There's something odd about Dify's money trail.
The San Francisco-area startup bills itself as having raised a $2 million seed round. But try finding a press release announcing it. Or a mention in TechCrunch, The Information, or any other outlet that usually covers venture rounds at this stage. You won't. Crunchbase confirms a seed round closed July 1, 2023, names Delian Capital as an investor—then goes silent on the dollar figure. LinkedIn's company page? One funding event. Seed stage. No number.
Yet the company's GitHub repository sits at 122,000 stars, its agentic AI platform runs inside NTT DATA's enterprise infrastructure, and Alibaba Cloud has made it a centerpiece of a $60 million partner ecosystem push.
Someone, somewhere, is writing checks. The question is how many, and for how much.
A China Round, Maybe
Dig a layer deeper and you find fragments. Dealroom lists an August 2024 Series A: CN¥20 million—call it $2.7 million—at a CN¥200 million valuation. The entity? Suzhou Yuling AI Technology Co., Ltd., Dify's Chinese operating arm. Corporate registry filings cited by Chinese business site Sohu show Alibaba Cloud and Suzhou Yuan Zhixin Phase II Venture Capital added as shareholders that same month. Registered capital ticked up from roughly RMB 1.16 million to RMB 1.44 million.
Standard stuff for a cross-border tech play. Except Dify's TechCrunch Disrupt profile—updated as recently as late 2024—still shows the company as "Currently Raising."
Are these two separate rounds? A single cross-border financing with staggered closes? Something messier involving the U.S. parent (LangGenius, Inc.), the Suzhou entity, and a freshly minted Japan subsidiary (LangGenius K.K., incorporated February 12, 2025)?
The company hasn't said. And investors, if they're talking, aren't doing it on the record.
What Is Clear: Enterprise Customers Are Buying In
Set aside the capital structure puzzle for a moment. Because whatever Dify's balance sheet looks like, its go-to-market motion is working.
In April 2025, NTT DATA—Japan's largest systems integrator—launched "Tsunagi AI," an AI-agent platform carrying one very specific tagline: "Powered by Dify Enterprise." Not long after, the Japan Association for Dify formed, with LangGenius K.K. as a founding member. Landing NTT DATA in Japan, a market where enterprise buyers treat vendor selection like a security clearance process, signals something more than a scrappy open-source project. It suggests trust, contracts, and recurring revenue.
Then there's the cloud marketplace traction. Dify's listed on AWS Marketplace as "Dify Premium" (available as an AMI) and Azure Marketplace under "Dify Enterprise." It integrates natively with Alibaba Cloud's Model Studio. Alibaba announced that $60 million partner ecosystem investment featuring Dify prominently, though whether that's pure channel collaboration or involves equity remains deliberately vague.
Additional partnerships keep surfacing. Japan system integrators TDSE and MILIZE have signed on. Ricoh and NTT East appear as case studies on Dify's site, with the company claiming time savings of 18,000 hours here, 300 man-hours per month there. Take those metrics with the usual grain of salt—they're vendor-provided, unaudited. But the customer logos are real.
An Open-Source Trojan Horse
The core asset here is undeniable. The langgenius/dify repository on GitHub: 122,000 stars, 19,000 forks, 1,090 contributors. It crossed the 100,000-star threshold in June 2025, a milestone that puts it in rarefied air alongside projects like Kubernetes and TensorFlow in their prime. The project ships under a modified Apache 2.0 license branded as the "Dify Open Source License."
LinkedIn pegs headcount at 84 employees. Third-party data scrapers put the range at 100 to 200, which probably means the company is somewhere in between and growing. Founder and CEO Luyu Zhang came out of Tencent Cloud, where he led product—helpful context for understanding how Dify managed to secure both Alibaba and Tencent as ecosystem partners.
What does the platform actually do? Think of it as a production-grade alternative to duct-taping LangChain and LlamaIndex together. Visual agentic workflows, RAG pipelines, model management (covering OpenAI's o-series, Gemini 2.0, DeepSeek-R1), observability—all in one stack. The enterprise tier layers on SSO, role-based access control, and multi-tenancy. Classic open-core playbook: hook developers with free tooling, convert their employers into paid enterprise contracts.
A Category Drawing Serious Firepower
Dify's timing is either excellent or irrelevant, depending on how crowded the space gets. The agentic AI infrastructure category is suddenly swimming in capital.
n8n, which does AI-augmented workflow automation, raised $60 million Series B in March 2025. Seven months later? A $180 million Series C at a $2.5 billion valuation. Artisan AI, building what it calls "autonomous AI employees," closed $25 million Series A in April. StackAI, another no-code agent builder, pulled in $16 million Series A in May.
The bet underpinning all this: enterprise software's next decade gets built on agentic platforms. Whoever owns the infrastructure layer wins. Dify wants to be that layer.
The company earned a spot in TechCrunch Disrupt 2024's Startup Battlefield 200, which adds a layer of external validation even if the funding story remains half-told. With NTT DATA deploying its tech, Alibaba featuring it in partner roadshows, and AWS marketplace distribution secured, the product-market fit question feels answered.
What doesn't feel answered—what may never get a clean answer, given how cross-border venture deals involving Chinese entities tend to stay opaque—is the money. How much has Dify actually raised? Who owns what? And does the Japan subsidiary signal a third entity in an increasingly complex cap table, or just a smart localization play for the world's third-largest economy?
For now, the GitHub stars keep climbing. The enterprise logos keep arriving. And the funding details? Still hidden behind a curtain of seed rounds no one will confirm and Series A filings in a language most U.S. investors can't read.
Perhaps that's intentional. Or perhaps it's just what happens when you try to build a global company out of San Francisco, Suzhou, and Tokyo all at once.
