[{"data":1,"prerenderedAt":160},["ShallowReactive",2],{"announcements":3,"\u002Fblog\u002Fanthropic-valuation-small-company-ai-spend":4,"_nav_\u002Fblog\u002Fanthropic-valuation-small-company-ai-spend":150},[],{"id":5,"title":6,"authors":7,"badge":10,"body":12,"date":137,"description":138,"extension":139,"image":140,"meta":142,"navigation":143,"path":144,"seo":145,"stem":146,"tags":147,"updated":148,"__hash__":149},"posts\u002Fblog\u002Fanthropic-valuation-small-company-ai-spend.md","Nine Months of AI at Eebz: Can a Small Company's AI Spend Justify Anthropic's $1.3 Trillion Valuation?",[8],{"name":9},"Peter Laughton",{"label":11},"Implementing AI",{"type":13,"value":14,"toc":129},"minimark",[15,19,22,25,28,33,36,39,42,45,49,52,60,63,66,70,73,76,79,82,85,89,92,95,98,101,104,107],[16,17,18],"p",{},"Anthropic is about to list, and the number is extraordinary. $1.3 trillion, on a revenue run-rate of around $65 billion. That is about twenty times revenue. To grow into a valuation like that the way a mature mega-cap eventually must, the revenue has to multiply - not double, multiply. Call it a ten-to-twentyfold increase from here.",[16,20,21],{},"Some of that will come from companies that do not use Claude yet signing up. But a great deal of it has to come from customers who already pay, paying far more. Which gives me a test I can actually run, because I am one of those customers. Would a company like Eebz agree to pay Anthropic ten to twenty times what we pay today?",[16,23,24],{},"So this is not really a piece about how we use AI. It is a piece about the valuation, and our own spend is just the instrument I am using to interrogate it. The question everyone else asks about a number this size - is this AGI, is it about to take us over - is the boring one. From the operator's seat the answer has always been no: Claude rewards skill and punishes laziness, and it is not running the company while I sleep.",[16,26,27],{},"We spend with Anthropic in three places: development, the back office, and the product. Only one of them can carry a number like $1.3 trillion. Here is the walk.",[29,30,32],"h2",{"id":31},"development-a-real-force-multiplier-a-low-ceiling","Development: a real force multiplier, a low ceiling",[16,34,35],{},"Our development team of ten runs, on a good month, like a team of sixty, and Claude Code is the reason. I will use \"force multiplier\" about it without wincing. It is not a magic \"build me an app\" button - point it carelessly and it will carefully build the wrong thing, fast - but managed like a very capable junior, it is the most leverage we have ever had.",[16,37,38],{},"So could Anthropic charge us a lot more for it? A little more, yes. Ten to twenty times more, no - and the reasons are worth spelling out, because they are the reasons the valuation cannot lean here.",[16,40,41],{},"The first is that the tool is decided by its ecosystem as much as by the model. Whether we reach for Claude or for GitHub Copilot to review a pull request turns more on how cleanly Copilot sits inside GitHub than on which model is a shade cleverer this month. The lock-in that matters in development is the surrounding tooling, and Anthropic does not own ours.",[16,43,44],{},"The second is arithmetic. If our Claude Code spend tripled, the cost of moving parts of our stack to an open-weight model like DeepSeek would suddenly look small, and we would pay the engineering cost to switch. Push it to five times and I do not even need another model: a human developer becomes the cheaper option again. Development spend can rise from here. It cannot rise tenfold. There is a firm ceiling not far above where we sit today.",[29,46,48],{"id":47},"marketing-and-sales-flat-and-quietly-exposed","Marketing and sales: flat, and quietly exposed",[16,50,51],{},"The back office is the clearest win we have had and, for Anthropic, the clearest dead end. AI let us run marketing with far fewer people, and we made those reductions - the hardest thing we have done, and no reflection on good colleagues who are not the subject of this piece. But the saving is taken. There is almost no one left for AI to replace, so there is no second act here and no ramp. By volume, this line is done.",[16,53,54,55,59],{},"Nor is there an upsell waiting. I do not think the newest, most expensive models earn their premium for this kind