Steelhead and Xometry alternatives, and the number no shop is ever shown: what the marketplace keeps.

Start with what anybody can repeat in twenty minutes. On August 24, 2026 we asked each company the two questions a shop owner should ask before a demo: what does this cost me, and what does it cost me to leave. We requested each pricing page in a browser, read the rendered document rather than the raw response, and searched the whole page for a dollar figure. Then we went to the one place where a company has to answer, which is a filing, and read the relevant sections of Xometry's annual report and the whole of its most recent quarterly results release.

The asymmetry that came back is the reason this page exists. Xometry is listed on the Nasdaq, so the economics of its marketplace are a matter of public record whether or not its sales team wants to discuss them. Steelhead and Fictiv are private companies, and Protolabs Network publishes nothing about its own economics either. On every channel in the table below except one, the number simply is not available anywhere. On that one it is filed with a regulator every quarter, in a document no shop owner would think to open.

Read the first column on its own and every option looks free or unknowable. Read the third and the shape of the decision changes, because one of these channels has to file its numbers with a regulator and the rest do not. That is not a moral distinction. It is a reason to use the one published number as the reference point for all of them, which is what the rest of this page does.

Steelhead Technologies calls itself a job shop ERP on its own home page, and the product it describes is a system of record for a building: digital workboards on the production floor, quoting and order entry, scheduling and capacity planning, job costing, inventory, quality management, invoicing, a customer portal and accounting integrations to QuickBooks and Sage. Its vertical pages name powder coating, plating and anodizing, liquid painting, thermal spray, galvanizing, chemical processing, heat treat and fabrication, which tells you where the company came from. Metal finishing is a spec-heavy, certification-heavy business, and Steelhead leads with spec tracking and audit readiness because that is the part of the job a finisher gets audited on.

Xometry is not selling any of that. It is a marketplace. It owns the customer relationship, quotes the customer with its own instant quoting engine, and then routes the job to a shop in its supplier network. Its partner materials describe a job board of pre-matched work you can claim or counter-offer, and its own annual report describes the platform as connecting buyers and suppliers of custom manufacturing. A shop does not buy Xometry. A shop is sourced by Xometry.

The two names keep landing in the same conversation for one reason: Xometry now ships shop software of its own. Its annual report describes Workcenter, the partner-facing application, as a cloud-based manufacturing execution system, and lists it beside Thomasnet advertising and financial products as services that, in the company's own words, deepen its relationships with suppliers. That is a candid sentence and it deserves to be read as written. Workcenter is not sold to you as an ERP competitor. It is given to you because a supplier who runs their day inside the marketplace's software is a supplier who stays.

Once you separate the two, the real shortlist is obvious and it is not a shortlist of products. It is a shortlist of demand channels, sitting on top of a separate and much more boring question about whether your shop knows its own cost per job. We work the second question through in the manufacturing automation consulting guide, and the quoting-software half of it on the Paperless Parts alternatives page, which covers the products a shop buys to quote its own work. This page is about the channel.

Xometry's supplier recruitment page states the position plainly: no requests for quote, no fees, no subscriptions. Its frequently asked questions repeat it in answer form, saying that joining the partner network and accessing jobs is completely free and that there are no subscriptions or platform fees. Both of those statements are true. A shop pays Xometry nothing.

Xometry's annual report for 2025 explains the model to a different audience. It defines marketplace gross margin as the economic value driven by the spread between the price to the buyer and the cost to the supplier. It then says where each side of that spread comes from: the price to the buyer is set primarily by artificial intelligence through the instant quoting engine, and the cost to the supplier is set by a matching algorithm that finds the optimal supplier in the network. And it reports the result. Marketplace gross margin was 34.7 percent for the year ended December 31, 2025, against 33.5 percent the year before. In the quarter ended June 30, 2026, on marketplace revenue of 215.4 million dollars, marketplace cost of revenue was 140.6 million and marketplace gross margin was again 34.7 percent.

Be careful with that figure, because overstating it would be as bad as ignoring it. Cost of revenue on a marketplace of this kind is not purely what the shops are paid; it carries logistics, shipping and the company's own quality and applications engineering costs alongside the supplier payment. So 34.7 percent is not a per-job commission and nobody should quote it to a customer as one. What it is, on the company's own definition of the line, is the share of the buyer's dollar that does not leave the platform as a payment to the network. That is a number worth carrying into a meeting, and it is the only one of its kind available anywhere in this category.

