Europe’s fastest-growing software company sells subscriptions and buys inference. Its Series C funds the half of the P&L it does not yet own.
Lovable closed a $400 million Series C on August 12, 2026 at a $13.3 billion valuation, double its December mark, on roughly $500 million of annualized revenue. The revenue line already works. The cost line belongs to Anthropic, OpenAI and Google, and this round buys three things that change it: models Lovable trains itself, a multiyear compute commitment, and a tripling of headcount into machine learning, infrastructure and security. The price shows up in the number Lovable spent all year advertising. Revenue per employee drops from $2.77 million to roughly $1.33 million at the plan.
On August 11, one day before the round went public, Lovable published an engineering post titled “The model picker is a dead end.” It reads like a product philosophy essay. It functions as a cost structure disclosure. Somewhere inside it sits the sentence that explains the $400 million: the unit Lovable optimizes is the finished application, and the company measures each build by what the system tried, how long it took, and what it cost.
What Happened
Menlo Ventures led the Series C, co-led by the Scaleup Europe Fund, an EU vehicle managed by EQT. Balderton Capital and Carmignac joined from Europe, Kaszek Ventures and LTS Growth from Latin America, Tencent and World Innovation Lab from Asia, and Regent from the United States. Accel, Antler, CapitalG, DST Global, Evantic Capital, HubSpot Ventures and Salesforce Ventures all returned. (Regent also owns TechCrunch, which disclosed the conflict in its coverage.)
Lovable hit $500 million in annualized run rate in June and now guides toward about $600 million. That prices the round at 26.6 times current ARR, or roughly 22 times the guide. The company says it hosts 60 million projects drawing 900 million monthly visitors, and that close to two-thirds of Fortune 500 companies have employees using the product, up from half six months ago. Nvidia, Adidas, Zendesk, Klarna and HubSpot appear on the customer list.
Osika framed the raise around product, infrastructure and team. The stated uses: grow headcount 50% to 450 this year, hiring across ML engineering, infrastructure and security; expand offices in London, Boston, San Francisco and New York; and build out automated operations, payments and multi-agent orchestration.
The Backstory
Anton Osika, a physicist who worked at CERN and sold Depict.ai before this, founded Lovable with Fabian Hedin in 2023. The product launched publicly in November 2024. It crossed $100 million in annualized revenue eight months later, then $200 million in November 2025, $300 million in January, $400 million in February and $500 million in June. The valuation ladder ran alongside it: $1.8 billion on a $200 million Series A in July 2025, $6.6 billion on a $330 million Series B in December, $13.3 billion now.
Two numbers from that run matter more than the rest. In February, Lovable reached $400 million ARR with 146 full-time employees, which TechCrunch calculated at $2.77 million of revenue per head. Gartner had predicted a new class of unicorn hitting $2 million per employee by 2030. Lovable beat the 2030 forecast four years early, and the company let that figure carry a lot of narrative weight.
The second number cuts the other way. Barclays analysts found Lovable’s traffic down 40% from its peak as of September 2025, with Vercel’s v0 off 64% and Bolt.new off 27%, and warned investors that growth in this category may rest on month-to-month subscribers who leave as fast as they arrive. Lovable has never published a churn figure. Osika’s answer has been net dollar retention above 100%, a measure that counts the customers who stay and ignores the ones who do not. Lovable also walked back its August 2025 projection of $1 billion ARR within twelve months, declining in March to reaffirm it.
The Plan
Follow the money and it lands on the cost of goods sold.
Lovable is training its own models. The company started with routing, summarizing and commit messages, the high-frequency, low-difficulty jobs that generate the most API calls per unit of value. Those post-trained models now handle what Lovable calls a meaningful share of app-building work in production, and the roadmap points at harder problems and more of each build.
Lovable is building the switch that decides who gets paid. Its control plane assigns different parts of a single build to different models, adapts prompts and tools per vendor, and reroutes to another provider when one runs hot. Osika’s team frames this as quality engineering. It doubles as a procurement engine. Every model competes for traffic on a per-task basis, including Lovable’s own.
Lovable locked in its compute. On June 3, the company signed a multiyear expansion with Google Cloud worth a fivefold increase in its footprint, bundling expanded access to both Gemini and Anthropic’s Claude, a Wiz security integration, and distribution through Google’s Gemini Enterprise Agent Gallery.
The hiring plan finishes the picture. ML engineering, infrastructure and security are cost-control functions, not revenue functions. Osika told Web Summit in November that security engineers were the fastest-growing part of his engineering organization. That was eight months before the money arrived to pay for them.
The Business Model Angle
A dollar of Lovable revenue and a dollar of classic SaaS revenue are different objects. Airtable ran roughly 90% gross margins because a marginal seat cost it nothing. Every Lovable prompt carries a bill. The company learned this the hard way in 2025, when Osika found the platform losing money on its heaviest users and moved to credit-based pricing to cap the exposure.
Which makes the valuation gap worth staring at. Eight days before this round, Bending Spoons agreed to buy Airtable at 2.7 times recurring revenue. Lovable priced at 26.6 times. The company with the worse unit economics fetched ten times more per revenue dollar. That looks backwards until you notice the two buyers were purchasing different things. Bending Spoons paid for cash flow that exists today. Menlo paid for a cost curve that does not exist yet, and the $400 million is the capital that builds it.
