When the Workstation Began to Shrink Again

On the sixteenth of October 2026, laptops built on NVIDIA’s RTX Spark platform started reaching the market, five months after the platform was first unveiled in May. For anyone outside the profession this reads as a dry hardware note, another chip, another press release. For me, as an architect who has carried a mobile workstation every working day for years, it landed differently, and not because of the specifications, though the specifications are serious, a Blackwell generation graphics processor paired with a Grace processor, up to a hundred and twenty eight gigabytes of unified memory, something on the order of a petaflop of local artificial intelligence capability, inside a machine far thinner than the heavy slab I am used to hauling. I did not first think about the core count. I thought about my back, and my charger, and the fan that spins up as though the thing intends to take off, and the battery that begins the day with me and then, a few hours in, decides it has had enough of architecture. And then I thought about something larger than any of that. The workstation has begun to shrink again. And that single fact is a whole history of my profession compressed into one product launch.

I do not want to announce the death of the workstation, because every time a new chip appears someone declares the heavy office machine obsolete and they are always wrong, at least on the timeline they promise. Real projects need verified benchmarks, certified drivers, thermal stability, the blessing of the software vendors, and long life under genuine load, and a desktop tower does not vanish because a slide deck says the future has arrived. But the direction of travel is unmistakable, and it is older than this chip, older than this company, older than the computer itself. So let me trace it properly, from the drawing table to whatever it is I will be holding ten years from now, because the story of what the architect carries is really the story of how power keeps migrating into smaller and smaller boxes while the work itself keeps getting heavier.

Before the Computer, the Hardware Was the Table

An entire generation of architects never once asked which graphics card you used. The question was which ruler, which pen, which drawing table. The real hardware of the profession was a piece of furniture, a large inclined surface, a T square, a set of triangles, a Rotring pen, rolls of tracing paper, and drawers full of instruments. The architect did not carry the office. The architect went to the office, because the tools of the trade lived there and could not be moved. The place was the machine. This is easy to forget now, but it set a pattern that computing would inherit without anyone noticing, the pattern of the architect being bound to a location by the sheer physical weight of the means of production.

Then the computer arrived, and it did not free us from the place. It simply swapped one heavy object for another. We gave up the enormous drawing table and received, in its stead, an enormous machine. Early computer aided design lived on large and costly systems, and only through the nineteen eighties and nineties did practices migrate from mainframes and specialized terminals down to desktop and personal machines, a shift that reorganized the office itself. The decisive popular moment came in 1982 with AutoCAD, which helped push the profession off the board and onto machines that more and more people could actually own. The craft changed, and we paid a new kind of price for the change. We no longer needed the big drawing table, but we now needed a machine capable of running whatever we drew, and from that moment the architect was chained to real hardware in a way the pen and the triangle had never demanded.

Then the Architect Became a Prisoner of the Workstation

As we moved from lines to three dimensional modeling, then to building information modeling, then to photorealistic rendering, the computational appetite of architecture grew without limit. Each new program, the modeler, the visualizer, the simulator, delivered the same quiet verdict in a slightly different accent, that the machine you bought two years ago is no longer enough. That was the new digital tax, levied every couple of years, and it had a cruel symmetry to it. The lighter architecture became as a practice, freed from paper and physical models, the heavier it became computationally, until the weightlessness of the drawing was paid for in the mass of the box required to hold it.

At the start of the new millennium a category appeared that we fell in love with immediately, the mobile workstation, a professional machine you could, in principle, carry. Around 2001 the first serious portable workstations arrived carrying professional graphics aimed at exactly our software. The workstation had become portable, in theory. I say in theory because my spine knows the truth. Anyone who has actually used these machines for years knows that the word mobile was often wildly optimistic. The machine is heavy, the charger is heavy, the battery collapses under load, the heat is considerable, and the moment you open a building model, then an image editor, then a heavy scene, then a render, you understand that portability is not the same thing as freedom. You were no longer tied to the desk. You were tied instead to the nearest power outlet, and you dragged that tether behind you through every airport and every site visit and every client meeting for twenty years.

Why This New Chip Is Genuinely Different

I do not want to turn this into an advertisement, so let me be precise about what actually matters, because it is not raw speed. The significant thing about this generation is the fusion of graphics, artificial intelligence, and a large pool of shared unified memory inside a portable body. For most of our computing lives we have thought about two separate memories, the system memory and the graphics memory, and the particular architectural tragedy begins when you have a model too large to fit while the graphics card announces, flatly, that it is done. Unified memory of real size changes part of that equation. It is not magic and it does not automatically retire the professional desktops, but it lets a smaller machine handle workloads that only a few years ago demanded far bulkier hardware. That is the thing I care about, not a higher benchmark but a shifting ratio, the ratio between size and capability, which is the one number that has governed my profession’s relationship to its tools from the drawing table onward.