The report contains a second sentence that matters more to a shop owner than the first, and it is written entirely without malice because it is written to investors. It says that as the company adds to its supplier base, its pricing becomes more competitive, and therefore more attractive to buyers, leading to higher revenue and improved margins. Read from the shop side, that is the whole thing in one line. Every additional shop that joins the network improves the algorithm's ability to find a lower supplier cost. The network's growth is the mechanism, and the network is deliberately large: the same report counts 4,996 active suppliers for 2025, defined as suppliers that used the platform at least once in the previous twelve months, up 17 percent in a year, against 81,821 active buyers.

None of this makes marketplace work a bad idea. Fill-in work at a known margin, on a machine that would otherwise be idle, is good business, and it has been good business since long before anybody put it on a website. It makes marketplace work a thing you have to be able to price, which is exactly the capability most shops in this size band do not have, and it is why the argument on this page ends up being about your own system rather than about theirs. The general form of that argument, across verticals, is on renting AI against owning it.

The spread is the part of the marketplace decision a shop can at least estimate. The part it usually cannot estimate is what the work is worth over five years, and Xometry's own documentation answers that question more directly than most vendors would.

Its partner network page states that projects fulfilled by the network are assigned unique internal identifiers that are not specific to the customer or to the ultimate project, and that customer data is not shared outside the platform's services. The stated purpose is confidentiality for the buyer, and that is a legitimate purpose. The consequence for the shop is structural. You make the part and you do not learn whose part it was. There is no relationship to develop, no second call to earn, no engineer to build a habit with.

The same page addresses repeats. For tooled processes such as injection molding or die casting, the same partner performs the repeat work, because the tool lives in your building. For lower-volume processes such as computer numerical control machining, the page states plainly that the same supplier is not guaranteed to get the repeat order. That is the difference between a customer and a job, stated by the platform itself, and it is the single most useful sentence on the whole site for anybody weighing the channel.

Protolabs Network describes the same architecture from its side in warmer language: it tells prospective manufacturing partners that becoming a partner is totally free, that they get access to pre-vetted and pre-sold orders, and that the platform handles all customer communication and payment processing. Handling the customer communication is the service and it is also the trade. Fictiv's partner page, read the same day, is a contact form under four benefit headings and carries no figure at all. Three platforms, three tones, one architecture: the platform holds the customer.

What follows from that is not that marketplaces are traps. It is that marketplace revenue and direct revenue are different assets and should never be added into one line on a management report. A shop with sixty percent of its hours on marketplace work has a smaller book of business than its revenue suggests, and an owner planning a sale should know that before a buyer's diligence team explains it to them. The wider version of that argument, for platform owners rather than single shops, is on the manufacturers page.

Three products sit around the Xometry marketplace, all pointed at the shop rather than the buyer, and reading them together is what turns this from a sourcing question into a strategy question.

Workcenter is the partner application: a job board of pre-matched work, filters to match your shop, the ability to accept a job or send a counter-offer, drawings and design-for-manufacturability notes in one view, team permissions, performance tracking and a mobile application. Xometry's frequently asked questions say new jobs are added daily, thousands each week. Its annual report calls the same product a cloud-based manufacturing execution system. Both descriptions are accurate and they are aimed at different readers.

Thomasnet is the industrial directory, and Xometry owns it. The annual report describes it as connecting industrial buyers with over 500,000 listed North American suppliers across over 80,000 industrial categories, and says that in late 2025 it launched a performance-based listings model where advertisers set budgets and pay only for potential buyers who interact with their profiles. Services revenue, most of it Thomasnet advertising, was about 8 percent of total company revenue in 2025. So the company that quotes your customer also sells you the advertising that would let your customer find you directly.

The third is money. The Workcenter page describes the standard partner payout as net 40 and offers FastPay, which converts it to net 3 for what the page calls a nominal fee, with the fee itself not published anywhere on the page. The annual report separately lists financial products, naming a product called Instant Pay, among the services the company sells its suppliers. Early payment is genuinely useful to a shop making payroll, and it is also a second margin on the same job.

Set out plainly, that is a coherent and rather impressive business, and the shop-side reading of it is simple enough to say in one sentence: the channel that sets your price also runs your shop floor software, sells you your leads and finances your receivables. No single one of those is a problem. The concentration is the thing to notice, and the antidote is not indignation. It is knowing your own numbers well enough to price each of them separately, which is the boring capability nobody sells because it is not a product.

Steelhead deserves a fair reading on its own terms, because the interesting thing about it is not that it competes with a marketplace. It does not. The interesting thing is that it is built for a slice of manufacturing that most shop systems treat as an afterthought.