Run the arithmetic on the headcount plan and the trade becomes explicit. At 450 people on a $600 million guide, Lovable lands near $1.33 million of revenue per employee, roughly half the February figure and a third below Gartner’s benchmark. Bending Spoons, the most efficient listed software operator in Europe, went the opposite direction over the same period, lifting revenue per employee from $1.12 million in 2023 to $2.57 million in 2025 by cutting. Two European software companies, one week’s news, opposite strategies. Lovable is spending its best metric to buy control of its worst one. That is a defensible trade and a deliberate one, and the efficiency ratios investors use to grade SaaS businesses will register it as a step backwards for at least two years.
There is a stranger loop underneath the Google deal. Google sells Lovable expanded access to Claude, a model built by a company Google invested $10 billion into in April at a $350 billion valuation, with another $30 billion promised if Anthropic hits performance targets. Lovable’s largest variable cost is therefore Google Cloud revenue, Anthropic revenue, and a contributor to the milestones that unlock Google’s own follow-on investment. That is not a supplier relationship. It is a position, and it is exactly the margin migration down the stack that has been reshaping AI economics since the model layer started commoditizing.
One more asymmetry sits in the traffic figure. Lovable charges the person who builds the app. It does not charge the 900 million people who visit one. Across 60 million projects that averages about 15 visits per project per month, and every one of those visits costs Lovable serving capacity while paying it nothing. Credits burn at build time, hosting costs run forever, and the projects that succeed are the ones that cost the most. Shopify solved this by taking a cut of merchant volume. Lovable, so far, sells subscriptions and nothing else.
The Risk
Owning your cost base means fixing it. Training runs and a multiyear cloud commitment convert variable expense into a floor, which is a gift at $1 billion of revenue and a trap at $650 million. Lovable has already declined to reaffirm the billion-dollar target once.
The Cursor precedent should worry Lovable’s board more than the Airtable one. SpaceX bought Anysphere in June for $60 billion, about 15 times its $4 billion ARR, after a source told TechCrunch that the $2 billion Cursor planned to raise would not get the company to breakeven. The largest and fastest-growing business in AI coding could not finance its own path to profit and sold. Lovable is one-eighth its size with the same input costs.
Commoditization cuts both ways here. If frontier inference gets much cheaper over the next 18 months, Lovable will have spent hundreds of millions solving a problem the market handed out for free, while Anthropic and OpenAI keep shipping coding agents on top of the models Lovable rents.
Then there is the cap table. An EU sovereign-backed fund and Tencent now sit alongside each other in a Swedish company whose core supplier chain runs through American clouds and American labs. Nothing about that is a problem today. It becomes one the first time a government decides model access is a strategic asset.
Quick Questions
How much is Lovable actually worth per revenue dollar? $13.3 billion on roughly $500 million of June ARR works out to 26.6 times. On the ~$600 million the company guides toward, closer to 22 times. Replit priced at $9 billion in March on far less revenue; Cursor sold at about 15 times.
Is Lovable profitable? Unknown. The company has never published gross margin, net margin or churn. Reporting has suggested margins worsened after the switch to agentic mode, which consumes more tokens per build.
Why does revenue per employee matter here? It was the proof point for the whole thesis: software built by 146 people generating $400 million. Tripling headcount into infrastructure roles is the company conceding that the thin, model-renting version of itself has a ceiling.
Who is Lovable competing with? Cursor, now inside SpaceX. Replit at $9 billion. Cognition at $26 billion. Base44 inside Wix. Bolt.new, Vercel’s v0, and the coding agents from the two labs that supply Lovable’s models.
What should I watch next? Whether Lovable ever publishes a gross margin, and whether the share of builds handled by its own models keeps climbing. Those two figures decide whether this round worked.
The Business Model Analyst Take
Most of the coverage is reading this as a European champion story, and Brussels putting money in alongside Tencent makes that easy to write. The more useful reading is that Lovable just told the market its business model has a hole in the middle and raised $400 million to fill it.
The hole is real, and the fix is the right one. A company that rents its core input from three suppliers, two of whom sell competing products, does not own a business. It owns a distribution channel with a variable cost problem. Training your own models for the repetitive work, routing the rest to whoever is cheapest at each sub-task, and locking multiyear pricing on compute is how you turn that into something durable. Osika is buying the layer that Google’s own capex disclosures suggest will hold margin once model prices converge.
The steelman for the price is straightforward. If ARR reaches $1 billion in 2027, 26.6 times looks cheap in hindsight, the fixed cost base becomes operating leverage, and the revenue-per-employee dip reads as an investment year. Menlo made this its largest position outside Anthropic, which is a real signal from a firm that can see the diligence.
The bear case rests on one absence. Lovable has published ARR at every milestone and churn at none. In a business where customers buy month to month, burn their credits building, then stop, the retention number is the whole model, and the company has answered every question about it with a metric that excludes the people who left. Watch for a gross margin disclosure. The day Lovable volunteers one is the day this round is working.