And this fusion carries a second consequence that has nothing to do with rendering at all, which is local artificial intelligence. Until now most of us have reached for these tools through the cloud, sending our work out and waiting for it to return. When the machine on the desk can run large models itself, the whole arrangement changes, because the client’s files can stay local, the building models and the contracts and the site photographs and the sensitive data need never leave the office, and an assistant can work on the device itself. For an architectural practice this is not a luxury feature. It can become a question of privacy, control, and information security, and it may turn out that the real power of these machines for us is not that they render faster but that they fold the intelligence into the workstation itself, where our data can stay under our own roof.

This Is the Entire History of the Computer

Step back from architecture for a moment and look at the shape of computing as a whole. A room. Then a cabinet. Then a workstation. Then a desktop. Then a laptop. Then a tablet. Then a phone. At every stage, the power that once required an entire dedicated space contracts until it slips inside something smaller, and the thing we thought of as the irreducible minimum becomes the oversized relic of the next decade. This is why I refuse to read RTX Spark as an endpoint. I read it as a station on a line, exactly as the mobile workstation was a station, impressive in its moment and quaint within fifteen years. The history of the computer is a single long act of miniaturization, and there is no reason to believe architecture’s corner of it has reached the bottom.

There is a beautiful irony specific to our profession hidden in this. The person writing an email could work from a feather light laptop many years ago. So could the journalist, the writer, and to a large degree the accountant. The architect, every time he tried to escape the desk, was dragged back by the graphics card asking where he thought he was going with a twenty gigabyte model and a render that wanted all the memory in the world. So even in the age of the cloud, the architect kept carrying a heavy piece of the office with him. If this new class of silicon genuinely lightens that burden, the value to us is not a higher score in a benchmark. The value is freedom of movement, the thing every other knowledge worker received long before we did.

But the Next Stage Will Not Be a Better Laptop

Here I move from history into guessing, and I will say plainly that what follows is not news from any company but my own forecast. I believe we are moving toward a stage in which the device itself matters less than the interface to the work. Today we work on a screen. Tomorrow we may work inside the space. The early spatial computing headsets have already given us a clear prototype of this direction, with demonstrations of large engineering models and three dimensional data streamed into a headset and used for review, collaboration, and handling structured assemblies inside a volume of space rather than on a flat rectangle. Today that hardware is heavy and relatively costly. Tomorrow, if computing keeps its promises, it will be lighter, smaller, and cheaper, like everything else that ever happened in this field, and the honest research on these headsets already shows the real trade, that deep immersion genuinely improves solitary focused work while measurably weakening the sense of being present with others, which tells us exactly what problem the next decade of architectural research on these tools has to solve.

Consider the strangeness of what we currently do. The building is three dimensional. The site is three dimensional. The cities we plan are three dimensional. And we compress all of it onto a flat pane in front of us and push a mouse around until we understand a place. If the glasses ever become truly light and reach professional reliability, the building model could stand in front of me at real scale. I could walk around it, cut it, enlarge a detail, set it onto the actual site, compare the design against the reality in front of my eyes. At that point the device stops being a tool for displaying architecture and becomes part of how architecture is perceived, which is a deeper change than any increase in rendering speed, because it alters not the output but the act of seeing itself.

Then the Phone, and the Device With No Name

I also suspect the phone has not said its last word in our profession. Today we use it for photographs, messages, reviewing documents, occasionally opening a model viewer, scanning, augmented reality overlays, documentation. We do not yet treat it as a full design station. Three things hold it back, the size of the screen, the method of input, and the raw power some workloads demand. The first is already changing with folding devices. The second can change through voice, artificial intelligence, and the glasses. The third changes with faster local silicon and with the cloud. If all three resolve, we may not in the future need a dedicated object called an architectural workstation at all. We may simply carry a personal computing device that opens into a screen when we need one, connects to glasses when we need them, and draws on cloud or local intelligence according to the task. The construction site, of all places, is where this would matter most, the architect standing in the half built space with the full model available in the hand and in the air around him rather than left behind on a tower in the office.

And beyond that lies the genuinely uncertain territory, the quantum machine, where I want to be careful rather than fashionable. Quantum computing is not a faster laptop and not a stronger graphics card, and I do not expect any architect to be running a quantum edition of a modeling program any time soon. If it matures, its value to architecture will almost certainly lie not in drawing but in the hard problems of optimization and complex simulation, structural optimization, material discovery, energy systems, urban networks, impossibly tangled scheduling, design problems with too many variables for ordinary machines to explore. The architect will not draw faster on a quantum computer. But he may be able to test a vastly larger space of possible solutions, and that is a different and more profound kind of power than speed, because it changes not how quickly we execute a decision but how many futures we can examine before we commit to one.

The Architect Did Not Shrink, the World Around Him Did

There is something quietly moving in all of this, and it is the counterweight to every anxiety about the machines. The tasks themselves did not get smaller. The opposite is true. Buildings grew more complex, the models larger, the data denser, the renders finer, the simulations deeper, and artificial intelligence added a whole new layer on top of everything. And yet the object that carries all of this keeps shrinking. That is the paradox I keep returning to, and I think it is the real subject hiding underneath a chip announcement. The size of the problem keeps growing, and the size of the machine keeps falling, and the gap between those two curves is precisely the measure of how far the tools have come. Each shrinking of the device is not a loss of seriousness but a victory over it, the same enormous labor folded into an ever smaller space.

