Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.
Acquire to Automate: When the Robot Maker Buys the Payroll
Security officers walked off three office towers in downtown San Francisco on September 18, 2026, for one day. Their union, which represents about 14,000 guards in California, wanted a path to $30 an hour. By the union’s account, the contractors across the table, Allied Universal and Securitas among them, had opened with a raise of 25 cents spread over four years. One officer, Keven Adams, thirty-six years on the job, told the San Francisco Standard that guards who asked for more had been warned they could be replaced by AI and advanced camera systems. He was not impressed. “AI doesn’t make decisions. You program it.”
Four weeks later the warning turned up in the financial press as an investment plan. On October 16, Bloomberg and Reuters reported that SoftBank was in early talks to buy Allied Universal, described as the largest security company in the United States, for about $18 billion, or close to $30 billion with its debt. People familiar with the talks said SoftBank would eventually replace some guards with robots on certain tasks, and that humanoids would take longer. Nothing is signed. No price is agreed, SoftBank declined to comment, and as of November 20 no deal had been announced.
The deal may never happen. The idea behind it is already at work in a dozen other industries. Call it acquire to automate: a company that builds or finances the machines stops selling them to employers and buys the employer.
The salesman who skipped the pitch
A robot maker normally has to sell. The prospectus that took Agility Robotics public says its warehouse humanoid rents “at a discount to a customer’s fully burdened labor cost,” which is a polite description of a negotiation. The customer can haggle, wait a year, or buy from someone else. Every robot has to get past a person whose job is to say no.
Masayoshi Son has spent two years assembling the selling side of that table. SoftBank agreed in 2025 to buy ABB’s robotics division for $5.4 billion, a purchase still waiting on approval in China, and it plans a US stock listing for Roze, its robotics company. In an interview on Blackstone’s podcast published on October 13, three days before those reports, Son said, “We have more than 30 robotics companies in our portfolio,” all of them quietly getting ready. People still do a lot of tiring jobs, he added, and who enjoys that?
Then he went looking for the other side of the table. The Financial Times reported in October that Son is seeking up to $100 billion from Gulf investors for a fund that would buy established companies and rebuild them with AI and robotics. One person familiar with the matter, The Paypers reported, suggested Allied could be a target for such a fund. A fund like that has no need to persuade a security company to rent a robot. It owns the security company, and it owns the robots.
What a guard costs
The arithmetic is public. The Bureau of Labor Statistics counts about 1.3 million security guards in the United States, and the median one earned $38,020 a year in May 2025, about $18.28 an hour. Allied says it employs approximately 770,000 people in more than 100 countries. A guard company sells hours. Its invoice to the building is the guard’s wage plus a markup, post by post and shift by shift.
That makes guarding a clean test of the Substitution Threshold, the moment the cheapest reliable provider of a service is no longer human. The line will not be crossed everywhere at once. Patrol loops, camera monitoring and badge checks are the easy posts. Talking down a drunk at a stadium gate is a hard one, and the reporting on SoftBank’s plan says as much.
The Bureau’s own outlook explains why the crossing would be quiet. It projects about 157,200 openings a year for guards and gambling surveillance officers, and says most will come from the need to replace workers who leave. It also expects surveillance technology to “somewhat limit demand for these workers.” An owner who wants fewer guards does not have to fire anyone. It only has to stop refilling the overnight lobby.
Who leaves the room
When a robot is sold, at least three parties get a say in what happens to the wage it saves. Buying the employer removes them one at a time.
The customer. In the rental model the employer is the customer, and the discount is the thing it bargains for. In the acquisition model the employer is the vendor. The office tower still buys a body at the door by the hour. Nothing obliges its supplier to say when the body becomes a machine, or what the machine costs. Vendors already prefer prices pegged to the wage being replaced. Here there is no vendor price at all, only a transfer inside one company.
The competitor. A robot on the open market can be bought by every guard company at once, and competition then pushes the savings toward clients. A robot kept in-house cannot. Amazon ran the experiment from the other direction. In March 2012 it agreed to buy Kiva Systems for about $775 million, when Kiva’s warehouse robots were also working for Staples, Gap and Walgreens. After the purchase Kiva stopped marketing its systems, and most of its sales staff left. Retailers that wanted the same robots had to wait for other companies to build comparable machines. By 2019 Amazon had more than 200,000 robots of its own. An employer that buys the robot maker and a robot maker that buys the employer end up in the same place: one owner holds both the machine and the payroll it replaces.
The worker. A union can bargain with an employer that needs guards. David Huerta, president of the California guards’ union, put that leverage in one sentence that month: “You cannot tell your clients and the public that security officers are indispensable, then come to the bargaining table and treat them like they’re disposable.” The argument works on a company whose business is renting out people. It has less grip on an owner whose reported plan is to need fewer of them, and who can set the pace through attrition instead of layoffs.
What remains is one owner making three decisions that used to be made in three places: how fast the posts are automated, what the client pays afterward, and where the difference goes.
It already has a playbook
Security guards would be the largest case, not the first. Venture capital has been running the same move on office work for two years under the name “AI roll-ups.”
