The AI XR Podcast Sep 18, 2026 56m 38m saved
With Jan Goetgeluk, founder and CEO of Virtuix, a former JP Morgan investment banker who took the omnidirectional treadmill company public on Nasdaq this year
Virtuix does not need scale to work. Jan Goetgeluk put the arithmetic plainly: at 3,000 units a month the company does $100 million of annual revenue, and he expects hundreds of millions from defense alone.
Almost every consumer virtual-reality company built its plan around Meta's headset market growing large enough to leave room for it. Virtuix sells a treadmill that needs a headset, and more than half of its customers had never owned one before they bought it.
"Yeah, Peloton for gamers, absolutely — and we do, it's in a lot of our ads and marketing and messaging, is, play games, stay fit, stay in shape, stay active when playing games, absolutely."
Goetgeluk was an investment banker at JP Morgan before founding Virtuix, started work on the treadmill in 2012, appeared on Shark Tank in 2013 without a deal, and took Mark Cuban's money six months later at a conference. The company now sells to consumers through Meta, to the Army and Marine Corps, to the Navy's morale program, and to Tesla and Figure AI for robot teleoperation.
The full episode is covered here so you can skip it. 56 minutes of audio, 18 minutes of reading.
Here are the 18 takeaways that matter.
Key Takeaways
At 3,000 units a month Virtuix does $100M of annual revenue, and he expects hundreds of millions from defense alone
Orders are up 3x since the Meta launch, in the thousands rather than tens of thousands
More than half of buyers had never owned a VR headset — the company sells the system, not the accessory
The military has had flight simulators since 1929 and never had one for infantry
Virtuix sold its first systems to the Navy's morale program, which runs centers on 68 naval bases
Tesla and Figure AI are repeat customers for robot teleoperation, which one host called the company's biggest market
Omni Arena built over half a million players before the company stopped selling new units
One user lost 40 pounds in four months, and the product qualifies for health-spending accounts
Autism therapy is the next vertical, across 12,000 US therapy centers already reimbursed for VR
The hosts argued the pacing debate is really about product liability ahead of an IPO
Snap's new see-through headset is $2,200 and aimed at enterprises, not consumers
1. Snap's See-Through Bet
The hosts opened on the launch of Snap's glasses, which one of them had spent 30 to 40 minutes with at the company's event. He came away impressed against low expectations, and with a clear view of who the buyer is.
Pre-orders at $2,200 ship within weeks. The device runs AI on board and lets a developer program experiences on the glasses themselves using Snap's own tool, which removes what he called the main obstacle to enterprise deployment: previous hardware needed someone who knew game engines, was hard to upgrade, and was expensive. He said that unlocks enterprise applications the show had only speculated about three or four years ago. His judgment on the market was blunt: the device is for early adopters, developers, lens creators and enterprises rather than consumers.
The competitive position is unusual, because nobody else is shipping see-through optics: "They're the only one on the field since the HoloLens and the Magic Leap are no longer available except to hackers." The talent came partly from Apple and partly from companies Snap acquired, including one bought years earlier by Google.
2. The Teardown Strategy
The second host's argument was about who profits from that position. Samsung, Meta and Apple will have bought the device and taken it apart, and teams inside them will now be forecasting which internal components can be replaced over five years to get smaller and cheaper. He said the same thing happened to Magic Leap, whose first and second headsets he believes are visible in Snap's design.
His model of how the large companies operate is to let someone else prove the category, buy the result, work out the single unsolved problem and ship later. His example was the folding phone: Samsung iterated for years, Apple identified the crease as the problem, spent years on it and then launched. He applied the same reading to AI, with Google watching the labs at the frontier fight it out before releasing something better. His summary of the strategy was deliberately ugly: the large companies eat the bodies of the pioneers and step on them.
3. Pacing and Liability
On the essay arguing that frontier labs should slow their rollouts and accept independent evaluation, and the endorsements that followed it, the hosts went straight past the safety argument to the securities one.
