Do you remember where you were on November 30, 2022? Many investors do. On that day, OpenAI introduced ChatGPT-3, and many informal investors, venture capitalists and financial adventurers were immediately aware that artificial intelligence (AI) would become big in the coming years. Very big. In the first few months they seemed to be correct: ChatGPT had a million users within a week and 100 million within two months, making it by far the fastest growing app ever. The share prices of companies such as Microsoft (major shareholder in OpenAI) and Nvidia (which supplies the chips for applications such as ChatGPT) shot up.

But from late 2023 to early 2024, doubts crept in, mainly due to the discrepancy between gigantic investments that generative AI requires and the operational costs on the one hand and the revenues of AI companies on the other. For example, training the Large Language Model (LLM) that ChatGPT-3 launched in 2022 appears to have cost around 5 million dollars, and in the first months of 2023 it costed around 700,000 dollars daily to answer questions with ChatGPT. The company's turnover in 2023 was 'only' 700 million dollars, reports The New York Times. However, the forecasts are full of optimism. OpenAI is hoping for 2.7 billion dollars in revenue this year and 100 billion dollars by 2029. A recent investment round valued the company at around 157 billion dollars – more than three times the value of ING Bank.

But judging by the share prices of Microsoft and Nvidia that have not yet fully recovered to previous heights, there is still doubt among investors. And there are still cold feet among investors in unlisted AI companies. Investor Johan van Mil, co-founder and partner of venture capital company Peak Capital, even speaks of a binary market: “The large AI companies such as Nvidia and the providers of large language models attract enormous amounts of capital. Other AI companies are having a much more difficult time.”

This also applies to the approximately 750 AI companies in the Netherlands, according to the report AI Scaling Challenges for Dutch Founders by accounting and consultancy organization Deloite and Techleap, an organization that stimulates technological innovation. These often concern starters in a phase where product development and coordination with the market are central. These companies – together around 10 percent of all Dutch startups – often have difficulty attracting investments, according to the Techleap report even in 67 percent of the cases.

“Nevertheless, I expect that we will see higher investment amounts for AI companies in the Netherlands next year”, says Thomas Mensink, startup analyst and CEO of venture capital company Golden Egg Check. According to him, there is also every reason to be optimistic about the prospects for the sector: “AI can help companies increase turnover, reduce costs and improve margins. Every company should be working on AI in some way, whether it is internal process optimization or developing AI products for customers.”

In fact: “If you don't use AI as a company, you will lose out”, says Bas Rieter, investor at the Dutch Founders Fund (DFF) and the Samen Slimmer alliance of the DFF, Slimmer AI and Keen Venture Partners. AI is the future, according to him: “100 percent sure.” And that justifies further investments in AI companies. Van Mil thinks exactly the same: “AI is not just a trend. Artificial intelligence has become an essential component for any software company that wants to be taken seriously. Without AI in your core functionality, you run the risk of falling behind and not being able to find investors at all.”

Rieter thinks it’s only a good thing that the hype has passed its peak: it means that a time of realistic valuations has arrived. Although investors must still be cautious of ‘AI-washing’, as Rieter calls it: they need to separate the wheat from the chaff – companies that use the term AI merely to mislead investors – from genuine providers of valuable AI applications and companies that successfully implement AI (think of Klarna, which can now operate with hundreds fewer employees thanks to the use of AI). Most of the ‘wheat’ will typically be AI application providers, especially since there are still relatively few companies that can demonstrate clear successes with AI use.

When it comes to those AI applications, most experts distinguish roughly 3 types that may offer opportunities for investors.

The application layer is also so popular because the business case here is concrete and the payback period is shorter than in deep tech with its long investment horizon. Things that investors pay special attention to. Like Van Mil: “We prefer companies that offer a 'painkiller', i.e. a solution to an acute problem for customers and for which they are willing to pay.” This is in contrast to companies that offer 'vitamins' to prevent possible problems in the future.

