At HubX, we’ve always prioritized transforming AI models into production-ready systems that are not only high-performing but also cost-efficient. This vision has been fueled by our close collaboration with Google Cloud, giving us early access to cutting-edge technologies and the opportunity to test and adopt them ahead of the curve.
By deeply integrating Google Cloud’s latest TPU architecture (Trillium) and leveraging the Google Kubernetes Engine (GKE) to its full potential, our teams have been able to design infrastructures that meet the demands of modern AI workloads without compromising on speed and scalability.
The results speak for themselves:
~35% reduction in latency
~50% faster cold start times
~45% overall cost savings
These gains represent not just technical milestones but a validation of how strategic infrastructure choices directly impact real-world performance and budget.
Together with Google Cloud, we took a deep dive into the deployment of large-scale AI systems using TPUs and GKE, unpacking the innovations and architectural decisions behind them at Google Next 2025. The session delved into the strategic use of Google's TPU (Tensor Processing Units) and the deployment on Google Kubernetes Engine (GKE), offering a deep dive into the technical intricacies and innovative approaches that power these AI solutions.
Representing HubX on stage was Mustafa Ă–zuysal – ML Researcher. Also joined us at Google Next 2025 and played a key role in this journey: Deniz Tuna – Head of Development, Yunus Emre, PhD – Head of AI Lab, Kaan OrtabaĹź - Co Founder, Levent Uysal – Backend Team Lead, and HĂĽsnĂĽ Sebik – ML Engineer.Â
We’re proud to share the stage and be part of this global conversation and even prouder to share our learnings with the community. Excited to showcase what’s possible when the right technology meets the right strategy. 🚀
Curious about what we’re building next?
Join us! http://hubx.co/jobs
See you at Google Next 2026. 👋🏻