General Catalyst set aside $1.5 billion to incubate AI companies and use them as vehicles to buy established service firms, and their customers, in seven industries with plans for up to twenty. Marc Bhargava, who leads the effort, gave TechCrunch the reason. Services take in $16 trillion a year worldwide against $1 trillion for software, and he reckons 30 to 50 percent of the work can be automated, up to 70 percent in call centers. The firm aims to at least double the profit margin of what it buys. The investor Elad Gil gave the logic in one line: “If you own the asset, you can [transform it] much more rapidly than if you’re just selling software as a vendor.”
The pattern has grown quickly. Thrive Holdings, started by the OpenAI investor Thrive Capital, was reported in July 2026 to be raising about $2 billion from Altimeter, D1 and SoftBank. OpenAI took a stake in it and placed its own engineers inside the acquired firms. One Thrive company, Current, has bought 48 local accounting firms. Dwelly, which buys British rental agencies, says one property manager now oversees more than 300 units, up from about 100. In May 2026 Long Lake, backed by General Catalyst, agreed to take American Express Global Business Travel private for $6.3 billion.
Read in order, the targets are forty-eight accounting firms, a travel company, and now perhaps 770,000 guards. The work has moved from spreadsheets to lobbies, and the buyer from a venture fund to a conglomerate with its own robot makers.
The honest caveats
Robots cannot yet guard a building. SoftBank’s own Pepper, a humanoid greeter, had its production paused years ago, and the reports on Allied put humanoid guards at the far end of the plan.
The roll-ups are unproven too. The margins quoted above are targets set by the investors, and TechCrunch’s headline was that the transformation “may be harder than VCs think.”
Guarding also sells something a machine does not carry. Clients pay for a licensed person who answers for what happens on the post, which is the liability gap in a uniform. And clients are not helpless. Guard contracts get rebid, and a rival would be glad to take a tower from a supplier that swapped its people for machines and kept the price.
Last, Allied carries about $12 billion of debt. In a company that leveraged, the savings from automation have a first claimant, and it is the lender.
What the framework asks
The book’s answer to the Substitution Threshold is not to stop the machine at the lobby door. It is the Foundation, a floor that no payroll decision can take away, so that the person at the door for $18.28 an hour is not ruined when the post goes. Until that floor exists, the useful question about any automation deal is who can still contest it.
Two of the book’s Five Laws apply directly. Law 2 is Truth Must Be Seen. Agility’s rental price and build cost are public only because securities law forced them into a prospectus. A private guard company owned by a conglomerate publishes neither. Clients who buy guarding, public agencies above all, can require it in the contract: which posts are automated, what they cost now, and how the saving is split.
Law 3 is Power Must Decay. A robot that only its owner’s subsidiaries may use is a wage anchor with a lock on it. Merger review exists for this case. A purchase that joins the supplier of the machines to the largest employer of the people they replace is a vertical merger by any definition, and the remedy is old: the machines stay for sale to everyone.
None of this protects a guard by itself. It keeps other people in the room: a client who can see the saving and ask for it, a rival who can buy the same robot, a union that knows the schedule it is bargaining against. Savings reach the public only when somebody is in a position to demand them.
Keven Adams was right about the machine. “AI doesn’t make decisions. You program it.” Under acquire to automate, the company that programs it also picks which posts go first, sets the price of what replaces them, and signs the paychecks of the people still standing there.
Sources
- California security guards strike (San Francisco Standard, September 18, 2026) and Bay Area security workers guarding Big Tech offices strike (KQED, September 18, 2026)
- Security officers in the Bay Area launch 24-hour strike (SEIU-USWW, September 2026)
- SoftBank wants to buy America’s biggest security guard company and use robots (24/7 Wall St., October 17, 2026, summarizing Reuters)
- SoftBank in talks to buy Allied Universal, valued near USD 30 bln (The Paypers, October 19, 2026)
- SoftBank seeks $100 billion from Middle Eastern investors (Tom’s Hardware, October 10, 2026, summarizing the Financial Times)
- SoftBank CEO Masayoshi Son: Superintelligence, AI Investing, and the Future of Robotics (Inside Blackstone podcast, October 13, 2026)
- Security Guards and Gambling Surveillance Officers (Bureau of Labor Statistics, Occupational Outlook Handbook, 2025 data)
- The AI services transformation may be harder than VCs think (TechCrunch, September 28, 2025)
- Thrive Holdings raises $2 billion to buy and rewire services firms with AI (PYMNTS, July 6, 2026) and Buy the firm, then rebuild it with AI (PYMNTS, August 4, 2026)
- Amazon Robotics (Wikipedia, for the Kiva Systems purchase)
Related Articles
- The Substitution Threshold: The line this deal is a bet on, crossed one post at a time.
- Wage-Anchored Pricing: What the robot costs the customer when the vendor still has to sell it.
- Anticipatory Displacement: Why the cuts arrive before the machine does.
- Human Reserved: Whether a post can be fenced off from the threshold by law.
- The Liability Gap: What a licensed person at the door sells that a machine does not.
- Robots Run on Electricity, Not Wages: The running costs an owner inherits along with the robots.
- Collusion Without Agreement: Another way a price stops being contested.
- The Humanoid Robot Revolution: Where the machines stand today.