Together AI is planting deeper roots in Europe. The company announced today that it has gone live with GPU infrastructure in Sweden while also launching a free workshop series to help AI engineers get hands-on with customizing open-source models. The dual push underscores the company’s bet that Europe’s AI boom will demand both high-performance compute and upskilled talent.
A Nordic Node for High-Performance AI
The new Sweden region becomes the latest addition to Together AI’s global GPU network. Customers can now spin up serverless inference APIs for models like OpenAI’s gpt-oss, Meta’s Llama, Qwen, and DeepSeek—or request dedicated GPU clusters directly in-country.
The location isn’t just about geographic diversification. By situating compute resources in Sweden, Together AI is targeting three critical needs:
- Data sovereignty: European firms juggling GDPR and data residency requirements can keep sensitive workloads inside the EU.
- Latency gains: Locating inference closer to Northern and Central Europe trims round-trip times by 50–70 ms, which Together AI says can translate into 25–30% faster response times for real-time AI applications.
- Scalable options: Customers can mix dedicated capacity with serverless elasticity, giving startups and enterprises alike more flexibility.
Mark McKenzie, founder of AI startup Caesar, said the company runs its high-concurrency knowledge platform on Together AI’s dedicated endpoints and plans to shift workloads to Sweden to cut latency and satisfy in-region data governance.
Training the Builders, Not Just the Models
Infrastructure is only one half of the story. Together AI is also launching a new AI upskilling workshop series, kicking off September 10 in Amsterdam. The first session, titled Upgrading and Customizing Open Models, dives into some of the more technical levers for getting open models to rival proprietary systems at lower cost.
The program covers:
- Fine-tuning and optimization: Post-training steps such as SFT (supervised fine-tuning), preference optimization, and reward-based training.
- Speculative decoding: A method for accelerating inference by pairing models, which Together AI claims can deliver 1.85× speedups on DeepSeek-R1.
- Quantization: Techniques to compress large LLMs, lowering compute costs and enabling deployment on smaller devices.
CEO Vipul Ved Prakash said the expansion reflects Europe’s role as “at the frontier of AI innovation” and emphasized that pairing infrastructure with education is key to supporting developers, researchers, and institutions.
Europe’s AI Arms Race
Together AI’s moves mirror a broader industry trend: hyperscalers and AI infrastructure startups are racing to add regional compute hubs to address sovereignty, latency, and sustainability concerns. Amazon, Microsoft, and Google have all announced new European data regions in recent months, while startups like Stability AI and Mistral continue to advocate for open models tailored to EU markets.
By blending GPU availability with workshops on customization, Together AI is positioning itself not just as another infrastructure provider but as a partner for Europe’s AI engineering community. Whether that strategy helps it stand out amid giants with deeper pockets remains to be seen—but for developers in Europe, having a new node in Sweden and a free seat in Amsterdam might be reason enough to pay attention.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
