AI: “Companies Need To Move Quickly To Stay Competitive”

Interview with Dr David Moser, Managing Partner at Becquerel-Institut Italia

In your presentation at the 2025 Intersolar Europe Conference, you described the current shift from predictive to generative AI, as well as the increasing use of AI agents. Could you please explain this shift to us, where we currently stand, and how you expect this process to unfold in the future?

We’ve been using what I call predictive AI for years. It relies on training data to predict outcomes. Generative AI, on the other hand, can learn from data and from there extrapolate and create new solutions. The two key aspects that make it revolutionary are its creativity and accessibility thus democratizing the access to AI. It can produce human-like responses, and anyone can use it through natural language, not just IT experts.
In the PV industry, this means humans can use natural language to better understand how a plant performs and make better decisions. Generative AI can adapt to different contexts — market segments, technologies, or site conditions — without needing complete retraining on new data but rather using domain-specific fine tuning. We already see many startups emerging across the value chain for all sorts of use cases, and the pace is accelerating. But not all ideas will succeed.

So there is a lot of movement in the industry, a lot of startups and new companies rising, but you expect not all of the ideas or projects to be successful on the market, right?

Yes, exactly. Many ideas are good, but differentiation will be key. In monitoring, for example, companies offer diverse solutions at the moment, all of them with their own corporate identity, dashboards, the way they plot data, the way you can visualize data etc. When these tools move toward generative AI, users will simply ask questions like “How is my plant working?”, so products will need to stand out by performance (for example model accuracy, data quality, integration capabilities, and domain expertise), not by individual distinctions. This explains why not all these new companies will survive. But it's also true that existing companies, if they don't embrace AI, they might find it difficult to retain their position in the market. Both sides need to move quickly to stay competitive.

You also highlighted the report's findings on potential automation levels, which range from 60 to 90 per cent across the different stages of the value chain. That sounds truly disruptive! But how realistic is it to exploit those potential levels, and what obstacles need to be overcome to do so?

Those numbers surprised me too: they’re high, but realistic as a potential, and obviously there will be positive and negative consequences. We assessed each phase of the value chain by asking how repetitive or structured tasks are, whether they require physical work, and how complex decisions are. The results reflect the maximum theoretical automation level we expect by 2030, assuming steady technological progress.

Some segments will reach those levels sooner than others. Key barriers include the pace of implementation of a certain technology, because robotics lags behind AI, as well as implementation costs, especially for small firms. Regulations are another obstacle: for instance, drones still require human pilots. For the use case of AI-generated documents in permitting, legal reliability remains still unclear. Who's liable if an AI-generated design fails to meet code requirements? Until that's clarified, adoption will be slower. Finally, workforce transition is crucial. Some jobs will disappear, but we’ll also need many AI and robotics experts — and we may not have enough of them yet.

What are the key improvements and opportunities manufacturers, developers, service providers and asset managers can achieve in their business models using innovative AI applications and robotics?

Automation mainly reduces the time needed for tasks. In Operation and Maintenance (O&M), for example, what now takes weeks could be done in hours with automated fault detection, diagnosis, and intervention. This cuts costs and boosts efficiency. The goal isn’t to reduce staff, but to manage larger portfolios or improve margins with the same workforce, or even hire more people to expand your business.

For manufacturers, where we currently already have very high levels of automation, the next step is decision automation — production optimization, quality control, and supply management handled autonomously.

For developers, AI can speed up permitting by generating documents automatically, while authorities could use the same tools to review them faster. Generative design is another promising area: imagine designing a PV plant through natural language prompts and optimizing it instantly. Asset managers can use AI agents for automated reporting and performance analysis.

What impact could the increasing use of AI have on the shortage of skilled workers in the solar industry? Could there also be a negative impact?

I can see the message can be rather scary. As an example, let's take the one where it's less obvious that we can fully automatize, which is building a PV plant: If automation rises from the actual 10% by 50% to 60%, it seems we need 50% fewer people. But that’s too simplistic. Many workers can be retrained to focus on complex analysis, data quality, and supervision instead of repetitive tasks. So rather than a net job loss, we’ll see a transition toward higher-skilled roles – it actually means that people are becoming more important than before as the nature of their contribution becomes higher-value!

Not everyone can or will be reskilled though, and per megawatt installed, we may need fewer people. But since the PV market keeps growing exponentially, AI and robotics actually help sustain that growth despite labor shortages. Managed properly, the overall effect will be positive.
A new challenge is sociological: how people adapt to working alongside AI and robots, and whether they accept taking instructions from non-human systems. Coordination between human and AI agents will be both technically and psychologically complex.

Could you describe some of the key outcomes of your report “Transforming the PV Sector: The AI & Robotics Revolution”?

At the end of the report, we use an analogy with autonomous cars. Initially, humans supervised them closely; now, they operate independently and alert control rooms only when needed. PV plants will evolve similarly: becoming intelligent, autonomous systems through AI, drones, and robots. Pilot projects could appear soon, and large-scale adoption may follow within five years, though cost-effectiveness will take time.

However, as systems become fully autonomous, cybersecurity and control become critical. If someone gains access, they could manipulate or even harm the grid. So autonomy brings great potential, but also serious risks: The implementation of AI and Robotics is not as easy as I say.

Other key outcomes are that through these new automation levels, the entire PV sector will be reshaped by AI and robotics. We therefore recommend allowing sandboxing to safely test new concepts and supporting workforce reskilling and transition programs. For companies, the key advice is to embrace AI early and develop an internal AI roadmap to identify high-ROI use cases. Waiting too long could mean being overtaken by AI-native competitors.

Would you like to find out more about the AI and robotics revolution, and how these solutions can be adapted for your business? The full report 'Transforming the PV Sector: The AI & Robotics Revolution' of Becquerel Institute is available to purchase here.

If you have any specific questions about the report, you can also ask Becquerel's AI agent, SolarIntelligence.ai.