TL;DR

• YC W24 open-source IaaS/PaaS cloud, founded 2023, HQ San Francisco
• Three-founder team from Citus Data / Microsoft Azure / Heroku Postgres
• $16M seed (Mar 2024, YC + 500 Emerging Europe); no subsequent round disclosed
• Software abstraction layer running on leased bare-metal (Hetzner, Leaseweb, OVH, AWS Bare Metal) — not a hardware lessor
• Core services: VMs, block storage, networking, managed Postgres, managed Kubernetes (beta), GitHub Actions runners, AI inference
• Positioned as 3x–10x cheaper than AWS; Postgres claimed 9x price/performance vs RDS/Aurora
• AGPL-3.0 — core code is self-hostable; managed service runs the same code
• ClickHouse partnership (Jan 2026) — ClickHouse's native managed Postgres now runs on Ubicloud (private preview)
• The sovereign/open pitch is stronger than the certification stack that backs it

Company Basics

• Founded: 2023
• YC batch: W24 (primary partner Garry Tan)
• Offices: San Francisco (HQ), Amstelveen NL, Istanbul TR (Levent)
• Team size: ~10 at seed (Mar 2024); YC page later listed 15; current not published
• Legal entities: Ubicloud Inc. (Delaware/US) + Ubicloud B.V. (Netherlands)
• Mission framing: "What Linux is to proprietary operating systems, Ubicloud is to cloud"
• Core theses: radical cost compression, eliminate vendor lock-in, architectural transparency / self-hosting option

Founders at a Glance

Three-founder team's common thread: managed Postgres-as-a-service across Heroku, Citus, Azure, and Crunchy Bridge. Ubicloud is Daniel Farina's 4th managed-cloud control plane.

Umur and Ozgun met at Stanford (with third Citus co-founder Sumedh Pathak, who is not in Ubicloud).

Umur Cubukcu (Co-CEO)

• Education: BS Boğaziçi (Istanbul); MS Management Science & Engineering, Stanford (~2001–2003)
• Prior roles: BCG management consultant → Citus Data co-founder & CEO (YC S11, 2011–Jan 2019) → Microsoft Azure Data, product lead for Azure Database for PostgreSQL / Hyperscale (Citus) (Jan 2019–Oct 2022) → YC Visiting Group Partner W23/S23 (Oct 2022–Oct 2023) → Ubicloud
• Public presence: O'Reilly Strata NY 2018 speaker; Citus blog author; heavily quoted in TechCrunch/SiliconANGLE launch coverage
• Signature framing: "OpenStack takes an army of people; Ubicloud is signup-to-VM in two minutes"
• Profiles: LinkedIn /umurc · X @umurc

Ozgun Erdogan (Co-CEO / CTO)

• Education: BS Galatasaray (Istanbul); MS Computer Science, Stanford
• Prior roles: Amazon distributed systems engineer (Seattle, ~2006–2010; holds patents on distributed cache consistency and load balancing) → Citus Data co-founder & CTO (technical lead on Citus distributed-Postgres planner/executor) → Microsoft Azure engineering lead for Citus/Hyperscale (~4y) → Ubicloud
• Public presence: QCon SF 2017 speaker; PostgreSQL Person of the Week; Heavybit community speaker; General Assembly instructor; Startup Reporter EU interview (2026)
• Signature framing: "The entire stack is open-source, from bare metal to application layers, so businesses can audit our privacy and security claims"
• Profiles: LinkedIn /ozgune

Daniel Farina (Co-founder, Infra)

• Education: not publicly disclosed
• Prior roles: Plumtree Software (early career) → Heroku Postgres core engineer ~2010–2015 (widely credited as primary author of WAL-E, the Postgres continuous-archiving tool) → Citus Cloud control plane ~2016–2019 → Microsoft Azure ~2019–2021 → Crunchy Bridge at Crunchy Data ~2021–2023 → Ubicloud
• Public presence: US Patent 8,484,243 (stream query processing, 2013); active PostgreSQL mailing-list contributor; RubyConf 2024 talk "Build a Cloud in Thirteen Years"
• Signature framing: Ubicloud as the 4th iteration of a 13-year Postgres-as-a-service arc; Ruby chosen for infra orchestration because REPL + mature libraries = productivity advantage for a small team
• Profiles: LinkedIn /danfarina