There is no shortage of such ‘painkillers’. "You can apply AI to virtually any work process", says JC Heyneke, CEO of Slimmer AI, an investor that can provide companies not only with funding but also with software engineers. Slimmer AI – which, like the DFF, is part of Samen Slimmer – invests in and builds ‘vertical applications’: AI applications designed for specific industries or professions, characterized by user-friendliness, integration into existing workflows, and domain expertise. For example, Slimmer AI has invested in EvidenceHunt, a company that creates software to make the outcomes of medical research more accessible to researchers and pharmaceutical companies. Curiosity also favors companies that focus on the application layer, like Strise, a Scandinavian provider of intelligent software for screening business clients in line with anti-money laundering regulations.

One of the 'painkillers' that Peak itself has in its portfolio is HealthSage AI, a company that focuses on making unstructured health data searchable with AI, which can reduce the administrative burden for doctors and accelerate patient care. Because HealthSage AI uses patient data and not the data offered by the major providers of LLMs, HealthSage AI also saves significant costs. Another advantage is that the large LLMs themselves will not be so likely to come to the market with a comparable application.

Brouwer sees opportunities in three areas in the short term, she says. Starting with services, where tasks can be taken over by software. For example, Brouwer mentions AI consultancy companions that can partly take over the work of consultants. A second category is AI aimed at high-value and/or labor-intensive tasks. For example, Curiosity VC has invested in QA Tech, which offers applications that can perform quality checks on software and thus relieve developers of work, and in Neople, which develops virtual, AI-driven employees that can be deployed in the customer service department. Thirdly, Brouwer sees potential in AI applications that can unlock difficult-to-access or unstructured internal and external data. Consider applications to extract data from e-mail boxes and make it usable.

Certainly in the long term, AI can also be used for completely new applications, Brouwer thinks. The pioneering work of an AI startup like Cusp.ai may offer a preview. This company of the well-known Dutch machine learning expert Max Welling and the British chemist Chad Edwards has developed a platform that can discover new materials to bind CO2 and remove it from the atmosphere. “A crazy idea that shows how AI can be used to solve problems that previously seemed unsolvable", says Brouwer. She is not the only one who sees the potential. In June, Cusp.ai received an initial investment of 30 million euros from Hoxton Ventures and managers from Deepmind, Google's AI division.

Raising capital is a challenge, say the various venture capital experts. Especially when it comes to larger investment amounts. There are exceptions. For example, look at the investment at the beginning of this year in the Amsterdam Datasnipper, a company that can take over all kinds of tasks from accountants (think of searching for data in all kinds of documents, cutting it and pasting it into Excel): the American Index Ventures paid 100 million for a stake of 10 percent in the company.

“A financing round of 100 million or more is very difficult, perhaps even impossible, with only European players”, says Sander Verbrugge, partner of Innovation Industries. In doing so, he highlights one of the biggest threats to the European AI industry – companies, investors, buyers, employees and other parties in the 'AI ecosystem'.

According to Verbrugge, the lack of European capital can have far-reaching and particularly unpleasant consequences for Europe. “Foreign investors are gaining stronger control over European companies, which means that the weight and orientation of such a company can quickly shift abroad”, he says. And dependence on foreign capital means that European vulnerability is increasing: “At the moment the world becomes complicated, you run the risk of investors withdrawing to their home market, as happened during the corona crisis.”

But even without such a crisis, this dependence is already bad news: it quickly leads to European AI companies shifting their focus to other markets, selling their intellectual property to foreign companies or even being completely taken over by foreign parties or establishing a business and get a stock listing elsewhere. As a result, the value that these companies create does not benefit the European economy and society. It could slow down Europe's ability to innovate and lead to a 'brain drain': an exodus of many talented AI researchers and developers continuing their work in the US or elsewhere outside Europe.

Verbrugge sees the lack of players capable of competing with the power of American investors as a major gap in the European/Dutch investment landscape. However, there seems to be a gradual increase in the number of deep-tech funds; Verbrugge mentions DeepTech XL, which, like Innovation Industries, is based in Brainport Eindhoven. Governments are also increasingly recognizing the importance of AI, which requires a long-term investment horizon. For instance, InvestNL has invested money through its deep-tech fund into Axelera and Innatera, which designs microprocessors that mimic brain mechanisms to process sensor data.

