Building World-Class AI from Armenia: five founders on talent, compute and beating incumbents
Key takeaways from the DigiTec "Building World-Class AI from Armenia" panel featuring SuperAnnotate co-founder Tigran PetrosyanDealroom has a profile for this one. Try Dealroom →, Collectly co-founder Levon BrutyanDealroom has a profile for this one. Try Dealroom →, PicsArt's Artavazd MehrabyanDealroom has a profile for this one. Try Dealroom →, ArloopaDealroom has a profile for this one. Try Dealroom → co-founder Arman AtoyanDealroom has a profile for this one. Try Dealroom →, serial founder Arto MinasyanDealroom has a profile for this one. Try Dealroom →, and investor Big Story VCDealroom has a profile for this one. Try Dealroom →:
SuperAnnotate's data-infrastructure play. Petrosyan said the company — started around 2018 by him and his brother while doing PhDs in Switzerland and Sweden — now serves over 100 enterprises, including more than half of the world's seven largest companies, with data annotation and evaluation tooling. It has raised over $65m, most recently a Series B earlier this year, with investors including NVIDIA Ventures and Databricks Ventures (both also customers).
Collectly's Armenia-built healthcare AI. Brutyan said the company helps US healthcare providers bill patients, runs a roughly 120-person team with offices in San Francisco, LA and Yerevan, and builds its AI locally in Armenia — already in production serving around 500,000 patients daily.
PicsArt as an AI-native creativity platform. Mehrabyan positioned it as a design platform with 100m-plus users that has invested in AI since 2015, shipping on-device offline "magic effects" models as early as 2016 and building near-decade-long research in areas such as salient object detection and inpainting, used both in-product and internally for productivity.
Arloopa's move from AR into generative 3D. Atoyan said the AR/VR company recently launched Genie AI (text-to-3D) and Arloopa Factory (image-to-3D) to attack content creation — the biggest bottleneck in AR/VR. He also recounted a failed year-long AR watch-try-on project abandoned after ByteDance revealed a far more precise version, trained on TikTok's vastly larger dataset.
Building deep-tech from Armenia. Minasyan, who runs four companies including Krisp, described pioneering real-time noise cancellation (2017) and real-time accent conversion — a technology his ~15-person Armenian team built over 3.5 years to become number one in its category, beating Palo Alto competitors with far more funding and Stanford/MIT talent.
Competing globally from a small ecosystem. Panellists framed Armenia's constraints — talent depth, time-zone friction, a sales-averse culture and delayed access to cutting-edge knowledge — against the upside of a hungry, hardworking, tightly networked community, arguing niche focus (rather than head-to-head battles with US or Chinese labs) is the realistic path.
The role of compute and Firebird. On Armenia's $500m data-centre push, founders argued cheap local compute matters less for their own cloud-served workloads than as a magnet to attract top talent back to Armenia and to recirculate export revenue into the economy.
An easier but noisier funding market. Brutyan said raising for AI is easier than ever — a recent term sheet took just two meetings — but investors struggle to tell genuine AI from thin "AI-washed" features; local Armenian funds can write $500k–$1m cheques (worth 3–4x that given lower costs), while tens of millions still require Silicon Valley, London or Berlin.
"What if Google does it?" The panel argued startups don't compete with Google but with a product manager inside Google, and that incumbents' large legacy customer bases are as much a liability as an advantage — citing Adobe shutting down its Adobe Aero AR product and Amazon staff preferring SuperAnnotate's tooling over their own.