Funding & Capitalization

• Seed: $16M, closed Jan 2024, announced Mar 5, 2024
• Lead: Y Combinator + 500 Emerging Europe
• Other disclosed: Pioneer Fund, Liquid 2 Ventures, ScaleX Ventures (Turkish), e2vc, Rainfall, Maxitech, angels
• Valuation: not publicly disclosed
• No Series A publicly announced as of Apr 2026
• Capital efficiency thesis: software abstraction layer, not hardware lessor — avoids the multi-billion CapEx of CoreWeave-style plays
• Implied runway: strong for lean SF/NL/TR distributed team with no owned datacenter

Product Portfolio

• Elastic Compute — x86_64 and ARM64 Linux VMs; standard and burstable classes
• Block Storage — non-replicated, AES-XTS encrypted at rest, backed by local NVMe
• Virtual Networking — VPC-style private networks, dual-stack IPv4/IPv6, IPsec-encrypted tunnels, nftables firewalls
• Load Balancer
• Managed PostgreSQL (flagship) — HA across AZs, PITR, read replicas, connection pooling, ParadeDB full-text extension, automated backups
• Managed Kubernetes — public beta, single-node and 3-node HA control plane, UbiCSI driver for local NVMe PVs
• GitHub Actions runners — Standard and Premium tiers; x64 and ARM64; 10x larger cache on Premium
• AI Inference Endpoints — OpenAI-compatible API on vLLM V1 via per-model subdomains {model}.ai.ubicloud.com/v1; open-weight models only; streaming, JSON mode, function calling; 500k tokens/month free
• EuroGPT Enterprise — €19/user/mo, Llama 3.1 405B + Llama Guard 3, all GPU processing in Germany, GDPR-compliant
• IAM with ABAC — attribute-based access control from day one
• Strategic note: deprecated raw GPU VM rentals Dec 31, 2025 — moved up-stack to managed inference

Architecture (the "Clover" stack)

• Control plane: Ruby + Roda (HTTP) + Sequel (ORM) + Rodauth (auth) + PostgreSQL (state); orchestrates hosts over SSH (no heavy agent, net-ssh library)
• Host "cloudification": Prog::Vm::HostNexus workflow installs Rhizome host-agent code, SPDK, nftables, configures hugepages, caches boot images
• Virtualization: Linux KVM + Cloud Hypervisor (Rust-based VMM; lighter and more security-focused than QEMU); QEMU 10.1+ used specifically for Blackwell B200 GPU topology
• Tenant isolation: each Cloud Hypervisor instance in its own Linux namespace, runs unprivileged, seccomp-bpf supported
• Block storage: SPDK user-space stack; bdev_aio → vbdev_crypto (AES-XTS + envelope encryption + auto key rotation) → bdev_ubi (custom COW module for instant VM provisioning from base images)
• Networking: IPsec tunnels, nftables, Linux namespaces, dual-stack IPv4/IPv6
• Opinionated: single stack, deliberately rejects OpenStack's "support everything" complexity

Open Source Footprint

• Repo: github.com/ubicloud/ubicloud
• License: AGPL-3.0 — strong copyleft, prevents hyperscaler repackaging-as-SaaS (the "Elastic/MongoDB problem")
• Primary language: Ruby (~92.5%) — deliberate, inherited from Heroku-era experience
• Stars/Forks: ~12k / ~558
• Dual deployment: managed service at console.ubicloud.com OR self-hosted via docker compose + cloudify-your-own-bare-metal
• Third-party OSS leveraged: Cloud Hypervisor, KVM, SPDK, nftables, strongSwan/IPsec, QEMU, PostgreSQL, vLLM, Tailwind

Pricing & Cost Positioning

Representative prices (Germany region, 2026):

Public IPv4: $3/mo. Egress: free up to ~0.625 TB per 2 vCPUs, then $3/TB (≈30x cheaper than hyperscaler egress). Free tier on inference: 500k tokens/month. Per-token pricing for most chat models (Llama 3.3, Mistral Small 3, DeepSeek V3/R1) is dashboard-only.

