Back

OVHcloud AI Summit 2026: 55 launches to take AI into production


OVHcloud AI Summit 2026: 55 launches to take AI into production

Why do so many AI projects stay stuck at the prototype stage? It’s rarely for lack of a model. It’s because going into production means juggling several products, several providers, several pricing grids, and data transfers that the compliance team refuses to approve. That’s the roadblock we removed at the OVHcloud AI Summit 2026, on 19 November at the Carrousel du Louvre in Paris, when we announced 55 cloud and AI launches, from GPUs to applications, all from a single European provider.

An overview of the new products

On the cloud foundation side, 38 launches strengthen compute, storage, networking, security, and private cloud. The other 17 cover the entire AI stack, from infrastructure to applications.

But what should you take away from this extensive list, and where does it bring you? Here’s how to understand it:

1. Ready for production

First, read through the list of launches announced at the Summit and count nine “private” in their name: Private Boot, Private NAS-HA, Private Endpoint, Private PKI, VPC, Private DNS, Private Endpoint Kubernetes, Site-to-site VPN, Private VCF 3-AZ. Add Cloud HSM, disk encryption, Confidential Computing, and the 17 new OVHcloud Connect points of presence, and you see a common theme with a large share of the Summit announcements.

This is no accident: it’s the whole point. After all, an AI agent doesn’t live in a demo. It reads your documents, queries your databases, sends emails, and places orders. It lives in your network. And as long as that network leaves a public address on every building block, no one in compliance will sign off on putting it into production. So we spent the year strengthening the network where an agent can work without anything leaving it: a VPC with subnets across three zones, a DNS that resolves your services internally, private endpoints to your buckets and clusters, keys that never leave a dedicated device, and an attestation that proves — with cryptographic evidence — that the server is running exactly as you believe it is.

2. A complete AI stack

On this foundation, our AI services are organised in five layers. Everyone comes in at the level that suits them, from customers who want to manage their own GPUs to those who prefer to delegate everything to us.

Layers 1 and 2: infrastructure and orchestration
 The long-awaited NVIDIA B300 GPUs are here, on bare metal and in Public Cloud. The Rubin NVL8 generation is on its way for the most demanding workloads. To orchestrate your training runs, AI Fabric, built with Blade.sh, turns the GPU into a simple argument in a Python function: no more clusters to size, and billing stops when the container does.

Layer 3: inference
 AI Endpoints is expanding with voice from Gladia, in more than 100 languages, and with Shadow’s models, all on a single invoice. AI Routers will pick the right model for each request based on your rules. A simple question goes to a lightweight model, reasoning to a more powerful one, and code to a coding model, with automatic fallback if a model fails.

Layer 4: the AI PaaS
 AI Agents handles everything around the model: persistent memory, MCP connectors to your tools, channels, scheduled tasks, delegation between agents, and human approval for sensitive actions. One API call, and the agent is deployed. AI Search feeds it: you bring your documents, and the service chunks them, indexes them, and returns the right passages to your applications via API or to your agents via MCP.

Layer 5: applications and SaaS
 Our customers design and deploy their AI applications on this AI-ready cloud. We use it ourselves to build the services they expect. Website Builder creates a complete website from a few sentences and updates it on simple instruction. Receptionist answers your phone line day and night, and creates appointments and tasks directly in your tools. Finally, OVHai Workspace builds on all the layers below (see further down).

3. Olala, our first in-house model, in five open variants

Olala is OVHcloud’s first family of LLMs, developed by OVHai LLM, its lab dedicated to agentic AI. Its first model, Olala 1.0 mini 7A1B, is built for efficiency: it has 7 billion parameters, but only 1 billion are active for each request. As a result, it runs on a GPU with as little as 8 GB of memory, and even on a standard CPU. It will be available as open source on Hugging Face, under the Apache 2.0 licence, with its AI Act compliance documentation.

Olala 1.0 mini 7A1B was trained on three European public supercomputers (Jupiter, MeluXina, and Jean Zay), and the family is already aiming for more than 100 billion parameters with version 2.0.

Olala is an open-weight, open-source model that you can use in several ways. Follow the guides:

“Know-how in model training has become widely accessible, and the cost of training has dropped twentyfold in three years,” explains Octave Klaba, CEO of OVHcloud.

