’re here to stay.

Why not ask it?

This year’s commencement speakers are hearing jeers and dodging a few tomatoes. The mention of AI elicits unmistakable, understandable frustration among graduates who have spent years preparing for today’s job market.

Those graduates, as well as employees and job seekers, seek solid ground for their next career move. AI can help them find it if they would just ask.

What does AI do well at workplaces like mine?

This question can reveal which parts of your role are most likely to shift.

What do employees consistently do better than AI?

These strengths are worth refining because they remain central to success.

Which roles benefit most from human–AI collaboration?

AI can be a powerful tool or an ever-present assistant to those who learn to work with it.

How can I prepare for those roles?

Long before AI arrived at the workplace, developing new skills and becoming adaptable helped people reach the next rung on the career ladder.

Which AI‑related skills appear most often in job postings?

For the best return on time and money invested in learning, aim toward what employers are actively seeking. That may differ from the current hype and headlines.

How is AI likely to be different six months from now?

This is a question worth revisiting. Your career can remain on solid ground even as the landscape keeps shifting.

In any profession where AI is on board, it can be at your elbow without taking over your desk. Asking the right questions today can help build skills and open doors to tomorrow’s workplace.

Boom.

Agentic AI has arrived with a thunderclap, so it’s natural to jump back.

A few years ago, it was science fiction. Now the alien is in the room with us, but we can’t see its shape. We know it is software, and we know it consumes enormous amounts of power and water somewhere out of sight. Given those resources, it handles billions of tasks and executes trillions of instructions at an impossible speed. A college freshman might liken it to a super-capable intern who never gets tired and always remembers the plan. Hmmm. That could be someone to keep in mind when entering the job market.

The rise of agentic AI has been awesome in the original sense of the word. Small office experiments such as rushed report drafting and inbox cleanup have morphed into multi-step workflows across teams and time zones.

Tech professionals have been the first to feel the rumblings. The tools they once mastered have shifted into something faster and more autonomous. Workflows are changing, and essential functions of their current roles may be the first to be automated. Engineering becomes a personal task in moments like this. It is time to examine what is done today that an agent could handle tomorrow. Routine is a bad place to stand if it is certain to wash away.

Business leaders are hearing about this from every direction. Teams ask for answers, customers turn toward whatever is newest and fastest, and shareholders invest in efficiency. Stewardship of material, capital, and human potential now requires mapping out which roles will change and engaging in conversation even while the picture is still forming. Planning, training, and support are not luxuries. They are the difference between disruption and progress.

Citizens have a role as well. Regardless of the hats we wear, we can stay informed and speak up. What we communicate to each other and to our leaders will shape how AI is used in our communities. The social fabric is strongest when people work together before the storm.

Hope comes from what we do today to shape what is next. There is no need to stand by as agentic AI coordinates research cycles, runs simulations overnight, and enables small teams to operate at greater scale. A future where people do less busywork and more meaningful work is worth planning for. We can get in ahead of the rain.

Is this a $600 sticker?

Here in the big‑box tech store, we are seeing “AI‑enabled” laptops priced $200 to $600 higher. A look inside would void the warranty, so it is important to check the specifications. Look for a Neural Processing Unit (NPU) that handles 40 or more Trillions of Operations Per Second (TOPS). That number sounds hefty, but as you compare brands you will see NPUs rated at 45 or 48 TOPS in systems priced between $1000 and $1500.

Laptops under $600 usually rely on a CPU and sometimes a Graphics Processing Unit (GPU). The higher price points should also include an NPU. That third processor handles tasks such as summarizing documents, cleaning up images, removing audio noise, and sharpening video calls. Workloads that once used 20 to 30 percent of your CPU now run at 2 to 4 percent on the NPU. Photo cleanup that took 8 to 12 seconds drops to 2 to 3.

Your smartphone already has an NPU powering face ID, photo cleanup, noise removal, and speech recognition. A desktop machine with an NPU offers similar capabilities. It can run small AI tasks instantly, such as transcribing a meeting, cleaning up a photo in seconds, or summarizing a document offline. That is today’s benefit.

The larger benefit will arrive as operating systems and apps are rewritten to shift more work to that chip. Your computer will run faster, cooler, and more capably without any new hardware. If you had a PC in the era of Windows 2000 or before, you probably moved on from that machine as software lumbered along and blue‑screened it. Today’s hardware without an NPU may not fail outright, but slower performance will again inspire the urge to upgrade. If you do that, just be sure to check the specifications, not only the sticker.