Performance Claims (Postgres)

Self-published benchmarks vs AWS (independent third-party verification not found):

• TPC-C (transactional): 1.4x more TPS than Aurora at 5.8x lower cost; 4.6x more TPS than RDS at 2.8x lower cost
• Latency: 1.91x lower than Aurora, 7.65x lower than RDS
• TPC-H (analytical): 2.42x faster than Aurora; 2.96x faster than RDS
• Headline: "9x price/performance" vs RDS/Aurora
• Driver: SPDK + local NVMe + Cloud Hypervisor = less I/O overhead per dollar
• Caveat: all numbers sourced from ubicloud.com — no external benchmark surfaced

Competitive Positioning

Key Customers & Partnerships

• ClickHouse (Jan 22, 2026) — strategic wedge. ClickHouse launched its own native managed Postgres service in private preview, powered entirely by Ubicloud. Coincided with ClickHouse's $400M Series D (Dragoneer-led). ClickHouse engineers now contribute upstream. Shifts Ubicloud toward B2B2B infrastructure play.
• Direct customers with public stories: Felt, Hatchet (formal case studies); Resmo, Windmill, PeerDB (homepage logos)
• AudienceKey — cited by third-party research as achieving 50% DB cost reduction post-migration (not independently verified on Ubicloud's site)
• Claimed scale: ~400 paying customers per a Reddit-sourced figure — unverified
• No public Turkish enterprise, government, or bank customers announced

Office & Data Center Footprint

Offices

Production data center regions

Marketing materials reference future regions (Frankfurt, Oregon, Singapore, São Paulo) and additional bare-metal partners (OVHcloud, Latitude.sh, AWS Bare Metal). No broader MENA or APAC presence. Ubicloud owns no physical hardware.

Recent Developments (2025)

• ARM64 VMs and ARM GitHub Actions runners GA; "100x price/performance" on certain ARM CI workloads
• Premium Runners launched (2x faster builds, 10x larger cache, 100 GB free cache)
• Managed Kubernetes moved to public beta (Germany + Virginia); UbiCSI local-NVMe PV driver in preview
• Postgres dashboard overhaul (June 2025)
• AI Inference Endpoints — OpenAI-compatible API on vLLM with open-weight models, managed multi-GPU
• SOC 2 Type II certified (Feb 2025 changelog)
• Deprecated raw GPU VM runners (effective Dec 31, 2025) — strategic exit from CapEx-heavy GPU race
• B200 HGX GPU launched in Türkiye (Istanbul) Private Location (Oct 2025); 4- and 8-GPU partitions added Nov 2025
• B200 HGX GPU virtualization (Dec 15, 2025) — deep technical post on QEMU 10.1+, VFIO-PCI, NVIDIA Fabric Manager, Shared NVSwitch Multitenancy; HN front page

Recent Developments (2026)

• ClickHouse partnership (Jan 22, 2026) — ClickHouse native Postgres powered by Ubicloud; private preview; engineering cross-contributions; tied to ClickHouse's $400M Series D
• Blog output — LLM coding practices, VLM-based OCR, documentation automation, CPU-performance myths ("Does MHz still matter?"), AI Coding sober review
• EuroGPT Enterprise continuing to scale (launched Nov 2024) — privacy-first ChatGPT Enterprise alternative, €19/user/mo, Llama 3.1 405B hosted in Germany
• No new funding round publicly disclosed — most recent remains the Mar 2024 seed

EU/EMEA Regulatory Posture — the credible parts

• Dual-entity controller structure: Ubicloud B.V. (NL) and Ubicloud Inc. (US) — Schrems-II-aware
• EEA-only storage of Customer Account Data (personal data of customers themselves)
• Transfer basis: Article 45(1) adequacy + Article 46(2)(c) Standard Contractual Clauses
• SOC 2 Type II confirmed (Feb 2025 changelog; dedicated /docs/security/soc2 URL currently 404s)
• Matomo for analytics (not Google Analytics) — GDPR-friendlier choice
• Penetration test referenced, available on request
• Proactive engagement on EU Data Act — Nov 2023 blog post welcoming cloud-switching/portability provisions is their most substantive regulatory communication
• EuroGPT residency guarantee: all GPU processing stays in Germany; no customer data used for training

EU/EMEA Regulatory Posture — the gaps

Silent or not-yet-claimed despite their EU sovereignty pitch:

• No ISO 27001 / 27017 / 27018
• No C5 (German BSI — often required for Bundesverwaltung procurement, conspicuous given the German region)
• No SecNumCloud (France / ANSSI)
• No ENS (Spain)
• No EUCS claim, no Gaia-X participation
• No public DORA posture — notable given ClickHouse partnership targets financial services; DORA in force since Jan 17, 2025
• No public NIS2 posture — Ubicloud's IaaS would normally be in scope
• No public EU AI Act role classification — despite operating EuroGPT and inference APIs
• No published BAA process for HIPAA — ToS prohibits PHI absent separate written agreement
• No public SLA posted

Contract Gotchas (Terms of Service)