At Qwant, Olala 1.0 mini 7A1B powers the flash answer displayed at the top of search results. Running on Blade, Shadow’s GPU platform, it handles 3 million requests a day with three times fewer resources than its predecessor, delivering answers 20% faster. This production deployment is part of an alliance between Qwant, Shadow, European Search Perspective (EUSP), and OVHcloud, which aims to deliver end-to-end sovereign AI, hosted and operated in Europe. OVHai LLM models also rely on EUSP’s independent web index to provide sourced, up-to-date answers.

“Europe has the index, the models, the GPUs, and the use cases. What it lacked was a way to connect them,” says Boris Lecoeur, CEO of Shadow, Qwant, and EUSP.

4. The developer experience

There’s one thread that often gets forgotten in announcements: the one that runs from your terminal to all of this. You asked for it, so we give it to you.

On the experience side, the Control Panel has been completely rebuilt. The Developer Hub brings together all 6,500 API operations, billing included, with SDKs, CLI, Terraform, and changelog in one place, and is home to the Public Cloud API v2. The documentation has redesigned navigation, and Cost Center exports your costs in the open FOCUS format for your FinOps tools. Finally, OVHcloud Labs lets you test new products in alpha and beta before they’re released.

5. OVHai Workspace brings your company’s data together in an agentic platform

With OVHai Workspace, OVHcloud aims to put agentic AI within reach of millions of businesses, starting with its web hosting customers. The platform brings together three spaces. The Productivity space combines email, calendar, documents, storage, and video conferencing around a single AI, so a simple prompt is all it takes to organise a meeting. In the Web AI space, Website Builder (in alpha) generates a complete website from a few sentences. Finally, the Knowledge Platform helps your teams and your AI agents search across company data.

The Productivity space launched as a free alpha on 19 November 2026, starting in France, with 5 GB of storage per user. Your data is hosted in Europe and isn’t used to train models. Olala 1.0 mini 7A1B already powers Workspace, which you can try as a free alpha.

Compute, Bare Metal and GPU

What’s newLearn moreKey takeaways
Serverless ContainersTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/serverless/Deploy a container image from any registry, or straight from your source code. The platform builds it, publishes an HTTPS URL, handles logging, and scales to zero: no cluster to manage, and you only pay for execution.
Serverless FunctionsTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/serverless/Write a handler in the Control Panel or deploy it from a repository, then trigger it via HTTP, a cron schedule, or a Public Cloud event. One revision per version, and one-click rollback.
Bare Metal Private BootSee the product page: https://www.ovhcloud.com/en/bare-metal/ | Available in the Control PanelBoot your High Grade and Scale servers over the private network: installation and reboots no longer depend on the public network. Available since 13 November on High Grade Gen 2+ and Scale Gen 2+, in single-AZ regions (excluding APAC and Toronto).
NVIDIA B300 on Public CloudSee the product page: https://www.ovhcloud.com/en/public-cloud/gpu/b300/NVIDIA B300 Blackwell Ultra GPUs as Public Cloud instances, with nearly 50% more HBM3e memory and 1.5 times more FP4 compute than the B200. Available from November 2026 in more than eight regions.
NVIDIA B300 on Bare MetalSee the product page: https://www.ovhcloud.com/en/bare-metal/high-grade/hgr-ai-b300/A full HGX B300 node on bare metal: 8 Blackwell Ultra GPUs, 2.1 TB of HBM3e, and NVLink 5 at 1.8 TB/s per GPU, with no hypervisor and no shared resources. Root access, plus your choice of OS and CUDA driver. A trillion-parameter model fits on a single node.
NVIDIA Rubin NVL8See the product page: https://www.ovhcloud.com/en/public-cloud/gpu/rubin-nvl8/The next generation of NVIDIA GPUs is coming to Public Cloud: 8 Rubin GPUs linked by NVLink 6, HBM4 memory at 130 TB/s, up to 400 PFLOPS NVFP4 for inference, and 265 TFLOPS FP64 for simulation.