Its own pages lead with the metal finishing processes: powder coating, plating and anodizing, liquid painting, thermal spray, galvanizing, chemical processing and heat treat, alongside general fabrication and manufacturing. Its home page describes a digital production floor where operators see instructions with pictures and videos, and it puts spec driven processing near the top of its own list of differentiators, saying its spec tracking simplifies audits for quality-intensive operations. It states that support is handled by a Michigan-based team and that its platform runs with what it calls turnkey cyber security on AWS Gov Cloud. For a shop that runs specifications and certifications on paper today, that is the right shape of product, and the compliance framing lines up with what we hear from the buyers described on the specialty manufacturing page.

The pricing page is where a buyer should slow down. It is titled around flexible pricing, it lays out a core platform plus a long list of add-ons (spec management, quality management, scheduling, inventory, assemblies, maintenance, an analytics tier it calls Sonar Pro, an automation bundle it calls Power Tools, and accounting integrations), and it ends in a request form. We searched the served page for a dollar sign on August 24, 2026 and found none, in the visible text or anywhere in the page source. There is no seat rate, no plant rate, no deployment figure and no starting-from number. That is the norm in this category rather than an outlier, and we said the same thing about it on the Epicor page and again about seven of the nine vendors checked on the quoting software page. It is worth noting here only because the add-on architecture means the number you are eventually quoted depends on which modules you take, so an early quote and a year-three quote can be very different documents.

Two of Steelhead's marketing claims should be carried into a demo as questions rather than as facts. Its quoting page states that 78 percent of buyers award the job to the first company that responds to a quote request, with no source given for the figure. Its order entry section says shops cut data entry costs by up to 80 percent, and its customer portal page says the portal reduced phone calls for a named customer by 80 percent. Those are vendor claims, they may well be true, and none of them is measured in your shop. The only version of any of them that should influence a purchase is the one you measure yourself in the first ninety days, which is the discipline we set out in the production planning and scheduling guide.

Every one of these is a legitimate answer for some shop. The order below runs from cheapest and fastest to slowest and most specific, and the deciding input is the same throughout: whether you can already tell, per job, what the work cost you.

One. Take marketplace work deliberately, as capped fill-in, and instrument it.

The honest case for the marketplace is idle capacity. A machine that runs a marketplace job at a thin margin beats a machine that does not run, and the platform genuinely removes the sales cost of finding that work. Where it wins: no fee to join, no subscription, a job board of pre-matched work and a payment you do not have to chase. Where it costs you: the price is set by somebody else's algorithm, the customer is not yours, and the repeat is not guaranteed on lower-volume processes. Do it with a written cap on the share of your hours it may consume and a per-job margin floor below which you decline, and revisit both quarterly. Neither the cap nor the floor can be set without job costing, which is why this option quietly depends on the third one.

Two. Buy the demand you can attribute instead of the demand you cannot.

Directory advertising, a real website with a quote form on it, and a named salesperson are the unglamorous alternative, and they have one property the marketplace does not: the customer is yours, and you can measure what an inquiry cost. Xometry's own Thomasnet listings now run on a performance-based model where you set a budget and pay when a buyer interacts with your profile, which makes the cost per inquiry knowable. Where it wins: you own what you build. Where it costs you: it is slower, it needs somebody to answer the inquiry within the day, and it fails completely if your quoting turnaround is measured in days. That last constraint is a quoting problem, not a marketing problem, and it is addressed on the quoting software page.

Three. Put a system of record under the shop before you argue about channels.

This is the option that makes the other four decidable, and it is where a product like Steelhead earns its money. If you cannot say what job 4471 cost you in labor, material, rework and oven time, then you cannot tell whether the marketplace price was good or terrible, and no amount of reading about take rates will help. Where it wins: one system, spec and certification records that survive an audit, and a per-job number you can defend. Where it costs you: it is a real implementation with a real change-management cost on the floor, and the price is not published, so budget for a discovery process. The ERP-side version of the same argument, for larger plants already on a tier-one system, is on the Epicor alternatives page and in the Epicor Kinetic playbook.

Four. Take the free shop software and accept what it is.

Workcenter is capable, it costs nothing, and for a shop whose marketplace work is a meaningful share of the schedule it is a reasonable place to manage that work. Where it wins: zero cost, and it is built around the job board you are already using. Where it costs you: it manages your work with one channel, not your shop. Your direct customers, your own quotes, your certifications and your job costing are not in it, so a shop that runs its day inside it is running its day inside a system owned by the party that sets its prices. Use it as a channel console, not as a system of record, and keep the number that matters somewhere they do not host.

Five. Commission a layer that prices the incoming job against your own history.