• Governing law: California
• Data residency not contractually guaranteed by default — ToS permits Ubicloud to move Services Content between regions at its sole discretion absent a written addendum (EuroGPT is a named exception)
• No SLA in the ToS — no uptime commitment, no service-credit regime
• Backups are the customer's responsibility — "Ubicloud does not promise to retain any preservations or backups"
• Termination at sole discretion, with or without notice; may result in immediate data destruction
• PHI and GDPR Article 9 special-category data prohibited without separate written agreement
• DPA not published — available only on request via [email protected]
• Trust Center URL resolves to an empty SPA shell for anonymous visitors
• Sub-processors (Mar 30, 2026): Hetzner (DE/FI), Latitude.sh (DE), Leaseweb (US) for workloads; Cloudflare, Stripe, GitHub, Slack, Matomo, Hubspot, etc. for account data

Risks & Open Questions

Due-Diligence Verdict Summary

Fit-for-Purpose Matrix

Key Sources

Ubicloud primary

• Homepage
• About
• Pricing
• Introducing Ubicloud (blog)
• Cloud virtualization internals
• B200 GPU virtualization
• ClickHouse Postgres powered by Ubicloud
• EuroGPT Enterprise
• Terms of Service
• Privacy Notice
• Sub-Processors
• GitHub repo

Press & founders

• TechCrunch: $16M seed
• SiliconANGLE: $16M coverage
• BusinessWire: ClickHouse native Postgres
• Webrazzi (Turkish)
• Cubukcu LinkedIn
• Erdogan LinkedIn
• Farina LinkedIn
• Erdogan PostgreSQL Person of the Week
• Farina RubyConf 2024 talk

HN threads: 37154138 (Aug 2023), 39598826 (Mar 2024), 44167607 (2025), 46312792 (B200, Dec 2025).

TL;DR — AI Strategy

• Pivoted from raw GPU rentals to managed inference PaaS — GPU GitHub Actions runners deprecated Dec 31, 2025; GPU VMs repositioned as private/enterprise-only
• Open-weight only — no Claude/GPT/Gemini re-hosting; every model on the platform is open-weight
• Three product surfaces: inference endpoints (dev API), EuroGPT Enterprise (SaaS), private B200 VMs (enterprise/BYOC)
• Production runtime: vLLM V1 with FlashAttention-3, FlashInfer, speculative decoding, prefix caching
• Signature technical work: open-source virtualization of NVIDIA HGX B200 using QEMU 10.1+ + Fabric Manager Shared NVSwitch Multitenancy
• AI footprint: Germany (Falkenstein, Helsinki, EuroGPT processing) + Türkiye Istanbul Private Location for B200

Product Surface

Two API surfaces:

Per-model subdomain pattern — each model gets its own hostname (e.g. llama-3-3-70b-turbo.ai.ubicloud.com/v1). There is no unified inference host.

SDK support: any OpenAI-compatible SDK (Python openai, JS); first-party Ruby SDK + ubi CLI (beta).

Free tier: 500,000 tokens / month.

OpenAI Compatibility

Documented and working against the per-model base URL:

• POST /v1/chat/completions — non-streaming
• POST /v1/chat/completions with stream=True — SSE streaming
• POST /v1/chat/completions with response_format={"type":"json_object"} — JSON mode
• POST /v1/chat/completions with tools=[...], tool_choice="auto" — function/tool calling
• /v1/embeddings — implied by Qwen3-Embedding-8B launch (endpoint path not explicitly documented)

Model Catalog (Confirmed Public)

Open-weight only. No Llama 4 in public materials. Context windows and quantization not published per-model.

Public Pricing

Per-token pricing is dashboard-only for most chat models. Only two models are publicly priced on web:

March 2026 addition: new GET /project/{id}/inference-endpoint API returns full price table programmatically with separate per_million_prompt_tokens and per_million_completion_tokens.

Hardware Stack

Not offered in public materials: H200, L40S, MI300X.

B200 partitioning via Shared NVSwitch Multitenancy

Inside a partition: full NVLink/NVSwitch bandwidth. Across partitions: isolated. Fabric Manager enforces routing.