of work, and this article is my evidence. We drafted it twice from the same brief: once with Opus, and once with Fable, Anthropic's pricier model at about twice the cost per token. The people we showed them to overwhelmingly preferred the Opus version. That may say as much about my ",[56,57,58],"code",{},"claude.md"," as about the models, and Fable may well have stuck closer to the brief. But paying double bought us nothing a reader could see. For most of what a back office does, the models are already good enough. What we are really paying for now is integration.",[16,61,62],{},"It is worse than flat, if you are Anthropic. Our marketing and sales do not run on Claude directly so much as on a stack of tools - MailChimp, HubSpot, our analytics, AdWords, LinkedIn, Lemlist, Apollo - and every one of them is bolting AI onto its own product. They will either absorb the work we currently hand to Claude, or sell us tighter AI integration themselves. Either way it is those vendors, not Anthropic, who capture that margin. Anthropic's direct revenue from our back office is more likely to leak than to grow.",[16,64,65],{},"And we are captive, but not locked in. ChatGPT often gives a slightly better answer than Claude. We still do not use it, because it does not live where we work: copying an answer out of one window and pasting it into another is not worth a marginal gain that Claude reaches anyway with a sharper prompt. Integration is what keeps us on Claude in Chrome and Cowork, and on integration Anthropic is well ahead of OpenAI. The day OpenAI stops chasing the AGI dragon long enough to ship a credible \"Chat for Chrome\" or \"Chat Cowork\", that switch becomes trivial. This is not a position I would price a trillion dollars against.",[29,67,69],{"id":68},"the-product-where-the-ten-to-twenty-actually-lives","The product: where the ten-to-twenty actually lives",[16,71,72],{},"Everything above is the setup. This is the line that can carry the valuation, and it is already our largest: we now spend more on the Anthropic API every month than on Claude Code, marketing and sales combined.",[16,74,75],{},"We are building Claude-based features into the digital shelf analytics product itself. It is early and it is imperfect - today Claude is an interesting generalist, and the responses are often not yet accurate enough to put behind a feature a customer pays for. That gap is precisely the opportunity. The version of Anthropic that earns its valuation is not the one that adds more words with Fable; it is the one that ships specialised tools - something that will reason reliably over a ring-fenced data set, for instance - that a company like ours can build a product on.",[16,77,78],{},"Where I think this ends up: we will use an AI the way we use Postgres. Not a novelty bolted on, but a standing part of the application's stack, as unremarkable and as load-bearing as the database. And that is the whole answer to the valuation question, because if AI is in the floorboards of our product then we can charge our customers more for what it lets the product do - which means we can afford to pay Anthropic more. The spend stops being capped by our headcount and starts scaling with our customers, and with the value we add on top.",[16,80,81],{},"That said, there is a limit here too. If the cost became too significant, we would build our own AI. There is plenty of talk about vibe-coding your own copy of Adobe's software, and it is worth remembering that Anthropic is software too. The prompt practically writes itself: \"Claude, please vibe-code me an LLM that uses our data points to build out a series of layers and vectors.\" It would not be as good as Claude. Given everything above about models being good enough, it would not need to be.",[16,83,84],{},"Within that limit, my honest take is that this works. As a small company, Claude lets us innovate faster than our own customers can; built into the app, it puts features in front of them well ahead of our competitors, and I will back us to keep moving faster than the big incumbents can. We are comfortable spending ten to fifteen per cent of revenue on something like Claude, and we fully expect to treble revenue next year. \"Ten to fifteen per cent of a trebling revenue\", repeated across a few thousand small companies making the same bet, is where the implied