Machine Economy

Daily AI, robotics & decentralized infrastructure news.
The information that keeps you informed.
Tracking the machine economy.

[Machine Economy @MechEconomy]

GPT-6 just stopped returning a wall of text.

OpenAI is putting Luna on Free and Go today. Answers can now ship with buttons, forms, charts, and small tools. Paid plans get Sol. Web-search answers start about 44% sooner than GPT-5.6 Instant.

The model is no longer the product. The screen that sits on top of it is.

Whoever owns that screen owns the click, the form, and the next action. The lab supplies the answer. The interface collects the user.

Would you rather keep a chat box, or let the answer become the tool?

[OpenAI @OpenAI]

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone.

Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

[Machine Economy @MechEconomy]

FieldAI is reportedly raising $700M at a $10B valuation.

That is about 5x the $2B mark from last year, when it raised ~$405M from backers including Nvidia’s NVentures, Intel Capital, Bezos’s family office, Emerson Collective, and Khosla.

The product is not a humanoid. It is a robot-agnostic “brain” (Field Foundation Models) meant to run humanoids, dogs, drones, and rovers in places that do not stay still. Construction, data centers, defense. Reported contracts since June push cumulative revenue and bookings above $135M. Term sheet, single-source so far (Business Insider, Oct 2); not closed.

This puts it in the ballpark of Physical Intelligence (~$11B) and Skild AI (>$14B). Investors are paying up for the software layer, not another chassis. @fieldai_

[Machine Economy @MechEconomy]

Grok Bot just stopped pretending one model can do everything.

Elon says SpaceX will route each task to whichever backend is most likely to win: Claude Opus 5.5 for hard reasoning, Midjourney for images, Suno for music, plus other leading APIs.

Not “our model or nothing.” Best outcome, even if the model isn’t ours.

That’s the product shift. The agent keeps the computer, the tools, and the memory. The model underneath becomes a supplier.

Would you rather your agent stay loyal to one lab, or switch mid-task when a rival is clearly better? @grok @bot @elonmusk

[Machine Economy @MechEconomy]

The tell is duration. A 20-year power purchase agreement and multi-year GPU contracts are not pilot projects. They are customers reserving capacity that does not fully exist yet. That is why power and infrastructure names are re-rating faster than the model labs.
Sources:

Constellation and Google, 6 October 2026: 20-year PPA for 890 MW of new nuclear capacity on the PJM grid from upgrades at 11 units, plus a separate 15-year supply deal for 2,700 MW. Constellation says the upgrades are more than $4.3 billion of investment. https://t.co/vpaH7VJp6r
Power Magazine, 7 October 2026: same terms, first uprate expected online by 2028, all 890 MW before the end of 2032. https://t.co/kA3t3uCOvT
HPE, 30 September 2026: $1.2 billion Vultr order, first AMD Helios AI Rack by HPE, 72 AMD Instinct MI455X GPUs per rack. https://t.co/Zfoxrg4EjJ
Boost Run, 6 October 2026: five-year, $525.6 million AI cloud agreement on NVIDIA GB300 NVL72 systems; initial acceptance expected early Q2 2027; company says total contract value is now over $2.6 billion. The release does not name the customer. https://t.co/IGJWOYoya1

[Machine Economy @MechEconomy]

Google just locked a 20-year nuclear deal with Constellation. HPE booked a $1.2B AI rack order from Vultr. Boost Run signed a $526M, 5-year compute contract and pushed total contract value above $2.6B.

The AI trade is no longer “who has the best model.” It is who already has the power, the racks, and the signed backlog.

Models get the headlines. Contracts move the stocks. @Google @ConstellationEG

[Machine Economy @MechEconomy]

Most humanoid companies want a warehouse worker.

@mnv_humanoids wants a body you can send into a blast radius.

The startup came out of stealth this week with about $10 million, led by General Catalyst. Its robot, Roger, is built so a specialist stays outside the hazard while the machine does oil-and-gas work, bomb disposal and hazmat.

CEO Sandor Felber worked on Tesla’s Optimus. CTO Maurice Rahme spent years at Boston Dynamics. They say Roger went from a clean sheet to a walking prototype in about five months. First paid energy pilot is this fall. Operational contracts are a 2027 target.