A company heading for a listing goes quiet, the host who has taken a regulated high-technology company public said. Everybody stops talking while the registration document is reviewed, because the lawyers are examining every risk. Against that, the damaging event was not a departing employee saying there was a greater than 10% chance the company kills everyone — it was a safety and alignment colleague publicly agreeing with the figure, cheerfully. That cannot be withdrawn, and the underwriters and funds now have to price the litigation exposure, because every attorney general and law firm will look for anyone harmed by the product, now or later.
His comparison was to medical technology, where companies go out of their way to show the probability of harm is tiny, because they do not want to hurt anyone, do not want to be sued, and are regulated. Which produced the question the segment kept returning to: "you don't need the government to regulate you to self-regulate first, you just don't build things that are that dangerous."
4. The Airplane Analogy
The hosts' argument against the claim that an unsafe frontier is unavoidable was drawn from industries that already solved it. No commercial aircraft maker could say it cannot build a plane without one exploding every day; the accident rate is roughly one in a hundred million. No carmaker could tell buyers that one vehicle in ten blows up, because a competitor would advertise reliability instead.
Their conclusion was that a company unable to build something powerful and safe is either choosing not to or does not know how, and that other industries have demonstrated it is possible. On the argument that a rival state will build it anyway, the host said that powerful and unsafe is the worse outcome for the country doing it, because it hurts more than it helps in reaching any equilibrium with a foreign adversary.
5. Local AI vs API Bills
The cost argument the hosts made is for running models on your own hardware. A high-end Nvidia consumer card is expensive, they said, but it keeps everything inside the organization's own network, removes the security question, and runs open-weight models that cost nothing to download. Against what companies are now spending on interface charges for leading models, one host expects this to get steadily more attractive, and said it is the reason he thinks Nvidia bought Hugging Face.
6. China's Practical Bet
The hosts contrasted the West Coast pursuit of a general intelligence with what they read about elsewhere, which is companies building narrow, practical systems. One said the all-in bet on a general system now looks like the wrong one, because so many practical and useful bets are in play without the same hazards. On China specifically, one host said the reporting he reads is that the goal is not a general intelligence at all, and that the goal is considered a bad one.
Their example was video generation. A Chinese company focused only on video has produced a model the hosts rate an order of magnitude better than anything else, at a moment when video models had been leapfrogging one another every few months. It is expensive and it produces edited video with sound, which changes the economics of advertising: "you make for $50 what used to cost $5 million." One host said the model wants context and expects prompts written like a novelist rather than a screenplay, filling in direction and performance itself, and that this has changed how people inside generative AI think about making content.
7. Electrodes in the Brain
The last news item was a brain-computer interface from a company one host had not previously heard of, described as an implant using 400 electrodes to capture neural signals. A woman who has been unable to speak for years can now speak through it, and it reads only her intention to speak rather than her thoughts generally. He rated it an order of magnitude beyond what he had seen from earlier implant demonstrations.
The other host split the field in two. Restoring function to people with paralysis, Alzheimer's or degenerative disease is medicine, and he named several companies working on it. Implanting electrodes in healthy people to add capability is a different proposition, and he said he is not in favor. His question was what a person looks like in twenty years with a very large electrode array and effectively merged intelligence, and whether that produces a separation event between enhanced and unenhanced humans.
8. Why a VR Treadmill
Goetgeluk started on the problem long before the current hardware existed.
The gap he set out to fill
I started looking at this problem way back when, in 2012, before VR was cool again — I wanted to build an omnidirectional treadmill that was compact and not too expensive, and would work well, and nothing existed.
Jan Goetgeluk
His first prototype video predated any consumer headset shipping. The company began with a Kickstarter for consumers, moved to out-of-home entertainment and arcades, built a larger multiplayer system for venues, and has now come back to a home product.
The mechanism avoids moving parts entirely: a concave, low-friction dish, over-shoes, and a harness.
The footprint
You're held in place — you're not walking off this platform, it's about, call it, five feet in diameter
Jan Goetgeluk
Six feet of clearance is enough, and inside it a user walks without boundaries in a virtual environment.