I will not throw away my mobile workstation tomorrow because a company released a new platform, and I would not advise any practice to retire its desktops because a press release claimed the future has landed. The profession has learned, the hard way, that the heavy reliable machine outlives the exciting announcement, and that certification and stability and endurance under load are not glamorous but they are what actually deliver a project. The glossy launches and the industry news will keep proclaiming the end of the workstation, and the design competitions will keep rewarding the rendered image without ever asking what hauled it into existence, and the serious sustainable practice of the craft, the sustainability of our own working lives and our own spines and our own freedom, will go on being decided by the quieter question of how much of the office we are still forced to carry. But the history says one thing without ambiguity. Every single time we were certain that computing power needed a box of a given size, another generation arrived and put it in a smaller one.

In 2001, carrying a workstation with you was an event worth a press release. In 2026, a company speaks of a petaflop and a hundred and twenty eight gigabytes of unified memory in a relatively thin laptop. Ten years from now I expect to look at these machines the way I now look at the workstation I used to carry, and ask myself how on earth I walked around with that thing. Perhaps the next station is a pair of glasses. Perhaps it is a folding phone. Perhaps it is a device we do not yet have a name for. But I am certain of one thing. The architect will never stop demanding more computational power, he will always ask for more, and the only real surprise is that the machine granting him that power will need less and less space on his desk, until one day the most powerful studio he has ever owned fits, impossibly, in his pocket, while the buildings he draws with it grow taller and stranger and more demanding than anything the drawing table ever held.

✦ ArchUp Editorial Insight

The essay’s governing observation is a ratio rather than a specification — that the single number which has governed the profession’s relationship to its tools from the drawing table onward is the ratio between a machine’s size and its capability — and that framing quietly dissolves the hardware-launch excitement into a structural account of dependency. The piece is honest about the pattern it names: the profession was never freed by the computer but merely handed a different heavy object, and each successive software layer delivered the same verdict in a new accent, that the machine bought two years ago is no longer enough, a biennial digital tax levied not by the work’s necessity but by the vendors whose release cycles define obsolescence. This is the same platform-dependency mechanism this archive traced in The Dematerialization of the Studio and Architectural Design for Twenty Dollars, where the conversion of tool ownership into permanent operational dependency placed the practitioner’s working life on terms set elsewhere — and the essay’s most analytically useful move is to notice the one place the migration points back toward the practitioner’s control: local artificial intelligence folded into the device, where the client’s files, contracts, and sensitive data need never leave the office, reframing the chip not as a rendering upgrade but as a question of privacy and information security, a rare instance of a technological shift returning custody rather than extracting it. Where the essay is weakest is precisely where this archive’s discipline would press hardest: it treats the shrinking device as an unambiguous victory over the growing problem, but the forecast toward spatial headsets quietly surfaces the cost its own cited research names — that deep immersion improves solitary focused work while measurably weakening the sense of being present with others — which is the same formation-and-collaboration tension this archive examined in Between Two Films, where the tool that enhances individual output can hollow the shared apprenticeship through which judgment is built; the machine may keep shrinking and the models keep growing, but the gap between those curves measures only computational progress, not whether the profession has decided who bears the cost of the dependency, the isolation, and the obsolescence that each smaller box quietly carries onto the architect’s desk.

Ibrahim Fawakherji — ArchUp

Sources

1. NVIDIA Newsroom, RTX Spark platform announcement and availability, unified memory and local AI capability figures, 2026.
2. Kim KS, Kwon J, et al. The future of two dimensional semiconductors beyond Moore’s law. Nature Nanotechnology, 2024.
3. Fuketa H, Uchiyama K. Edge Artificial Intelligence Chips for the Cyberphysical Systems Era. Computer, 2021.
4. Bittencourt LF, Rodrigues Filho R, et al. The computing continuum: past, present, and future. Computer Science Review, 2025.
5. Satyanarayanan M. Pervasive computing: vision and challenges. IEEE Personal Communications, 2001.
6. Barton JJ, Zhai S, Cousins SB. Mobile Phones Will Become The Primary Personal Computing Devices. IEEE WMCSA, 2006.
7. Wienrich C, Obremski D. Immersion, Attention, and Collaboration in Spatial Computing: A Study on Work Performance with Apple Vision Pro. IEEE Transactions on Visualization and Computer Graphics, 2025.
8. Bao DW, Burry J, Xie YM. Transforming architecture: the role of interdisciplinary collaboration in design and fabrication. Architectural Intelligence, 2024.
9. Autodesk, a brief history of AutoCAD and the shift from hand drawing to computer aided drafting.
10. Dell / The Register, Dell Precision M40 and the 2001 entry into the mobile workstation category.
11. University of Kentucky, CAD and the computerization of architectural labor.
12. Apple Newsroom, spatial computing for business and handling of complex assemblies in visionOS.