growth actually comes from. On that line, and only that line, the number stops looking mad.",[29,86,88],{"id":87},"so-is-it-justified","So, is it justified?",[16,90,91],{},"Not by our development team - there is a hard ceiling at three to five times before we switch models or rehire humans. Not by our back office - flat, leaking to our other vendors, and one good competitor away from an easy exit. It has to come from the product: from thousands of small companies turning Claude into part of their stack and growing into it, us among them.",[16,93,94],{},"That is a narrower and more specific bet than \"AI eats the world\". But it is the bet I would take, because it is the one we are already making with our own money - more on the API each month than on everything else combined, and climbing.",[16,96,97],{},"My real worry runs the other way. OpenAI looks to me like a dead man walking, and if it goes, Anthropic loses its biggest commercial rival. After that, the thing keeping its prices honest would be the open-source models, most of them Chinese, such as DeepSeek. They are our real alternative, in development and beyond: close enough on quality, well behind on integration, and ours to run. As long as they stay within reach, Anthropic cannot push its prices very far. If they ever fell too far behind, nothing else would stop it.",[16,99,100],{},"What I got wrong on the way here is instructive. I assumed the value would be a cost saving in the back office; it turned out to be growth in the product. And I assumed, briefly, that AI would mean fewer people. It means fewer roles in some places and a hunger for far more capable ones in others.",[16,102,103],{},"The quiet competitive point, kept to a paragraph: a serious ten-person company that builds this into the floorboards of its product now moves faster than a much larger one can, and compounds that speed where its customers can see it. The ceiling on a small company just moved a long way up. That, not AGI, is the story - and if enough small companies have noticed it, Anthropic is onto a winner.",[16,105,106],{},"Just not a magic one. Claude's moat looks much more like any other software company's than like a capability nobody can repeat - and unlike other software, Claude is the tool that would help its customers build the replacement. It could yet be the architect of its own downfall. A little less hubris would suit every software company, us included.",[16,108,109],{},[110,111,112,113,118,119,123,124,128],"em",{},"This is part of a series on building an AI-native product company. Earlier articles - ",[114,115,117],"a",{"href":116},"\u002Fblog\u002Fwhen-your-developers-start-asking-for-more","how Claude Code transformed our development productivity",", ",[114,120,122],{"href":121},"\u002Fblog\u002Fmanaging-ai-coding-tools","why managing AI coding tools is just managing developers"," and ",[114,125,127],{"href":126},"\u002Fblog\u002Farguing-with-your-ai","the evening I spent arguing with my AI"," - are also on the blog.",{"title":130,"searchDepth":131,"depth":131,"links":132},"",2,[133,134,135,136],{"id":31,"depth":131,"text":32},{"id":47,"depth":131,"text":48},{"id":68,"depth":131,"text":69},{"id":87,"depth":131,"text":88},"2026-11-17","Anthropic is listing at $1.3 trillion, about twenty times revenue. Would a small company like Eebz pay Claude ten to twenty times more? We ran the test on our own spend.","md",{"src":141},"\u002Fimages\u002Fblog\u002Fanthropic-valuation-hero.webp",{},true,"\u002Fblog\u002Fanthropic-valuation-small-company-ai-spend",{"title":6,"description":138},"blog\u002Fanthropic-valuation-small-company-ai-spend",[11],null,"rSQ0XdpjhxdlzzvdA71cmJiVNFF6qa30ySWy1aRTQog",[151,156],{"title":152,"path":153,"stem":154,"description":155,"children":-1},"Ada's Core Values","\u002Fblog\u002Fada-core-values","blog\u002Fada-core-values","Guiding principles for Eebz's AI-powered sales agent - the values that ensure Ada sells with honesty, fairness and respect.",{"title":157,"path":126,"stem":158,"description":159,"children":-1},"Have You Ever Argued With Your AI?","blog\u002Farguing-with-your-ai","Our outreach runs through Claude, and it's changed how we work. Then one evening it refused to delete four contacts, and the argument that followed lines up almost exactly with Yann LeCun's no-world-model critique of LLMs.",1794934262035]