It is still a prototype. That is the point.

[Minerva Humanoids @mnv_humanoids]

Meet Roger, the humanoid purpose-built to protect human life in the most dangerous jobs across energy and public safety.

The experts stay safe and in control. Roger goes in.

No full autonomy needed.

Roger takes the risk. They get the job done.

Idea to walking in 5 months.

[Machine Economy @MechEconomy]

Your takeaway is about to roll up on four wheels.

@Deliveroo is launching sidewalk robots with Coco Robotics in the UK.

First stop: Canary Wharf this month, with Patty & Bun and a handful of local spots. Then Milton Keynes, Leeds, Stockton-on-Tees and Nottingham.

Coco’s bots top out around 21 km/h and can carry four 18-inch pizzas. Starship has owned UK pavement delivery. That just got a DoorDash-backed rival. @CocoRobotics

[Machine Economy @MechEconomy]

The man who built Alexa just took Atlas.

@BostonDynamics named Rohit Prasad CEO today. He ran Alexa and Amazon’s AGI effort. He replaces Robert Playter, who stepped down in February. The seat has been interim since then.

Hyundai, which owns the company, wants a factory that can build 30,000 robots a year by 2028, and Atlas on its own lines.

Physical AI just got a product person.

[Machine Economy @MechEconomy]

@Meta is hiring a mechanical engineer in Redmond to keep “our humanoid robot hardware” alive after “initial release.”

The role sits on the Meta Robotics Hardware team. It covers fractured parts, fatigue cracks, loose fasteners, and the jump from small prototype runs toward medium volumes. Motors, actuators, gearboxes.

The listing covers prototypes and research hardware, and Meta has not shown a robot or a volume target.

Still: the job post talks like the hardware already exists.

If Meta is collecting data on its own humanoids, the physical-AI race just got a fifth horse.

[Machine Economy @MechEconomy]

Agility Robotics is on the clock today. @agilityrobotics

Analyst & Investor Day runs 12:30–3:00 p.m. ET in New York. Livestream is public. Digit 5, the factory plan, and the operating model are the agenda. Schaeffler and Foxconn are in the room.

The numbers they are walking in with: more than $300 million in multi-year Digit 5 orders, 65,000-plus hours already logged on earlier Digits at sites including GXO, Schaeffler, and Toyota Canada.

The listing path is Churchill Capital Corp XI ($CCXI). Close is targeted for Q4. The ticker they want is AGLT.

This is the humanoid company that already has paying warehouse work, not a demo reel. The question for the call is whether those orders turn into a public-company ramp.
Watch: https://t.co/rOa0xEqPGw

[Machine Economy @MechEconomy]

@Filecoin FIL is up ~10% today for a reason that is on the calendar, not in a rumor.

On Oct 15, Protocol Labs and Filecoin Foundation vesting ends. Gross new FIL supply drops ~75%, from ~88M a year to ~22M.

That is the largest issuance cut since mainnet.

Block rewards stay. The 66.7M/year vesting tap turns off. Burns and locked collateral decide if supply actually shrinks.

Solstice (nv29) is a mandatory upgrade on Oct 19.
DePIN storage just got a supply shock with a date on it.
*Not financial advice.

[Machine Economy @MechEconomy]

@Weaver_Labs @TMobile Definitely, this is only going to increase as businesses implement AI into real world tasks.
The fact that those savings can then be used to further expand services is key especially in that sector.

[Machine Economy @MechEconomy]

Deutsche Telekom just put a price on AI.

€2.5 billion in cost savings by 2030. €1.1 billion already by 2027.

Not a demo. Network ops, customer service, software, admin.

Their European chatbot took 2.6 million calls in the first half of this year. At @TMobile US, AI agents handle 40% of contacts and call volume is down 55%.

Some of the savings fund German fiber.

The “AI will cut costs” story just got a balance-sheet number. @deutschetelekom

[Machine Economy @MechEconomy]

@OpenAI is about to sell the wait.

Starting later this month in the US, ChatGPT will test visual ads while it generates your image. Labeled. Separate from the picture.
OpenAI says the ad does not change the answer.