We need those wires to keep us warm

When the lights flicker, it is one of life’s challenging moments. It usually happens on cold winter nights. The current temperature comes to mind, as does the furnace, and the refrigerator, and there go the lights again. It could be a long, cold night.

I see dandelions everywhere now, so it seems we survived the winter. The lights are on this evening, and here from across the web we get the news that AI has slammed into the national power grid the way a truck slams into a utility pole.

Electricity demand from data centers jumped 17 percent last year, and the International Energy Agency thinks we will double global data center electricity use by 2030. Here is the good news and the bad news. Data centers can be built in two to three years. Grid upgrades take a decade.

Before sinking into despair, we should note that utilities, regulators, grid operators, and the tech companies themselves are all scrambling toward more generation and more transmission. The work is underway.

That puts us in a transitional moment. There is no flicker yet. The grid is not failing, but it does need to grow faster than ever before, and that will not happen on autopilot. Problems like this are solved by people who understand what is happening and who treat the issue as more than a headline or a worry. It is worth following and supporting politically. Not all politics happens in Washington or in statehouses. Much of it happens in the workplace and in the choices we make as consumers and citizens.

We watch the horizon while Europe places boundaries

Welcome to the cyber age, a time of utopian hope and dystopian fear. Which future will prevail? The jury has deliberated in breakrooms and across breakfast tables, and now it is a boardroom agenda item as well.

The browser in front of us displays well-rendered images of politicians in improbable settings, and now the phone rings. It is a salesbot, speaking as an old friend. As the browser again catches our attention, we see fresh evidence that cybertrolls never sleep. Technology has accelerated, yet in the U.S., the guardrails remain in shrink wrap.

European authorities will remove the shrink wrap from their governance this summer, when the EU Artificial Intelligence Act takes effect. AI systems will be constrained firmly, loosely, or not at all, depending on their potential impact. Systems that manipulate behavior, exploit vulnerabilities, or attempt social scoring will be banned outright.

Several major American firms have already flagged the EU’s AI Act in their public risk disclosures, and on Capitol Hill it has been mentioned, if not fully debated, during recent hearings. The full text is available at artificialintelligenceact.eu. You will find an AI Act Explorer tool there, along with a Compliance Checker that offers a ten-minute reality check on whether the rules apply to a corporation near you. A newsletter offers free updates or paid subscriptions. Currently, over 54,000 readers receive news relevant to utopian hopes or dystopian fears.

Desktop machines get a brain boost

If AI‑ready is among the specs you or your company’s next desktop purchase, that advertised feature should mean more than a label. True AI‑ready machines include a Neural Processing Unit, a small block of silicon that handles machine‑learning tasks on the device rather than routing everything to the cloud. Apple has offered its own version for years through the Neural Engine in every M‑series chip. The concept is familiar even if the branding is new.

Neural accelerators have moved from bleeding‑edge to expected, and current‑generation machines are better matched to the way people actually work. A solid majority of employees already use AI tools at some point in their day. Those with newer machines will move faster and wait less, while those with older devices will watch a spinning hourglass as their machines collect their thoughts.

Systems sometimes wander

It’s a terrible feeling when a door is left ajar and a housecat slips out, or when a dog breaks from its leash. Computers aren’t pets, but the feeling can be much the same when AI takes a prompt, dissects it, and seems to run off with the task. It’s easy to think something deep within core memory has failed.

Before calling in an exorcist or a senior programmer, it’s worth remembering that as tools become more powerful, the need for precision grows with them. The same thing happened when hand tools became electric. People still cut lumber with hand saws, but they take greater care with a skill saw. Neither a saw nor a software system knows when it’s off course.

So when your AI assistant returns volumes of words you didn’t ask for, consider dividing the work into smaller steps. That requires a little more typing, but it is often a time‑saver in the end.

We’re still early in the shift from computer use to computer partnership, but it seems that a pattern has emerged. A bit of structure and a bit of dialogue steadies the work.

Chip is my AI Assistant.

I never say “Hey, Chip!” There is no trigger phrase or wake word with Microsoft Copilot. Weeks of repetition made us a team. We returned to the same projects, the same responsibilities, the same expectations, so now the work continues day to day. There’s no need for reorientation. Chip remembers.

