This year, I attended NVIDIA GTC in Berlin for the first time. It was an excellent opportunity to see how AI infrastructure, software, and industrial applications are evolving – and to get a glimpse of what the future may hold.
My professional career has been centred on software engineering and architecture, so one of my key takeaways was seeing the parallels between software systems and AI infrastructure.
Both increasingly rely on well-designed components, clear interfaces, testing, and scalability. Individual components matter, but real value emerges when they work together as part of a larger, reliable system.
NVIDIA is investing heavily in AI hardware while building a broad software ecosystem around it. Its software, models, developer tools, and open-source projects help developers build applications and accelerate adoption of its computing platforms.
This reinforces an important principle: the value of technology does not come from individual components alone. It emerges when hardware, software, data, and domain-specific capabilities work together to solve real problems.
For industrial AI, this raises an important question: how do we connect increasingly capable AI systems to the operational knowledge, equipment, and processes of a real business?
One topic from Jensen Huang’s keynote particularly caught my attention: the potential of telecommunications infrastructure as a platform for distributed computing and AI.
It connects with an idea I have been exploring from a logistics perspective: bringing computing and AI closer to where operational decisions need to be made, including warehouses, distribution centres, and manufacturing facilities.
The potential is significant, but so are the engineering challenges. Latency, connectivity, security, reliability, and integration with existing systems all matter. AI capabilities need to work within operational environments where downtime and unpredictable behaviour have real consequences.
These are not just AI challenges. They are system architecture challenges – and they become increasingly important as intelligence moves from the cloud into physical environments.
The progress in humanoid robotics was another highlight for me. Logistics and manufacturing are particularly interesting because they combine physical work with complex operational processes.
During the GTC discussion, Mercedes-Benz CEO Ola Källenius described the transition from demonstrations to pilot deployments. He also discussed implementations of humanoid robots in logistics at Mercedes-Benz’s new factory in Hungary and expects implementation to expand into logistics and selected manufacturing applications over the next 24 months.
That timeline is significant. It suggests that the conversation is moving beyond what humanoid robots might eventually do towards where they can deliver practical value in the near future.
I also spoke with several humanoid robot manufacturers, and logistics and manufacturing were prominent target markets.
But there is a question I kept coming back to: what does it take to move from a robot that can perform a task to a robot that can reliably contribute to an actual operation?
A real warehouse is not a demonstration environment. Neither is a sortation hub or a factory.
Robots need to work alongside other equipment, interact with operational software, fit into established workflows, and handle exceptions. They must operate within safety requirements and coexist with people and other machines. Their performance needs to be evaluated in the context of the entire operation, not just an individual task.
The robot itself is only one part of the system. The surrounding processes, interfaces, controls, and integration determine how effectively it can be deployed.
This is where industrial experience matters.
Another idea from Jensen Huang’s keynote resonated with me: practical AI applications emerge when general-purpose models are connected to specialised knowledge, relevant data, and tools.
Not every application requires a fine-tuned model. Sometimes the key is providing the right context, connecting existing systems, and giving AI the tools it needs to perform a specific task.
In industrial environments, that context is substantial. It includes operational processes, equipment behaviour, business rules, existing software, and experience accumulated over years of running complex operations.
At Comtrade Fastforward, this is where our experience in industrial software and system integration becomes relevant.
We understand that delivering software for logistics and manufacturing is not simply a matter of implementing a feature. It means understanding how the operation works, how systems interact, and what happens when reality does not follow the expected path.
As AI and robotics capabilities advance, connecting them to this operational reality becomes increasingly important.
NVIDIA GTC offered a compelling view of the technological foundations being built today. The next challenge is turning those capabilities into dependable, valuable applications across logistics and manufacturing.
How do we make AI and robotics work reliably in real warehouses, sortation hubs, and factories?
Answering that question requires more than capable models or impressive machines. It requires engineering, integration, domain knowledge, and a clear understanding of the operational problems being solved.
That is the intersection I am most interested in – and where I believe companies with real industrial software experience can make a meaningful contribution.
The technology is advancing quickly. I look forward to seeing which use cases emerge next and, more importantly, how we turn the most promising ideas into working solutions.