The argument for building shows up when the thing that decides yes or no on a job cannot be expressed as a template: three decades of part history, an oven schedule with real constraints, a margin policy that changes by customer and by lead time, a rule about what you will and will not run that your foreman can state but nobody has written down. A commissioned layer reads the inbound job, whether it came from the marketplace or from a customer, matches it against your own prior work, checks real capacity, applies your own floor and hands a person a decision to approve. Where it wins: the logic and the code are yours at handoff, and the measurement runs against your history rather than an industry average. Where it costs you: it is a project, not a subscription. Our commissions run between 45,000 and 180,000 dollars against a fixed scope on a six to seven week build cycle, which means it is the wrong answer for a small shop and we say so below. Why the handoff term matters is set out on off the shelf against a commission.

We commission custom systems for a living, so take the following in that light. If your shop cannot produce a defensible cost per job today, a custom build is the wrong purchase and we would rather say it here than discover it on a call. Our smallest band is 45,000 dollars. A shop that does not know its own numbers will not be able to tell whether the thing we built worked, which means the engagement fails on measurement even if the code is perfect. Buy or fix the system of record first. Spend the difference on the floor.

Two things move that answer, and revenue is not either of them. The first is the share of your hours running on channels you do not control. At ten percent, marketplace work is a scheduling convenience. At half your hours it is your business model, the price of your labor is being set outside your building, and the case for owning the logic that decides which of those jobs to accept becomes strong very quickly. The second is process complexity. A shop that runs one process with two variables can quote it on a spreadsheet forever. A finisher juggling spec callouts, certification requirements, rack configuration and oven capacity cannot, and that gap is where a build stops being an ambition and starts being arithmetic.

There is a third case, and it is the one owners talk themselves out of. Some shops have both an ordinary systems problem and one genuinely proprietary rule. The sequence for them is fixed: implement the system of record, run it for two quarters until the per-job number is trustworthy, and only then decide whether the remaining gap is worth a build. The order of operations is the decision. How to structure that so it ends in something somebody signs is on the manufacturing consulting page, and the readiness question is scored at the maturity index.

The shape is consistent in a shop. A layer that reads an inbound job, whether it arrived as a marketplace offer or as a drawing from a customer, matches it against prior jobs in the system you already run, checks real capacity rather than nominal capacity, prices it against your own floor and hands a senior person something to approve rather than something to assemble. The system of record stays where it is. We read and we suggest; we do not write to the plant floor, and the reasoning behind that boundary is in the production planning guide.

We have not shipped a commission for a specialty manufacturer yet. That is stated on the specialty manufacturing page and it is stated here, because a comparison page that quietly implies a track record it does not have would fail the same test we are applying to everybody else on this page. We have also not commissioned a build on Steelhead or on Xometry, and nothing above is a case study. What we have shipped, and what it actually proves, is listed on the manufacturers page. If you want the wider landscape rather than our view of it, that is the best AI consultants for specialty manufacturers, and the way to vet any of us, including us, is the manufacturer's guide to choosing a consultant. What the work costs by model is on the manufacturing cost page, and how it runs week by week is on the process page.

What is our cost per job, and who computes it. Ask it of your own controller before you ask it of any vendor. If the answer takes more than a day to produce, that is the first project regardless of what anybody is selling. A marketplace price is only good or bad relative to this number, and every argument on this page collapses without it.

What share of our hours runs on a channel we do not control, and what is the floor. Write the percentage down, write the per-job margin floor down, and put both in front of whoever accepts jobs. A shop that has never stated a floor accepts whatever the algorithm offers on a slow week, which is precisely when the algorithm is cheapest.

What leaves with us. Ask the system vendor what job history, part records, spec data and certifications you can extract in structured form without a professional services engagement, and get the answer before the first invoice rather than during a migration. Ask the marketplace what you keep if you leave, which is a shorter conversation, because the answer is your machines and nothing else. The same question, asked of a different vertical's vendors, is on renting AI against owning it, and the funnel-side version for manufacturers with a number attached is the slow quoting calculator.

Related reading.

On the rest of the manufacturing stack: what quoting software actually costs and who publishes a price, what Epicor ships against what it has announced, and the Epicor Kinetic integration playbook. On the plant itself, what actually gets automated in a mid-market plant, production planning and scheduling, and the manufacturing automation consulting guide. On who this work is for, the manufacturers page and the specialty manufacturing page, with the landscape on the best AI consultants for specialty manufacturers and the vetting questions on the manufacturer's guide to choosing a consultant. On the money, what this costs a manufacturer, renting AI against owning it and off the shelf against a commission. Put a number on your own quoting delay at the slow quoting calculator, place your operation at the maturity index, see the scope of the manufacturing benchmark we have designed but not yet fielded, see how an engagement runs on the process page and what it costs on the pricing page. The rest of the series is on the comparisons hub, the other verticals on the industries hub, and if you would rather talk it through than read another shortlist, talk to us.