B200 Virtualization — Signature Tech Work

Ubicloud wrote the "missing manual" on open-source virtualization of NVIDIA HGX B200. Stack:

• QEMU 10.1+ (not Cloud Hypervisor) — B200 needs multi-level PCIe topology that Cloud Hypervisor's flat topology can't produce; 10.1 added BAR-mapping optimizations critical for B200's 256 GB Region 2 BAR per GPU
• VFIO-PCI passthrough — vfio-pci.ids=10de:2901, intel_iommu=on iommu=pt; blacklist nouveau/nvidia/nvidia_drm
• nvidia-open driver on guest (proprietary stack can't drive B200)
• NVIDIA Fabric Manager in FABRIC_MODE=1 (Shared NVSwitch Multitenancy) on host; fmpm CLI for partition management
• Host/guest driver versions must match exactly (e.g., 580.95.05)

vLLM V1 Internals

Production runtime is vLLM V1. Three main components:

• AsyncLLM — async wrapper for tokenization/detokenization; talks to engine via IPC (bypasses Python GIL)
• EngineCore — busy loop: pull from input queue, run scheduler + one forward pass per step
• Scheduler — continuous batching via max_num_batched_tokens; all requests finish prefill before decode

Optimization layer

• FlashAttention-3 for forward passes
• FlashInfer (integrated Feb 2025) as high-performance kernel generator
• PagedAttention-lineage block-based KV cache, dynamically allocated
• Speculative decoding on DeepSeek R1 32B (Mar 2025)
• Prefix caching referenced in Dewey.py deep-research demo

Not covered publicly: multi-worker load balancing, health checks, auto-restart, model hot-swap.

EuroGPT Enterprise

The consumer/SaaS face of Ubicloud AI. Available at eurogpt.ubicloud.com.

• €19 per user per month — framed as 3x cheaper than ChatGPT Enterprise / Copilot
• LLM: Meta Llama 3.1 405B (open weights)
• Moderation: Llama Guard 3 (optional, input + output)
• Embeddings: Mistral E5 7B for RAG with private knowledge base
• Web search: DuckDuckGo (privacy-preserving)
• Data residency: "Data remains in Germany, including all GPU processing"
• Training: "No customer data or metadata used for training purposes"
• Security: encryption in transit + envelope encryption at rest, key rotation, file upload
• SSO: OIDC at platform level (Jul 2025); EuroGPT-specific SSO not explicitly documented

Not disclosed: quantization of the 405B deployment. Not offered: private API for EuroGPT — raw API consumers use Inference Endpoints directly.

Strategic Pivot: GPU Rentals → Inference PaaS

Before (2024)

Offered raw GPU rentals (RTX 4000 Ada / H100) as GitHub Actions runners and GPU VMs.

Inflection (2025)

Recognized the CapEx-heavy raw-GPU race against CoreWeave, Lambda, AWS P5, Azure NDv5 as structurally unviable for a seed-stage company. Moved up-stack to managed inference PaaS + dedicated enterprise GPU (private locations).

After (Dec 31, 2025)

• GPU GitHub Actions runners deprecated
• GPU VMs repositioned as private/enterprise deployments (B200, RTX PRO 6000 on request)
• Open-weight inference endpoints become the primary AI front door
• EuroGPT Enterprise becomes the productized SaaS face

Positioning

Differentiators actually claimable

• End-to-end AGPL-3.0 stack (hypervisor → vLLM → UI)
• Proven B200 virtualization (with public technical writeup)
• Germany-resident EuroGPT turnkey product
• Strong Postgres heritage → good RAG / vector story when paired with managed Postgres

Gaps & What's Missing

• No public SLA, rate limits, latency, or throughput numbers for inference endpoints
• No public per-token pricing for chat/reasoning models (only Qwen2.5-VL and Qwen3-Embedding priced on web) — dashboard-only
• Not offered: batch inference API, fine-tuning / LoRA, image generation, audio (Whisper/TTS), multimodal beyond vision-language input
• No public EU AI Act role classification (provider vs deployer) despite operating EuroGPT and open inference
• No named AI customers in public materials; no case studies beyond Ubicloud's own Dewey.py deep-research demo
• No benchmarks vs CoreWeave / Lambda / AWS P5 on B200 workloads; vs OpenAI / Groq / Together on inference throughput or latency
• Istanbul B200 hosting provider not publicly named — framed as "Private Location" / on-request

Key AI Sources

• AI Inference use case
• Python chat-completion tutorial
• Life of an inference request (vLLM V1)
• Virtualizing NVIDIA HGX B200 GPUs with Open Source
• EuroGPT launch blog (Nov 2024)
• EuroGPT Enterprise use case
• Oct 2025 changelog — B200 launch
• Nov 2025 changelog — 4/8-GPU B200
• Feb 2025 changelog — FlashInfer, free tier, speculative decoding
• Mar 2026 changelog — inference-endpoint API, Qwen3-Embedding-8B