Storage and data

What’s newLearn moreKey takeaways
Media TranscoderTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/media-transcoder/Convert your videos into streaming formats (H.264, H.265, or AV1 in HLS, DASH, CMAF, or MP4) straight from your Object Storage bucket, with encryption, DRM, and AI-generated subtitles. Your files stay in your bucket and are processed in Europe.
LakehouseTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/lakehouse/Centralise your data and analyse it with SQL, BI, or AI, on open standards: Apache Iceberg tables, Trino queries, and an Apache Polaris catalogue. Hosted in Paris across 3 AZs, billed for storage and compute, with no egress fees.
Backup CenterTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/backup-center/Schedule, monitor, and restore backups of your Public Cloud resources from a single interface, starting with Managed Kubernetes. A 14-day immutable vault, and restoration of a whole cluster, a namespace, or a single volume.
Private NAS-HATry it on OVHcloud Labs: https://labs.ovhcloud.com/en/private-nas-ha/The next generation of managed NAS-HA. Your volumes connect directly to your vRack and scale to the nearest gigabyte, from 10 GB (versus 3 TB previously), with native private access and a reduced “noisy neighbour” effect.
Object Storage: new metadata infrastructureTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/future-object-storage/No more requests-per-second limits: this new infrastructure targets up to 20 times better read and write performance, with stable latency even under heavy load. Free private beta.
Object Storage Private Endpoint S3Try it on OVHcloud Labs: https://labs.ovhcloud.com/en/private-endpoint-object-storage/Access your buckets from your Public Cloud Private Network, with no public IP, no gateway, and no internet egress. The endpoint is created in minutes (Control Panel, API, or Terraform), with end-to-end TLS.

Security and trust

What’s newLearn moreKey takeaways
Cloud HSMTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/cloud-hsm/Your own cluster of three HSM partitions, spread across three separate physical devices, with no hardware to operate. Keys never leave the HSM, and the PKCS#11 driver reaches up to 1,000 operations per second with RSA 2048.
Confidential Computing (OCCAM)Try it on OVHcloud Labs: https://labs.ovhcloud.com/en/confidential-computing-occam/An attestation service for your Bare Metal servers. OCCAM verifies the processor and the booted image before releasing the keys to your confidential VMs (AMD SEV-SNP, Intel TDX), Kubernetes nodes, or AI models on GPUs. Neither the host nor OVHcloud can read the memory.
Bare Metal Encryption (SED)See the product page: https://www.ovhcloud.com/en/bare-metal/ | Available in the Control PanelEncryption at rest for your disks, using the server’s self-encrypting drives, which you can enable or disable from the server page in the Control Panel. Included at no extra cost on Scale Gen 3 (Scale 2026), in Europe and Canada.
Volume CMKProduct page coming soonEncrypt your Block Storage volumes with your own keys, hosted in the OVHcloud Key Management System. You manage rotation and revocation, and a volume becomes unreadable as soon as you remove its key.
Anti-DDoS Customer ProfilesTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/antiddos-profiles/ | Available in the Control PanelDescribe your applications to the Anti-DDoS protection instead of working around it: choose a profile (VPN, streaming, blockchain, etc.) and a sensitivity level, then attach your IPs by block. Fewer false positives and more precise mitigation, at no extra cost.
Certificate Manager & Private PKIProduct page coming soonManage the TLS certificates for your OVHcloud services and issue your own internal certificates with a private authority, alongside your keys (KMS) and secrets. Identities and certificates stay under your control.
WAFTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/owaf/Protect your web applications and APIs against injections, malicious scripts, credential theft, bots, and application-layer denial of service, with rules maintained by our experts and no configuration required. No per-request billing.

