Case study · Qwant x Shadow GPU
How Qwant powers 100% of its production AI with Shadow GPUs — at ~50% lower cost
Qwant, the privacy-first French search engine, needed to embed generative AI in search results without compromising user data or blowing up infrastructure spend. By scaling from GTX 1080s to RTX A4500s on Shadow's OpenStack platform, the team now runs every production AI workload on Shadow GPUs while cutting costs nearly in half.
Production coverage
100%
All production AI models now run on Shadow GPUs.
Live GPUs
90+
RTX A4500 fleet powering NER, embeddings, categorization, query rewriting.
Cost reduction
~50%
Lower GPU cost vs comparable market rates.
Transforming search with generative AI — without sacrificing privacy or budget.
Challenged with growing technological needs, companies seek reliable, accessible, and cost-effective solutions. For Qwant, a French search engine renowned for its commitment to protecting user privacy, Shadow Pro has emerged as the perfect partner to overcome its computing power challenges.
The Challenge: Transforming Search with Generative AI. Founded in 2013, Qwant has differentiated itself by refusing to exploit personal data for advertising purposes. The company focuses on two main areas of development: its web page display and its complex search engine, which manages millions of stored URLs.
Qwant’s mission is to transform the user experience through a versatile tool. To achieve this, the company has integrated generative AI directly into its search results to offer intuitive summaries of relevant answers. However, training these AI models, processing data, and indexing web content creates an ever-increasing need for computing power. Qwant required a robust infrastructure capable of managing intensive workloads to meet these requirements.
Scaling up: from GTX 1080s to RTX A4500s
Qwant initially leveraged Shadow's infrastructure by using older units, specifically 1080 GPUs, which allowed them to rapidly mobilize resources for internal purposes such as crawling tests on millions of web documents. This strategy allowed the company to reduce its budget while benefiting from quickly available performance.
To support production-grade AI, Qwant has now transitioned from 1080s to A4500 GPUs for all its models. Today, over 90 GPUs are being utilized to run a much wider variety of complex model types, including:
- NER (Named Entity Recognition)
- Embeddings
- Categorization
- Query rewriting
Impact: 100% of production AI on Shadow GPUs
The result of this scaled partnership is a testament to the reliability of the infrastructure: 100% of Qwant's production models now run on Shadow's GPUs. Qwant can access these resources via OpenStack in just a few clicks, ensuring instant computing power while guaranteeing security and confidentiality via SSH connections.
“With Shadow, we can respond immediately to load spikes! Which is essential for a company like ours.” — Ludovic Logiou, Lead SRE at Qwant
Cost control and expert support
Shadow Pro’s GPUs are approximately 50% cheaper than the competition: while other providers typically charge around €500 to €600 for Qwant’s configuration specific needs, Shadow offers the same power for only €250. This significant price advantage, combined with a commitment-free subscription, allows Qwant to access high-end hardware without unpleasant surprises.
There are no additional costs for bandwidth or IOPS; only the computing resources ordered are billed. This has allowed Qwant to deploy a massive fleet of GPUs while keeping its budget strictly under control.
Furthermore, the collaboration offers dependable support and professional insights. Shadow’s hardware expertise has been invaluable to Qwant, notably in helping the team quickly resolve technical issues such as Nvidia driver errors.
About Qwant & Shadow
About Qwant
Qwant is a French search engine built on privacy, refusing to exploit personal data for advertising. Its team is integrating generative AI to deliver concise, trustworthy answers directly in search results.
About Shadow Cloud
Shadow Cloud delivers GPU infrastructure tuned for AI and rendering, giving engineering teams flexible access to right-sized NVIDIA GPUs through an OpenStack-based platform with transparent, usage-based pricing.