But more is needed, according to Verbrugge. He believes the solution lies in building ‘a robust European investment climate’. He envisions a role for pension funds and other institutional investors with deep-tech investors. He acknowledges that "pension funds are not ATMs" and will not simply invest directly in AI companies. Nonetheless, Verbrugge is optimistic. If more AI companies, especially specialized deep-tech funds, achieve success and a mature ecosystem develops around these players, he expects institutional parties will be willing to invest in those funds. This is already happening with the third fund of Innovation Industries, where, following the participation of PME and PMT in earlier funds, several pension funds are now participating. However, this will take some time.

According to the Techleap report, public investments could also be structured more efficiently, for example by speeding up procedures or by consolidating government funds to enable larger investments. Techleap also believes that tech entrepreneurs should be involved earlier in government policy processes so that outcomes are better aligned with real-world needs. All three layers of AI companies should be supported, according to Maxime Lübbers, AI Ecosystem Lead at Techleap. She advocates for a ‘diversified approach’, with relatively more focus on deep tech. "Because its applications are still relatively unclear to many investors, even though its impact is significant. If, in the long term, we want to remain strategically autonomous as a country, we need this type of company."

To prevent investments from becoming fragmented, Techleap suggests prioritizing specific sectors where the Netherlands is already strong, such as healthcare, the energy sector, and food. Lübbers explains, "We already have relatively many applications and a lot of data available. So we can also focus more on that as a country, and in doing so, strengthen that ecosystem."

Mensink believes that entrepreneurs who have sold their companies and reinvest the proceeds in AI can help initiate the investment cycle – investing, growing, selling, and reinvesting. Kudos to Adriaan Mol! Brouwer also has confidence in structures like the one Curiosity has chosen, with funds linked to advisors and (former) entrepreneurs who contribute both capital and knowledge, and who also form a ‘community’ with the founders of Curiosity’s portfolio companies. Heyneke and Rieter point out that collaborations such as Samen Slimmer also allow for ‘larger tickets’ than individual participants can afford. Moreover, it’s a way for funds to diversify their risks, adds Mensink – the main reason why his Golden Egg Check always co-invests with other funds, thereby building a portfolio of 35 participations.

Another major threat, according to Verbrugge, is the enormous energy consumption of AI applications. According to analysts, growth of 8,000 percent is in sight between 2024 and 2030. Here, global energy demand due to AI will rise to 16 percent of current US energy consumption, according to an article in Forbes, at least if those growth forecasts from OpenAI and Nvidia are representative of the coming years. Sam Altman, CEO of OpenAI, once said that he believes nuclear energy is the solution – and that ‘the marginal cost of intelligence will go to zero thanks to AI because the marginal cost of energy will also go to zero.’

Verbrugge thinks that even if there is a solution to supply energy at marginal costs, this solution will come too late to save the world from an energy infarction. He expects that innovative applications, such as the chips from Axelera and Innatera, can provide some relief, but the trend is worrying. Verbrugge is not happy about it. He hopes for other innovative disruptive technologies to generate and store energy cheaply, so that AI can grow without a major increase in energy consumption

The shortage of highly trained personnel is also sometimes mentioned as a threat. Verbrugge recognizes this, but puts it into perspective. He sees Dutch universities as a purveyor of 'real scientists' – physicists, materials scientists, chemists and other talented 'deep tech people' that companies from all over the world are after. And thanks to the presence of companies such as ASML, NXP and Philips in the Netherlands, there is sufficient employment for them. And if they want to change jobs, the doors of Innovation Industries' portfolio companies are open to them. In addition, many experts say that the Netherlands is still attractive to many foreign workers, especially as long as the 30 percent scheme for expats is not completely abolished.

Investors in the AI application layer are considerably more gloomy. Bas Rieter puts it as follows. “It is now a lot more difficult for a startup to gain a competitive advantage from technology. Where previously the distinctive character of a tech company lay in the tech, everyone can now build on open-source models or GPT-4 apps. In a sense, tech has now become a commodity.”

Given the rapid pace at which AI is developing, speed is now the distinguishing feature for many investors. “How quickly you can enter the market to collect unique data and thus train your model is of vital importance”, Rieter notes. If possible, with an extensive network that provides investors with valuable information that can accelerate the 'time to market' of AI applications. All the more reason for Samen Slimmer to provide not only financial resources but also people who can help accelerate the development of AI startups. Because who knows, OpenAI or another major provider may come up with a competing product in a few months – and you'll be gone as a slow, small party.