Network

What’s newLearn moreKey takeaways
VPCSee the product page: https://www.ovhcloud.com/en/public-cloud/private-network/A new regional private network foundation for Public Cloud, with your own addressing and subnets across the three zones of a region. A zone outage stays contained, and the VPC serves as the basis for WAF, VPN, peering, private DNS, and load balancing.
Application Load BalancerProduct page coming soonA regional HTTP, HTTPS, and TCP load balancer, active across the three availability zones of a region. If a zone goes down, traffic keeps flowing and the control plane stays reachable.
Private DNSProduct page coming soonPrivate name resolution inside your VPC: your instances reach your OVHcloud services (Object Storage, databases) and your applications by name, without going through public DNS.
Global Load BalancerTry it on OVHcloud Labs: https://labs.ovh.com/en/global-load-balancer/Send each visitor to the right region based on location, weight, or priority, automatically bypassing failed servers. Pure DNS, with no proxy and no TLS termination, so no extra hop on your traffic.
Site-to-site VPNTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/site-to-site-vpnConnect your Public Cloud Private Network to your data centre, a branch office, or another cloud through encrypted IPsec tunnels, with static or BGP routing. With two tunnels to the same site, failover is automatic.
BGP ServiceTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/bgp-service/ | Available in the Control PanelAnnounce your own IP addresses from several servers, with your own ASN if you have one. Failover in seconds, traffic distribution without an appliance, and migration without changing public addresses. Free during the beta.
OVHcloud Connect EdgeAvailable in the Control PanelA turnkey private link between your site and your vRack, without needing a presence in a point of presence. OVHcloud provides the local loop, the equipment, and the SLA, from 50 Mbps to 10 Gbps, with single or dual links.
OVHcloud Connect: 17 new PoPsView on our roadmapSeventeen new points of presence and new interconnection partners: your private network reaches OVHcloud as close as possible to your sites and other clouds, without crossing the public internet.
Private Endpoint KubernetesSee the product page: https://www.ovhcloud.com/en/public-cloud/kubernetes/ | Available in the Control PanelExpose the API of your Managed Kubernetes clusters on your private network only. The control plane is no longer reachable from the internet, and only your authorised networks can access it.
New weathermapLink coming soonA real-time map of the OVHcloud network (backbone links, points of presence, traffic load) so you can see at a glance how your data flows.

Private cloud and on-premises

What’s newLearn moreKey takeaways
Nutanix 3AZTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/nutanix-3az/A single Nutanix cluster stretched across the three availability zones of the Paris region, with VM and replica placement controlled by Strict Rack Awareness. Your Nutanix workloads survive the loss of a zone.
Private VCF 3AZSee the product page: https://www.ovhcloud.com/en/hosted-private-cloud/vmware/private-vcf-aas-3-az/ | Try it on OVHcloud Labs: https://labs.ovhcloud.com/en/vmware-3az/VMware Cloud Foundation 9 on a cluster stretched across three availability zones: synchronous replication, zero RPO, and automatic VM restart in the surviving zone. 99.999% SLA, fully managed by OVHcloud.
OPCP Cloud StoreSee the product page: https://www.ovhcloud.com/en/hosted-private-cloud/onprem-cloud-platform/cloud-store/ | Try it on OVHcloud Labs: https://labs.ovhcloud.com/en/opcp/The service catalogue for the OVHcloud on-premises platform. Your IT team deploys VMs, block and object storage, Kubernetes, or databases, and each hosted organisation manages its projects in self-service. Your data never leaves your premises.
OPCP KubernetesTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/opcp/managed-kubernetes/The public cloud Managed Kubernetes experience in your own data centre: clusters created via API, maintained versions, and failed nodes replaced automatically. Private alpha planned for February 2027 for OPCP Design Partners.
SNC Cloud PlatformSee the product page: https://www.ovhcloud.com/en/solutions/uc-secnumcloud-environment/The agility of the public cloud in an environment pursuing SecNumCloud qualification: multi-tenant, Infrastructure as Code, a self-service portal, and pay-as-you-go billing, to host and modernise your sensitive applications and data.

Experience and tools

What’s newLearn moreKey takeaways
New Control PanelExplore the Control Panel: https://console.eu.ovhcloud.com/A completely rebuilt Control Panel: unified navigation across products, shorter ordering and configuration journeys, and a faster, clearer interface.
Developer HubOpen the Developer Hub: https://developers.eu.ovhcloud.com/The single reference for OVHcloud APIs: 6,500 operations documented in one place, billing included, with natural-language search and ready-to-copy calls. SDKs, CLI, Terraform provider, and changelog are built in.
Public Cloud API v2Visible in the Developer HubA new way to explore the Public Cloud APIs, directly in the Developer Hub.
DocsBrowse the documentation: https://docs.ovhcloud.com/en/Product documentation with a new navigation experience. Guides, tutorials, and references have been reorganised so you can find the right page faster, from first steps to automation.
Cost CenterAvailable in the Control PanelSee what your OVHcloud services cost and export your cost data in the open FOCUS format, ready to use in your FinOps tools.
OVHcloud LabsExplore OVHcloud Labs: https://labs.ovhcloud.com/en/OVHcloud’s pre-launch channel. Test alphas and betas on real workloads, talk directly with the engineers, and help shape what gets delivered.
SMTP Mail OutProduct page coming soonA Public Cloud service for sending outbound emails from your applications.

