July 2026 put nine figures into construction robotics in two weeks, and neither round was raised on a roadmap. TerraFirma's machines were grading Texas jobsites while the term sheet was being priced. Monumental's robots had laid the walls of more than 100 homes before Khosla Ventures led its Series B. Capital has started scoring construction robotics the way it now scores all of robotics: on delivered work. The catch is that in construction, delivered work is structurally harder to convert into the next contract than anywhere else in hardware. The practice behind this note lives at GTM consulting for deep tech and hardware.
Two rounds, one month, one pattern.
TerraFirma announced $115 million on July 14, with a $100 million Series A led by Kleiner Perkins inside it, for semi-autonomous heavy construction equipment. The company was founded in 2024 by two SpaceX alumni, and its machines are not on a test pad. While the round was being priced they were doing commercial site work in Texas: pad preparation for a Starbucks in North Austin, grading for a sports arena, groundwork for a power substation. The productivity claim, up to 300 percent per operator, is being tested on live jobs where the owner can count the schedule days.
A day later, Monumental closed an oversubscribed $32 million Series B led by Khosla Ventures, with Plural and Hummingbird returning, to take its bricklaying robots from Europe into the US. The fleet numbers over 150 machines and has built walls for more than 100 homes in the Netherlands and the UK, half of that output delivered in the three months before the round. The commercial detail that matters most sits underneath the robots: Monumental does not sell machines. It bids for masonry scopes and gets paid when the walls stand.
Both rounds priced a record, not a promise. That is the same pattern the wider $540 million July window published across robotics, where Walden, BRINC, and TerraFirma all raised after the machines were earning. The cross-sector version of that argument is in the companion note, Robotics GTM: capital now follows deployment. This note is about what makes construction the special case, and it starts with what a jobsite is.
A factory pilot can graduate in place. A jobsite pilot cannot. The site it proved itself on stops existing.
Why the jobsite pilot is the hardest pilot in hardware.
Median pilot-to-production conversion across hardware is 12 percent (IDC 2025), and that number is mostly measured in buildings that do not move: factories, warehouses, labs. A factory pilot that performs can graduate where it stands. Same building, same line, same operations manager, same budget owner. The machine proves itself and then keeps working in the exact place it proved itself.
A jobsite pilot has an end date written into it before the robot arrives. The job finishes. The site demobilizes. The project team that watched the machine work scatters to the next three projects. There is no keep running here. Converting a jobsite pilot means winning a seat on the next project, and the next project is a new sale: new site conditions, new schedule, new trades, often a new owner.
The committee problem compounds it. The modal B2B buying committee already runs 6.3 to 6.8 people (Gartner and 6sense composite), and a construction buy assembles its own version on every project: the owner, the general contractor, the sub whose labor the machine touches, safety, field operations. Then the project closes out and that committee dissolves. Whatever your machine proved, it proved it to a room that no longer exists. The anatomy of the room is in the hardware buying committee.
And the number the pilot produced only counts if it travels. A result stated in robotics units, cycle times, autonomy rates, teleoperation ratios, dies in the estimator's office. The number has to arrive in the trade's units: cost per pad prepared, per course laid, per cubic yard moved, schedule days returned. An estimator can check those against the company's own production rates. That check, not the demo, is the sale.
What the July rounds actually priced.
Read as GTM evidence, the two rounds are two different answers to the same question: how does proof travel when the proving ground disappears?
Monumental's answer is to change what the buyer buys. Its outcome model reads like a pricing choice, but it is a go-to-market architecture. A contractor who buys finished walls instead of robots never convenes a capital committee for a machine purchase at all. The buy arrives in a unit the buyer already purchases every week, a masonry scope at a bid price on a schedule, and the pilot problem quietly disappears: every job is simply a job, and the machine re-proves itself inside a purchase order instead of beside one. The cost of that choice lands on Monumental's own book, where fleet utilization becomes the whole business. Which is exactly why the round was priced on output: walls for more than 100 homes, half of it in the last three months. The record is the collateral.
TerraFirma's answer is to make the record public. Put the machines on live commercial scopes and let named sites accumulate. A Starbucks pad in North Austin is small work for heavy equipment. It is also a reference an owner's rep can drive past. Named sites that a skeptical general contractor can call are how a machine wins its next jobsite from the outside.
Neither company waited for a customer to design the proof for them. Both built the mechanism that makes evidence outlive the project. That is the discipline the rest of the field is missing, and it is buildable at any size.
The record that travels: four moves.
If you run a construction robotics company between $1M and $20M in revenue, the July rounds are the diligence standard your next raise and your next enterprise deal will be held to. The readout is not raise more. It is: make every deployment produce evidence that survives demobilization. Four moves, in order.
Bid the pilot, do not demo it. A paid scope on a live commercial job outranks a dozen demonstration days on a test pad. Real money changes who watches: an unpaid demo draws engineers, a paid scope draws the estimator whose production rates you are beating and the superintendent who owns the schedule. The design rules for a pilot built to convert are in pilot to production.
Name the next jobsite before mobilizing this one. In construction, the conversion decision is which project the machine deploys to next, and a specific person holds it: the GC's operations lead or the owner's program manager, someone with a project pipeline. If that person is not named in the pilot agreement, the pilot dies with the site. A machine that finishes its scope without a named next project has to start the sale again from zero.
Produce one number an estimator can verify, then write it into a file that travels. One auditable economic number per deployment, in the trade's units, checkable against the contractor's own book. Then the reference file: site, scope, number, and the name of the person who will take the call. In a project-based industry the file has to outlive the project team that watched the work, because the next buyer was not on that jobsite. The record has to travel across companies as well as across projects; in construction, the company on the next job is usually not the company on the last one.
Sequence the raise after the record. Both July rounds were priced on delivered work. A construction robotics company that walks into a raise with named reference sites and an estimator-verifiable number is selling the 2026 sequence. A deck of test-pad videos is selling the 2021 sequence to a market that stopped buying it. The method that builds the record on purpose is Proof to Pipeline, and the sourced anchors behind every number on this page are on Hardware GTM Benchmarks 2026.
Two rounds are a month of sentiment, not a controlled study. Construction robotics is having a funded year, and funded years end. Round sizes prove appetite as much as method.
The structural half of this note does not move with the wave. Construction buying has been project-based since long before anyone financed a robot, and it will still be project-based when the wave recedes. A record that travels across projects wins in both markets, funded and dry.
Three questions, answered straight.
Why do construction robotics pilots fail to convert to production?
Three structural reasons stack on top of hardware's 12 percent median pilot-to-production conversion (IDC 2025): the jobsite ends, so there is no in-place graduation path; the buying committee dissolves with the project and re-forms on the next one; and pilot results stated in robotics units do not survive the estimator's office. Conversion in construction means winning the next project, which is a new sale.
What should a construction robotics pilot measure?
One auditable economic number in the trade's units: cost per unit of installed work, schedule days returned, or rework avoided, verifiable against the contractor's own production rates. A pilot that produces a highlight reel instead of an estimator-checkable number has produced marketing, not evidence.
Should a construction robotics company sell machines or outcomes?
July 2026 funded both answers. Monumental sells completed walls, which moves the purchase into a unit contractors already buy and puts fleet economics on the vendor's book. TerraFirma deploys equipment on live commercial scopes and accumulates named reference sites. Both are mechanisms for making proof travel across projects. The wrong answer is a machine purchase that asks a project-based buyer to underwrite a permanent capital decision from a temporary jobsite.
Keep reading.
This note sits inside a published method. Start with the page that ties it together, or go straight to the piece that names your stall point.