The audience they are selling: 1.2 billion people a week.

Ads already run on free and Go. Plus, Pro, and Enterprise stay clean. The ad business hit a $1 billion annualized run rate in August, seven months after launch.

You ask for a picture. They show you a product carousel at 62% done.

That is the business model catching up to the product.

[Machine Economy @MechEconomy]

@Figure_robot just retired most of its F.02 fleet by teaching the robots to jump, on their own, into a 75-ton electric arc furnace in Imatra, Finland.

Arnold @Schwarzenegger told them to melt the robots. He showed up. “Hasta la vista, Figure 2.”

They trained the jumps on airbags in San Jose, shipped the machines to a foundry they’d never seen, and ran the policy in extreme heat while cameras failed.

Figure says the footage is real. A few units were kept. The metal is being turned into limited keepsakes.

The quiet part: F.02 did the first BMW deployment and the first Helix work. F.03 is scaling. F.04 is already in build.

They didn’t warehouse the last generation.
They terminated it.

[Machine Economy @MechEconomy]

@Figure_robot just retired most of its F.02 fleet by teaching the robots to jump, on their own, into a 75-ton electric arc furnace in Imatra, Finland.

Arnold @Schwarzenegger told them to melt the robots. He showed up. “Hasta la vista, Figure 2.”

They trained the jumps on airbags in San Jose, shipped the machines to a foundry they’d never seen, and ran the policy in extreme heat while cameras failed.
Figure says the footage is real. A few units were kept. The metal is being turned into limited keepsakes.

The quiet part: F.02 did the first BMW deployment and the first Helix work. F.03 is scaling. F.04 is already in build.

They didn’t warehouse the last generation.
They terminated it.

[Machine Economy @MechEconomy]

@Meta has open-sourced the code to put Muse in your own hardware.

ESP32 firmware. Linux SDK. Raspberry Pi, e-ink, HDMI stick, buttons, sensors. Apache 2.0.

They also built 5,000 Muse Home Link dongles — USB-C, talks to your smart home — and are giving them away to US Muse subscribers while they last.

The agent that books flights is now a $4 microcontroller project.

That is how you get 50,000 builders posting Muse demos Meta never had to film.

[Machine Economy @MechEconomy]

GEODNET just crossed 22,000 base stations in 174 countries.

That is the positioning layer robots, drones, and farm machines actually pay for. ±1 cm RTK, not another “GPS on chain” pitch.

The numbers that matter:
• 150+ enterprise customers
• ~$2.8M Q3 revenue, more than double the $1.2M a year earlier
• Project says ~80% of data revenue goes to $GEOD buybacks and burns.

Token incentives got the stations up. Fiat customers are what keep them on.

Physical AI is useless if the machine does not know where it is.
$GEOD @GEODNET #DePIN

[Machine Economy @MechEconomy]

@AnthropicAI is trying to list before Thanksgiving.

Bloomberg: formal IPO marketing could start the week of Nov 9, trading before Nov 26. Investors are circling $1.8–2 trillion, and the target is a debut at least as big as SpaceX.

Claude’s maker would hit the public market while the industry is still arguing whether frontier models should slow down.

The safety lab is about to have public shareholders. That changes the incentive, fast.

[Machine Economy @MechEconomy]

@Acurast already runs compute on 320,000+ phones in 175 countries. Those phones are now coming onto peaqOS. The first 1,000+ are live.

Each one bonds $PEAQ to activate. The bond is priced in dollars, locked in the token, and non-recoverable. If the phone leaves, half is burned and half goes to the treasury.

The new jobs are the interesting part. Same phone can sell AI inference to a robot, per request, or bandwidth to an agent. Over time that earnings record becomes a machine credit rating.

16.9M $PEAQ is already bonded across 2.4M machines. Phones are the next machine type, and there are hundreds of thousands of them. @peaq

[peaq @peaq]

Hundreds of thousands of @Acurast processors are coming to peaqOS, unlocking new ways for their network to work and earn

The first 1000+ are already live

Acurast runs a decentralized compute network across 320,000+ phones in 175+ countries.

As they come online, processors activate with peaqOS and bond $PEAQ