9. Orders Up 3x With Meta
The home product now ships in a version built with Meta for its headset.
What that did to demand
our orders are up 3x since we launched a product with Meta
Jan Goetgeluk
Asked for the order of magnitude, he did not inflate it.
The absolute number
Yeah, it's not tens of thousands, thousands.
Jan Goetgeluk
Alongside consumers, he named defense training and simulation as a source of real traction, plus enterprise work with Tesla and NASA, and healthcare as an emerging application.
10. The Arcade Era Ended
The host's recollection was that Virtuix sold several hundred of its multiplayer arcade systems to family entertainment centers, a category he said includes five to ten thousand venues in the United States combining arcade games and mini golf.
The player base it built
we built up a player base of over half a million players with our Omni Arena systems nationwide
Jan Goetgeluk
Customers included large chains and a long tail of family-owned single sites. The design requirement was different from the arcade market it came from: an operator used to assembling equipment himself will tolerate complexity, but a chain venue needs something a 16-year-old attendant can run at high throughput. Esports competitions and leaderboards were added to drive repeat play.
How well that worked
some of those teams have played over 300 times
Jan Goetgeluk
Teams came back weekly for the contests. The business has since stopped.
What changed in the category
Ever since COVID, labor got so expensive, and it really started to go back to unstaffed, low-tech.
Jan Goetgeluk
Virtuix no longer builds new arena systems. It services the installed base, earns recurring revenue from game play and maintenance contracts, and helps arrange secondary sales. The market was close to saturated and the appetite for large staffed attractions had gone.
Geographically the product stayed domestic, with one system in Las Vegas. The venues, he said, are mostly outside city centers, because the square footage is too expensive in them.
11. Peloton for Gamers
One host asked whether the addressable market resembles scuba diving or skydiving — people who accept real cost and inconvenience for an experience. Goetgeluk said his consumer buyer is a gamer, and that the second reason they buy is fitness, because the product has them running.
The number he uses
We had one user lose 40 pounds in four months using our product.
Jan Goetgeluk
Which puts it in a different budget
you can even buy it with HSA, FSA funds
Jan Goetgeluk
Asked directly whether the positioning is an alternative to a connected exercise bike, he agreed without hesitation.
The company's own framing
Yeah, Peloton for gamers, absolutely
Jan Goetgeluk
That framing explains a fact that separates Virtuix from most companies in the category.
Where the customers come from
more than half of our customers, at some point, did not yet own a VR headset when they bought the Omni
Jan Goetgeluk
They are not buying a headset and then accessories, he said. They want the complete system because it moves them and is good for their health. The host's reading was that Virtuix is building a market with room for Meta rather than waiting for Meta to build one with room for Virtuix.
12. Life as a Public CEO
Asked whether he is glad he listed, Goetgeluk said yes, and was candid about the cost.
His background made the decision easier
before starting Virtuix I was an investment banker at JP Morgan, so I like the capital markets, I'm in my element
Jan Goetgeluk
The company went public at a high private-market valuation, which the public market then recalibrated painfully, and lockups expiring put the price in front of him daily while, on his account, the business was booming.
His summary of the job
being an entrepreneur, you really fluctuate between extreme lows and pain, and medium highs, kind of fine times
Jan Goetgeluk
Asked later whether being public constrains him, he said the opposite.
What the listing bought him
we have more options to raise capital, we have options to do M&A, to do acquisitions, since we have a public currency
Jan Goetgeluk
He feels more watched rather than more restricted, and noted that everything he says is now public, including on a podcast. The host suggested a shopping trip through a virtual-reality sector full of companies dependent on Meta's ecosystem and now looking for another route.
13. Defense Is a Pull
The host framed defense as a pivot other location-based entertainment companies are also making, and noted that the defense industry moves slowly despite virtual reality being an obvious fit for infantry training, which is expensive to run outdoors. Goetgeluk corrected the framing.
Not a pivot
the VR hype is maybe a bit dead, but VR is finding some real use cases
Jan Goetgeluk
The attention from that market has always been there, he said, and the movement component is what the military did not previously have. Then the arithmetic that makes the business work without consumer scale.
The unit maths
If we sell 3,000 a month, we're doing $100 million in annual revenues
Jan Goetgeluk
Virtuix does not need to sell a hundred million units, he said, because it is not Meta.
The defense ceiling as he sees it
we can do hundreds of millions in revenues in just defense space alone
Jan Goetgeluk
Two structural features of defense procurement work in his favor.
Why a civilian business helps
the defense business really likes it when you have a civilian use case
Jan Goetgeluk
The department calls it dual use. Without a commercial side, revenue waits and suppliers fail.
Their name for that failure
they call it a valley of death, they just unfortunately don't make it across
Jan Goetgeluk
The host's reading was that the government does not want to underwrite a supplier entirely, and prefers a contract that is a fraction of a company funded mostly by commercial revenue. The second feature is a change in purchasing policy.
The shift to off-the-shelf
the military finally decided that, hey, we should no longer reinvent the wheel and spend three years and a billion dollars creating our own wheel, when I can just buy it on Amazon
Jan Goetgeluk
Virtuix's treadmill is commercially available, which is what qualifies it.
14. The Virtual Terrain Walk
The simulations running on the platform in defense are bespoke rather than commercial games, and cover mission planning and rehearsal. The technique Goetgeluk described takes drone or handheld camera footage and reconstructs a location in three dimensions.
What they call the result
we call it the virtual terrain walk
Jan Goetgeluk
Soldiers walk the mission area physically before they are on it, to get a sense of space and distance. He said it is harder to do behind enemy lines.
The other program he named
we just demonstrated a counter-drone, counter-UAS training system to the US Army at Fort Benning
Jan Goetgeluk
The same system is being developed for the Marine Corps: drone swarms arriving over a replica of a base, and drills for defending against them.
Asked whether a harness restricts what a soldier can practice, he said less than the host assumed.
What the platform allows
with Omni One you can crouch, kneel — you can't go prone, that's the only limitation
Jan Goetgeluk
Crouching and seeking cover are critical to the training. And the historical claim he makes for the product is specific.
The gap in military simulation
never before have they had a simulator for ground troops, for infantry
Jan Goetgeluk
The military has had flight simulators since 1929 and simulators for tanks and vehicles, he said, and nothing for walking.
15. The Navy's 68 Bases
The most recent announcement is not a training contract at all. Virtuix sold its first systems to the Navy's morale, welfare and recreation program, which runs recreation centers for troops.
The size of that channel
they have 68 naval bases with a center for morale
Jan Goetgeluk
He called it full circle: selling to the military, for gaming. Beyond the revenue, he said it builds awareness and adoption across the service.
16. Teleoperating Humanoids
Asked about enterprise applications, Goetgeluk gave one he clearly rates.
The customers already buying it
We've sold a few repeat orders to Tesla, and also Figure AI, for robotics
Jan Goetgeluk
With the University of Central Florida the company built a demonstration in which a person on the treadmill wears a headset showing what a humanoid robot sees, and walking moves the robot in real time.
The mechanism
So you're tele-operating the humanoid robot from our Omni platform, with a VR headset
Jan Goetgeluk
He said he does not know how large the revenue opportunity is. One host disagreed at length, and told him this may be the company's largest market of the next decade. His argument: the valuations attached to humanoid robot companies assume the machines will do blue-collar and factory work, and teleoperators will be doing much of it behind the scenes for a long time, the way a remote operator sits behind an autonomous vehicle. That implies tens of thousands, possibly millions, of people needing exactly what Virtuix sells.
His second argument was about reinforcement learning. Training data for humanoids cannot be scraped, so it requires thousands or millions of people putting in very large numbers of hours, and people are already being paid to wear sensors and do household chores so that robots can learn them. He added that he has seen these machines up close and that many are not inherently safe, which may eventually mean legal requirements for human supervision.
He closed by noting that the practice is already standard in adjacent industries, and that behind a good deal of autonomous driving and drone flight there is a person piloting remotely. His advice was that a company founded today purely to teleoperate and train humanoid robots would start at a valuation of several hundred million, and he urged Goetgeluk to look at it before somebody else rebrands the same capability. The valuations attached to the sector, he said, are astronomically large.
17. Autism Therapy Next
Asked where the company will be in a year or two, Goetgeluk named three things: consumer growth on the back of the Meta launch, larger defense contracts and awards plus an acquisition in that sector the company has alluded to publicly, and one emerging vertical.
The market he is watching
it's 12,000 therapy centers in the US that do these kind of therapies
Jan Goetgeluk
Autism therapy for children already uses virtual reality and is already reimbursed in some states, he said, to teach life skills such as crossing a street or speaking to a police officer. Adding movement makes it more engaging and more effective. He was careful to say the company is trying to stay focused on consumer and defense.
18. Cuban and Shark Tank
On Mark Cuban, an investor the show raised, Goetgeluk corrected the premise.
The relationship
He was never on the board, but he invested three times, over various rounds
Jan Goetgeluk
He believes Cuban still holds the shares. The introduction came from television, and it did not work the first time.
How it actually happened
we were on Shark Tank — actually, we didn't do the deal on the show, because we were so early
Jan Goetgeluk
That was 2013, possibly before Meta bought Oculus, and most of the investors on the program were unaware of the renewed interest in virtual reality. Cuban was aware and still thought it early. Six months later the two ran into each other at a conference where both were speaking, Goetgeluk reintroduced himself, and Cuban invested in the seed round then underway.
Bonus Insights
The ticker
The hosts nearly guessed it wrong on air. Virtuix trades on Nasdaq as VTIX.
The Dutch pronunciation is a joke about luck
Goetgeluk told the host his surname means good luck in Dutch, which the host thought fitting for someone who has taken a virtual-reality company public — something he said does not so much happen rarely as not happen.
Where the fish are
On why defense concentrates in one place, Goetgeluk pointed to a single annual training conference in Orlando that he said every defense manufacturer in the training space attends. His phrase for the strategy was that you fish where the fish are.
Robots on Mars
One host asked whether a humanoid robot is now a likelier candidate than a person for a Mars mission. The other agreed, with the caveat that signal delay breaks teleoperation, and the first added that nobody has solved the radiation exposure of the journey for humans.
Goetgeluk's bottom line is that Virtuix does not need consumer scale to be a real business: 3,000 units a month is a $100 million company, defense alone could be worth hundreds of millions, and the movement platform he spent a decade building is now being bought by robot makers who need people to walk for their machines.
Products, Companies & Tools Mentioned
Virtuix (Trades on Nasdaq as VTIX; orders up 3x since the Meta launch, and $100M of annual revenue at 3,000 units a month on his arithmetic)
Meta (Its headset is the platform for the home product, and the launch partner behind the 3x order growth)
Tesla and Figure AI (Repeat buyers of the treadmill for humanoid robot teleoperation and training)
Snap (Its $2,200 see-through glasses shipped to pre-orders; the hosts rate the enterprise case above the consumer one)
Nvidia and Hugging Face (The hosts' case for running open-weight models locally on a high-end consumer card rather than paying interface charges)
Paradromics (Its brain-computer interface uses 400 electrodes and restored speech to a woman who had been unable to speak for years)
Dave & Buster's (A customer for the arena systems, and the reason the product had to be operable by a teenage attendant)
University of Central Florida (Partner on the demonstration in which a person on the treadmill drives a humanoid robot in real time)
JP Morgan (Where Goetgeluk was an investment banker before founding Virtuix, which he says is why he is comfortable as a public-company chief executive)
Shark Tank (Where Virtuix appeared in 2013 without a deal, before Mark Cuban invested six months later)
If this was worth your time, send it to someone who follows the name.
Get the latest market chatter as it happens: