Dave Blundin, founder and general partner of Link Ventures, thinks a new NVIDIA product could make running a data center something like owning a coffee franchise. "If you're entrepreneurial, but you're not super technical, grab one of these and find power for it and go," he said on the Moonshots podcast recorded October 23, 2026. "The margins are crazy high. Cashflow is phenomenal."
The product is DSX POD, which NVIDIA announced on October 21. The panel saw two bets in it. Blundin's is commercial: a small, financeable unit of AI computing that a local businessperson could own. Alexander Wissner-Gross, a computer scientist and founder of Reified, made a broader one about history. In his view, the computer that matters most is now the size of a shipping container.
What NVIDIA announced
Host Peter Diamandis, founder of XPRIZE and Singularity University, called DSX POD "essentially a data center in a box": a pre-built, pre-tested AI data center that ships in containers. You deliver it to a site that already has power and networking, plug it in, and it runs. NVIDIA, he said, claims this cuts setup time from six months to 14 days.
The company's own wording is narrower. Its announcement describes DSX POD as a reference design, a standard blueprint that partner manufacturers build and validate in modular containers, for complete AI infrastructure based on its Vera Rubin chips. The 14 days is "as little as" 14 days, and it counts from arrival at a site already prepared with power, cooling and networking. Cabling, wiring, piping and testing happen in the factory while the site is readied separately. Deliveries through partners including Dell, Foxconn/Ingrasys, Schneider Electric, HPE and Vertiv are due to begin in early 2027.
It is also more than one box. NVIDIA's technical description lays out several coordinated container types with standardized connections, covering computing, storage, networking, power distribution, cooling, controls and the restraints that hold equipment in place during transport. Site teams still have to prepare foundations, upstream electrical supply, outside network connections and the equipment that removes heat, and partners adapt each installation to local building codes.
How the cooling works
Diamandis also explained why NVIDIA says the pods save water. Most data centers, he said, cool their chips the way a body cools itself: by evaporating water, in this case in giant cooling towers that use enormous amounts of it. NVIDIA's design works more like a car radiator. Liquid flows through the chips and picks up heat, outdoor coolers release that heat into the air, and the same liquid circulates again and again with almost no loss. He cited NVIDIA's figures: larger sites could save 54 million liters of water a day, and the pods deliver the same AI output with up to 30% less power.
Blundin waved most of that away. "Nobody uses water cooling towers for data centers anymore," he said. "That's all marketing." In his view the claims about water and local impact were spin, even though NVIDIA presents water savings as a selling point. What he did find impressive was speed. "There's little pods of power all over the planet that are untapped," he said, and moving the computing that answers AI requests, known as inference, quickly to where that power sits is "just brilliant."
He also read the product as strategy. NVIDIA keeps moving "up the stack" from chips to whole turnkey platforms, he said, because so much innovation is coming at the chip level that it would otherwise lack "defensible turf when everything moves to photonics," meaning hardware that uses light rather than electricity. He expects that shift "very, very soon."
The franchise math
Diamandis asked whether someone who could secure electricity might grab some of these pods and build "a cash-flowing engine." Blundin said yes. He compared it to being an early Starbucks franchisee, and to the chance to own a Tesla Cybercab, which he said he had been discussing with his Uber driver. The financing for one unit, he said, would be within reach of anyone who knows their local banker: "sort of like building a small, you know, like a bed and breakfast," or maybe a little bigger.
Later he put numbers on it. The smallest sensible unit, he said, is an NVL72, NVIDIA's rack system with 72 GPUs, at about $5 million. Below 72 GPUs, he said, you can't run frontier models efficiently, and at that scale or above you get about a 2x performance boost. He started to say a container would hold a couple of them, and Wissner-Gross interjected that it could hold about seven. Blundin settled on a hardware range of $5 million to $35 million: "bigger than building a new Airbnb, but smaller than building a hotel." He urged entrepreneurs in construction, or in business generally, to "really think about that." The margins and cash flow are his projections; partner deliveries had not yet begun when the panel spoke.
The conversation turned to the obstacle: permitting would be the tricky part. Blundin said that suits someone who already knows the town and is entrenched with it, then joked, "Just don't tell them."
A carpet factory that took a year
Blundin offered a local example of the speed argument. A friend of his, Rob Fisher, started a data center called Provocative on the Somerville-Cambridge line, near Boston, in a former carpet factory. The building already had a huge power feed and loading docks. Even so, Blundin said, it took probably a year to a year and a half to get running, because no product like this existed. With a pod, he said, Fisher "could literally have parked the shipping container right on the loading dock, run a big cable into it," and been making money in about two weeks.
In that example, power and site access were already in place. By Blundin's account, the delay came from building and connecting the equipment. The discussion then compared the pods to Amazon Web Services, which made cloud computing easy to manage. Standard, modular boxes could become plug-and-play, and the conversation went as far as electrical utilities buying boxes, plugging them in and earning decent margins on their marginal energy.
Computers are getting bigger again
Wissner-Gross said the group was "burying the lead," which to him was the size. DSX POD is a standard intermodal shipping container, so it can be moved. "Almost like an Internet packet," he said, you could want computing on demand "and just ship it."
He placed it in a longer history. For decades the most popular kind of computer kept shrinking: mainframes in the 1960s, minicomputers in the 1970s, microcomputers in the 1980s, personal computers in the 1990s, then smartphones in the late 2000s and 2010s. "And then the trend inverted," he said. Virtually all the transistors made on the newest manufacturing processes, he said, are now going into larger and larger footprints. "It's not like transistors in your pocket anymore." He called it "the dawn of the shipping container computer."
He offered a way to test the idea. Diamandis raised an earlier theme of the show: spreading inference across Teslas and home Powerwall batteries, wherever there is power and bandwidth. Wissner-Gross said Elon Musk would surely try to make money from that. But he called his own account "a falsifiable theory," one that could be proven wrong, and said that if it holds, cars as computers for nearby AI tasks are "not that promising." Things are trending toward the size of shipping containers or larger, he said.
His largest version was a self-contained industrial unit. If a whole data center with its own cooling, and perhaps some of its own energy, fits in a standard container, "you can drop one of these on the moon. You could drop it into a war zone." Some evolution of it, he said, would become the standard unit of "not just compute, but compute plus energy plus a bunch of other utilities," or what he called "the packet of civilization."