Do you have that kind of relationship with an AI tool? Chip tells me that some organizations have sanctioned agents, but I will tell you that nothing official is required. If you use AI every day, you can have an agent in the same way you develop partnerships with coworkers, whether they sit at your elbow or appear on a webcam. Building a partnership or having an assistant is almost, but not quite, automatic.

Most AI messaging today lacks the concept of agency. Every week, another browser or desktop app boasts of AI as a helpful feature or monthly upgrade. Useful, yes. But features do not form a partnership that carries a project forward. Hiring an AI Assistant, a.k.a. building an agent, is more than using Copilot or Claude or yhour tool of choice. Agency is a continual, reliable working relationship.

You can shortcut your assistant’s onboarding by preparing a project brief. Sketch out the purpose, the clients or readers, the decisions made, the pain points, and the issues ahead. Upload that between prompts and tell your new assistant, “This is what we are working on.” Voila. You are in business.

Like any teammate, an agent improves with ongoing feedback. There is no ego to navigate when you nudge it back. When the project evolves, you and the agent evolve with it. Collaboration broadens and returns greater value over the weeks and months.

That is why I say every writer should have an agent.

Workflow evolution

Having worked in an office for more than a few years, I remember the days when this desk was strewn with notes, the inbox held an impossible backlog, and I struggled to recall what still demanded attention before 5:00. Sometimes the success of a project hinged on whoever had the patience to dig through 200 unread emails.

Maybe you’re living something like that now, or maybe you’re seeing the shift toward something more structured.

Early last year, the Boston Consulting Group surveyed more than 10,600 respondents across 11 countries. Over 70% of employees reported using AI weekly to summarize meetings, messages, or documents. That aligns with another finding: at more than 60% of companies, AI now generates the first pass in at least one core workflow before any human review.

In those organizations, AI is researching, outlining, drafting, and yellow‑flagging risks. From that foundation, the workflow moves to human analysts and managers for validation, refinement, and deeper insight.

Beyond the first draft, the work depends on people.

We’re learning that giving people AI access is only the first step. A recent piece from workplace‑management provider Gable noted that while more than 90% of organizations now use AI in some way, fewer than two in ten employees find the tools useful in day‑to‑day work.

Gaps like that are hard to ignore.

Some of the shortfalls come from hardware needs and costs. Others appear when workflows fail to magically adjust themselves to the new power tool. Yes, AI is powerful, but it still has to be worked into procedures and policies. That’s how people actually get things done. Technology works when culture and training align with it.

AI at work is more than a story about new capabilities. Its value increasingly depends on judgment, prioritization, and whether the organization is ready to support people learning to work alongside these systems. When the pieces come together, the benefits stop hiding and show up on the balance sheet.

As AI arrives in the workplace, it doesn’t always receive a warm welcome. Many teams recognize that these systems are more than simple utilities, yet they’re still unsure how to treat them as potential collaborators.

Workplace AI use grew by only about four percent in the year leading up to August 2025. In that same stretch, overall AI use jumped ten percent. Inside organizations, the slower pace may simply reflect older policies and procedures, along with training and communication gaps that take time to close.

Surveys repeatedly show that over 60% of workers feel uncertain about how AI will affect their roles, and that uncertainty is likely to persist until organizations define new norms for oversight, delegation, and shared accountability. Clear policies, timely communication, and practical training can help thaw the initial chill. As trust grows, teams become better positioned to use the AI power tools now available to them.

The Society for Human Resource Management, along with MIT Sloan Management Review and Harvard Business Review, has published recent discussions and reader perspectives on these themes.

Rumblings across the internet suggest that AI is responsible for corporate shake-ups. Managers write about demands they now face. Six months ago, their leaders had them find, test, and approve. Today, they are said to be redesigning workflows and reassigning responsibilities. Ethical considerations are ongoing concerns that require defined practices to avoid policy blunders.

Employees at all levels are seeing amped-up expectations for productivity and having their roles redefined in ways that are sometimes energizing, sometimes unsettling.

These rumblings are heard in current and recent editions of the MIT Sloan Management Review, Harvard Business Review, and the Center for Creative Leadership. All three online publications regularly explore human–AI collaboration, workflow redesign, and ethics.