AI services and applications

What’s newLearn moreKey takeaways
AI FabricTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/ai-fabric/The GPU becomes an argument in your Python function. Choose the card model and how many you need in one line, the CLI runs it remotely, and billing stops with the container. Developed with Blade.sh, with GPUs in France.
AI Endpoints: Gladia modelsSee Solaria in the catalogue: https://www.ovhcloud.com/en/public-cloud/ai-endpoints/catalog/solaria-3/Gladia’s transcription models (Solaria) join the AI Endpoints catalogue. Transcribe audio in more than 100 languages with the same account and the same API family as your other AI calls, hosted in Europe.
AI Endpoints: Shadow partnership (Blade.sh)Available in the Control Panel (AI Endpoints) | Try Blade: https://play.blade.sh/Each partner exposes the other’s models: the AI Endpoints catalogue gains dozens of additional models, all on a single invoice.
Gladia: voice as an interfaceAvailable in the Control Panel (AI Endpoints)Real-time and batch transcription in more than 100 languages, which you can combine with AI Endpoints LLMs to transcribe, summarise, analyse, and then act.
AI RoutersTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/ai-routers/A single OpenAI-compatible endpoint for all your AI traffic. The router sends each request to the right model based on your rules (a small model for a simple question, a coding model for code, etc.), with automatic fallback on failure. Your applications don’t change.
AI AgentsTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/ai-agents/Describe your agent, give it a model, tools, and permissions, and it’s deployed: persistent memory, MCP connectors, Slack and Telegram, web search, scheduled tasks, and human approval for sensitive actions.
AI Search (RAG)Try it on OVHcloud Labs: https://labs.ovhcloud.com/en/ai-search/Managed RAG. Bring your documents: AI Search analyses, chunks, and indexes them, then returns the relevant passages to your applications via API or to your agents via MCP. Your knowledge stays under European jurisdiction.
OVHai WorkspaceTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/workspace/Email, calendar, storage, documents, and video calls in one collaborative space where AI turns your requests into actions. Hosted in Europe, with no training on your data.
Website BuilderTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/website-builder/ | Try the demo: https://builder.website.ovh/Describe your business in a few sentences and the AI creates a complete website (pages, copy, style), then updates it on simple instruction. One-click publishing, with domain name, SSL certificate, email, and backups included.
ReceptionistTry it on OVHcloud Labs: https://labs.ovhcloud.com/en/receptionist/An AI phone receptionist that answers day and night, understands what callers need, transfers them to the right person, and creates appointments and tasks in your tools. No more missed calls.

Olala model family

What’s newLearn moreKey takeaways
Olala 1.0 mini 7A1BOpen the Olala playground: https://olala.ovh.comThe family’s compact, open model: 7 billion parameters, with about 1 billion active per token. Designed for agents, it answers accurately, quickly, and at a constant cost, millions of times a day.
Olala 1.0 TranslateOpen the Olala playground: https://olala.ovh.comThe translation-specialised variant: a small model trained for this task, which translates better and more cheaply than a much larger general-purpose model. Open-weight, served on AI Endpoints, and fine-tunable on your data.
Olala 1.0 SummariseOpen the Olala playground: https://olala.ovh.comThe summarisation-specialised variant: condense documents, tickets, and conversations at scale. Open-weight, served on AI Endpoints, and fine-tunable on your data.
Olala 1.0 BaseOpen the Olala playground: https://olala.ovh.comThe pre-trained base model, without alignment: the starting point for building your own variants through fine-tuning. Open-weight, and customer data is never used for training.
Olala 1.0 SFTOpen the Olala playground: https://olala.ovh.comThe instruction-tuned version of Olala, refined through supervised fine-tuning to follow instructions and hold a conversation. The model to use as is when you don’t need specialisation. Open-weight, served on AI Endpoints.
Olala on AI TrainingRead the guide: https://docs.ovhcloud.com/en/guides/public-cloud/ai-machine-learning/ai-training-fine-tune-olala-miniTrain Olala on your own datasets, for your use case, from a ready-to-use Docker image.
Olala on AI DeployRead the guide: https://docs.ovhcloud.com/en/guides/public-cloud/ai-machine-learning/ai-deploy-fine-tuned-olala-miniDeploy your fine-tuned Olala model without writing extra code or building an image yourself.

Share on: