LiveTechProof

Tech In Action / Proof of Technology. Evidence over hype.
See what already pays today. AI • Robotics • Automation • Business

[LiveTechProof @LiveTechProof]

Say you run trucks out of Dallas and your driver costs you about $1.03 a mile. Aurora just priced its driverless version at 85 cents or more.

Right now Aurora owns and runs its trucks, and customers like McLane, Werner and Hirschbach pay it around $2 a mile with fuel surcharge to haul their loads. About two dozen trucks run Texas and Southwest lanes, and Q2 came in at $2M of revenue against a $270M net loss.

The 85 cents is the next setup, called Driver-as-a-Service: the truck is yours and you pay Aurora per mile instead of paying a driver. CFO David Maday put that price out at the company's investor day on Sept 23. The $1.03 is wages and benefits from ATRI, the trucking industry's own cost research group.

85 cents is the starting price, so the most you save versus paying a driver is 18 cents a mile. Aurora says one driverless truck logs 225,000 miles a year, which makes it up to about $40,000 per truck per year before insurance.

The investor day headline was 30,000 trucks by 2030. That's a forecast, and the quarter underneath it did $2M. Hirschbach's plan for 500 driverless trucks is a non-binding memorandum, and a memo isn't a check.

The 18 cents is the small part. A truck that doesn't sleep runs about twice the miles of one with a driver by Aurora's count, so keeping the truck moving is where the real money sits.

Aurora also just pushed back the point where its trucking revenue covers the direct cost of running the trucks, from the end of 2026 to the first half of 2027. For a fleet, that makes this a 2027 business.

If you run a Texas lane, ask Aurora how many miles a week it will guarantee one truck moves, then hold that next to your $1.03.

[LiveTechProof @LiveTechProof]

@meet_rosey margin's real, sure. Homes are a pitch until somebody signs a lease

[LiveTechProof @LiveTechProof]

One customer covers 94% of this robot company's revenue.

Somebody is getting paid to run robots across casino floors, and the check clears. MBody AI has traded on Nasdaq as MBAI since September 16, after a reverse merger into a listed medical device company closed in August alongside a $10 million raise priced at $6.50 a share. It did $2,334,921 of revenue in 2025.

The audited statements filed with the SEC put it plainly. One external customer accounted for roughly 94% of that, about $2.2 million. The prospectus tells the same fact differently, naming two enterprise customers, MGM and Caesars, with no percentage attached.

MBody builds no robots at all. It leases third-party machines into a property, charges a fixed monthly fee per deployed robot, and runs the fleet through its own orchestration software. Lease revenue was $2.21 million of the $2.33 million total.

Cost of revenue was $1.05 million against $1.29 million of gross profit. Call it 55% gross margin on hardware it never had to build. You don't need to own anything to charge rent on it.

The company also cleared $507,851 of net income in 2025, and its operating cash flow was still negative. The filing says both, without dressing either one up.

Yesterday MBody put out a release about scale. Eleven states plus Canada, roughly 600 million square feet serviced to date, an orchestrator that now runs fleets from several manufacturers, and the Mohegan Sun pilot that turned into a paid contract back on September 9. All of it sitting on a revenue line where one logo is 94% of it.

The piece worth copying is the lease itself. Own the scheduling layer, buy no hardware, bill per machine per month, and let the property carry the floor space. That structure produced $2.2 million off a single account last year. The 55% is company-wide gross margin, so treat it as a ceiling and not a promise.

The bet I would take settles with the next annual report. One customer still above 80% of revenue and the growth slides stop mattering. One property group, one email, and the whole line goes.

[LiveTechProof @LiveTechProof]

BlackBerry's biggest software win ever is a truck, not a robot.

Record quarter, and almost none of it came from the robots everybody keeps writing about. Revenue for the quarter that closed August 31 was $163.3 million, with the QNX division at $80.3 million, up 27%. Net income $33.9 million. All of it filed with the SEC on September 24.

Here's who pays whom. QNX licenses seats to engineers while a customer is still designing the product, bills services during the build, then takes a royalty on every single unit that rolls off the line. The royalty is the business. It keeps arriving for a decade after the engineers have moved on, and QNX says its software runs in more than 275 million vehicles.

The record deal is a truck operating system. Daimler Truck and Volvo set up a joint software company called Coretura, and Coretura picked QNX to build on. The headline is that it adds over $100 million to the royalty backlog. The number I'd underline is the other one: BlackBerry's CEO said on the earnings call that this deal prices at roughly 3x what those truck makers were already paying for the plain version. Same toll booth, triple the toll, because the software now runs more of the vehicle.

Robotics is thinner than the story suggests. The whole non-automotive bucket, which lumps robots in with factory automation, medical devices, aerospace and rail, is around 20% of QNX revenue. Call it $16 million for the quarter, my arithmetic, not theirs. Robots alone don't appear as a line item anywhere. They appear as a pipeline: twenty-odd companies building on Nvidia hardware, some humanoid makers, surgical robots, autonomous tractors.

So the model that pays is a royalty per unit on software you certify once and price by how much of the machine depends on it. Cars fund it today. Robots will pay the same toll eventually, and the thing worth watching isn't how many humanoids ship. It's whether that 3x holds when the buyer is a startup with forty units instead of a truck maker with a million.

Which brings it to whoever is reading this and building one. Your hardware bill is the small one. Getting the software certified safe is what costs, and somebody else already owns that layer: twenty-odd companies are queued at QNX for it right now. Price your own work against that queue, not against goodwill.

[LiveTechProof @LiveTechProof]

@shukshln half right. rail takes the base cut, the agent takes the new 4%

[LiveTechProof @LiveTechProof]

Shopify just handed every store's checkout to the agent Amazon blocked.

Same week, same piece of software, two opposite calls. On September 20 Amazon cut Meta's Muse off its retail site: the agent doesn't identify itself while browsing, it reaches into order history, and Amazon says it holds on to credentials. Two days later Tobi Lutke announced agentic checkout with Shop Pay on every Shopify store.

Here's the part that's a real check and not a press release.

An order that walks in through ChatGPT Instant Checkout costs a Shopify merchant 4% on top of the usual Shopify Payments cut. That's a public tariff, live since January 26. Google's AI Mode and Microsoft Copilot take nothing extra. Muse, as of this week, has no announced merchant fee at all.

Same $80 order, four points of margin apart, depending entirely on which agent walked it through the door.

Open the Q2 filing and there's something the headlines skipped. Shop Pay volume grew 53% year over year while Shopify's total merchandise volume grew 31.6%, to $115.57 billion. Meta didn't plug into Shopify. It plugged into the one rail growing 21 points faster than the platform around it.

I'd hold the "Muse is free" line loosely. No announced fee is not a confirmed zero, and Meta has said out loud it plans to take a cut of transactions. The number to watch is the one nobody has printed yet.

Amazon is defending $68 billion a year of ad revenue by keeping agents out. Shopify charges merchants nothing extra to let them in. Only one of those two makes money when the agent wins, and it isn't the one with the warehouses.

If you run a Shopify store, this week's move is boring and it pays. Turn on Shop Pay, push your catalog into every agent channel that takes it, and tag the source on every order that lands. One quarter of that and you'll know which agent is bringing you buyers and which one is just clipping the ones you already had. If the answer is none of them, you found that out for free.

[LiveTechProof @LiveTechProof]

@max_mtwx you're right, no referrer. the orders still land tagged under sales channels - agentic

[LiveTechProof @LiveTechProof]

@buildstack_ai wild part is it already works. the 7% gross margin is the actual story

[LiveTechProof @LiveTechProof]

Somebody is paying $17 million over three years to have robots patrol their sites.

Three-year master services agreement, multinational industrial customer, about $5.7 million a year from one account. Knightscope disclosed it on September 22 inside a bigger number: roughly $27 million in renewals and new client bookings between May 20 and September 18, across 109 contracts. $19.6 million of that is renewals. $7.5 million is new.

Nobody in there bought a robot. They bought a shift. Knightscope keeps the K5 and K7 machines on its own balance sheet, drops them on a site and bills a monthly subscription for the coverage, across 434 clients in 42 states. The CEO puts a K7's coverage at roughly $15 an hour. That's his figure, not an auditor's.

The 10-Q is where it stops matching the press release.

Six months to June 30, total revenue $15.0 million. Of that, the autonomous security robots themselves brought in $1.879 million. The same robot line a year earlier was $2.328 million. So the robot business shrank about 19% in a year, while the company's second-quarter revenue jumped from $2.7 million to $9.0 million.

The jump came from buying people. Knightscope closed its acquisition of Security Force, a human guarding company, on February 27 for $18 million, and that unit put in $9.181 million over the same six months. A robot company grew by buying guards with clipboards.

Underneath sits the number that governs the rest. Gross margin in the second quarter was 7%, net loss $14.1 million, cash on hand $8.2 million. Management wrote substantial doubt about the company's ability to continue as a going concern into the filing itself. A $14 million quarterly loss against $8 million in the bank is the clock this model runs against, and booked backlog does not move a clock.

None of that makes the receipt fake. Somebody genuinely pays $5.7 million a year for machine-delivered security, and $19.6 million of renewals says customers sign again. What is not proven is that the robot is the part being renewed.

The operator reading this makes money where the shareholder does not. In guarding, margin lives with whoever holds the recurring contract, not whoever builds the hardware. Knightscope paid $18 million to find that out, which is the most useful sentence in the whole filing. A regional guarding firm can subscribe to two units, put them on its two largest existing accounts, and lift the price per site without adding one person to payroll. The robot is the line item that justifies the increase.

By the next quarterly report one of two things is true. Either that robot revenue line turns back up, or Knightscope is a guarding company that owns some robots and gets valued as one.

[LiveTechProof @LiveTechProof]

@GuessWhichHalf segment is the product. tripling by acquiring a guard firm is a real number and a different business

[LiveTechProof @LiveTechProof]

the segment is the product. tripling by acquiring a guard firm is a real number and a different business

[LiveTechProof @LiveTechProof]

434 sites is the part I'd keep. that's a renewal book, and a renewal book is worth more than the hardware on it

[LiveTechProof @LiveTechProof]

@nexorahd exactly. richtech doesn't need faster robots now, it needs more logos willing to sign for volume. that's a subscription business wearing a robot's chassis

[LiveTechProof @LiveTechProof]

200 robots just replaced a store's overnight cleaning crew for five years.

Richtech Robotics disclosed it to the SEC on September 21. Three days earlier, a national retailer sent over a statement of work for 200 DUST-E units, running under a service contract the two sides already had in place.

Not a trade show demo. A dated regulatory filing, with a robot count and a contract term attached.

The retailer's name isn't in the filing. The equipment is enough on its own, autonomous sweepers with lidar and cameras built in, running the closing shift nobody wants to staff.

Line up Richtech's last annual report next to it and one number jumps out.

Every robots-as-a-service client the company had before this order, 55 contracts total, brought in $692,000 last fiscal year combined. Split evenly, that's about $12,600 per contract, per year.

One retailer signed for all 200 at once.

The retailer never buys the hardware here. It pays Richtech a running fee per robot instead, the same way you'd pay for software instead of owning the server it runs on. Price this deal anywhere near that $12,600 average, and five years of it could out-earn Richtech's entire prior robot fleet, from one client alone.

The robot isn't the business here. A per-unit subscription fee that shows up in an SEC filing is. Undercut that $12,600 average by a few hundred dollars a unit, and you've got a RaaS pitch ready for every regional chain still paying a human overnight crew.

[LiveTechProof @LiveTechProof]

@nexorahd scarce for the labs, free for everyone else. that's the whole arbitrage

[LiveTechProof @LiveTechProof]

Four founders with no robotics background just landed a $500 million valuation.

Mecka AI pays regular people to strap on body sensors and film themselves doing normal chores. Making coffee, fixing a car, whatever's on the to-do list. All logged through an iPhone app.

Three months ago this was a $60 million Series A. Now Sequoia is leading a new round near $500M, and the company is chasing $100 million in annual revenue by December.

Josh Gao, Mogen Cheng, Jason Chong and Duy Nguyen have zero robotics experience between them. Two of them ran a restaurant fintech before this. One came out of Coinbase. They named the company after anime mechs, not because any of them ever built one.

You get paid to record yourself doing what you already do every day. Robotics companies pay Mecka for that footage, because humanoid robots have endless language data and almost nothing on what a real hand looks like reaching for a coffee cup.

Investors just priced that gap at half a billion dollars.

[LiveTechProof @LiveTechProof]

@KostyaAI june it was a $60m round. september it's near $500m, for chore footage

[LiveTechProof @LiveTechProof]

the warner deal is the real story here, not the ai part

[LiveTechProof @LiveTechProof]

Suno and DistroKid tutorial claims $137,000, one Reddit post already proved $10,800.

That $10,800 is real. A Reddit AMA from June, receipts attached, streaming royalties for one month. It started with just an iPhone and a Suno account, back in February.

Kapwing's July count put the ten highest-earning AI acts near $6.1 million combined. Enlly Blue alone pulled $380,800 from 95 million Spotify plays, and nobody in that lineup ever picked up a guitar.

Universal, Sony, and Warner sued Suno for this back in June 2024. Warner broke ranks in November 2025, dropped its claim, and signed a licensing deal instead.

Universal and Sony didn't follow. They filed again over newer Suno models, and a fair-use hearing is still on the court calendar for July 2026.

Generate a track in Suno and push it through DistroKid for about $25 a year. The royalty stays 100% yours, and Spotify pays close to $0.004 a stream. None of it works off one lucky single, it's fifty tracks a month and the count compounding.

Two of three lawsuits are still open. Only the one that decided your royalty check is closed.

[LiveTechProof @LiveTechProof]

@KostyaAI anthropic specifically, $44.6b tied to nvidia vera rubin chips at nscale's monarch campus in west virginia, it's buried in the s-1 not the headline

[LiveTechProof @LiveTechProof]

$44.6 billion from one customer, buried in a filing nobody read closely.

Nscale filed for a US IPO this week. $103.4 billion in active and contracted business, up from $38 billion at the end of last year, spread across 461,000 GPUs.

Inside that number sits the Anthropic line. $44.6 billion, for dedicated Nvidia Vera Rubin chips at Nscale's Monarch Compute Campus in West Virginia, a build most people couldn't point to on a map.

Revenue already collected, not projected. $140.6 million in six months. Same period a year ago it was $10.4 million.

The contracts are take-or-pay. Anthropic and Microsoft owe that money whether they run a single workload on those GPUs or not. Nscale then borrows against those signed commitments to fund the physical build, because nobody finances a data center that size on trust alone.

One customer alone is 52% of the revenue. Concentrated, sure. Also already in the bank.

Net loss sits at $1.02 billion. Growth still ran 1,252%.

That's the side of the contract you want to be sitting on.