March 6, 2023

Europe’s AI weaknesses could matter less in generative world, says Insight Partners

Insight Partners' Lonne Jaffe gives his take on the state of European AI innovation, which he says could benefit from generative models

Tim Smith

4 min read

Europe is far behind in the AI race; of the 10 best-funded startups in the space, none are European, while the biggest heavyweights in the space — Google and Microsoft — are US companies.

But, according to New York-based Insight Partners’ managing director Lonne Jaffe, Europe may not be at such a disadvantage in this new era of generative AI.

Insight has backed big AI startups like JasperAI and Weights & Biases , while these big-ticket investments are US-based, Jaffe has his fair share of skin in the European game.


His investments on this side of the pond include AI-powered insurtech Tractable, cybersecurity platform Featurespace and trading platform TradingView — but he makes no bones about Europe’s shortcomings when it comes to building AI.

Historically hampered

Europe has had two structural problems when it comes to AI innovation, says Jaffe.

The first is regulation around the use of data — which is much stricter in Europe, making it harder to train machine learning models on lots of information than in the States. And then there’s the shortage of talent (much of which has relocated to the US).

But Jaffe believes that both of these problems will be less of an issue in the generative AI context. 

Fine tuning

That’s because much of the value from generative AI going forward will be created by companies fine-tuning pre-existing models, he says.

This, he argues, is a task that's far less technically demanding than building previous types of AI functionality into a business. And it’s something where European companies could take a lead, even if they might struggle to catch up with the likes of OpenAI when it comes to training massive models. 

“It doesn't seem to be that hard to fine-tune these models. I had one example recently of talking to a large multinational corporate that you would not expect to have a deeptech team of any kind,” he explains. “In this generative AI use case their people — who were relatively not very technically sophisticated — fine-tuned a version of GPT-3 in [Microsoft] Azure, in about two weeks.”

Jaffe says this company was able to build a useful generative tool for navigating its document database, without needing to hire machine learning specialists.

He adds that the data regulation issue is also less important in the generative AI context, as a small amount of proprietary data can be used to fine-tune a pre-existing model, with great results.

“In the generative AI world many of these use cases require much less data to be able to take advantage of the systems,” says Jaffe. “You can just take some of your own private data, fine tune a model from one of the large foundation model companies and you can get a pretty powerful system. You don't need to have any specialised talent or to manage any specialised machine learning infrastructure.”

Picking battles

Jaffe also says that there might be some advantages that come from the open source strategies being led by European companies like Stability AI and Aleph Alpha, with one example being Apple’s integration of Stable Diffusion (Stability AI’s image generation tool) into its own processor.


“They took a version of Stable Diffusion and got it to work really well on their specialised chips on the iPhone. Having it be open source means that Apple can really get into the internals and play around with the nuance and subtlety of how it's working to make sure that it's good,” he says.

Use cases like this might give open source models like Stability’s penetration into places that would be harder for OpenAI, but Jaffe adds that, ultimately, whether open or closed source, it will be the foundation models (like GPT-3) that deliver the best results and outputs that will win out.

“The real question about open source versus closed source in the foundation model universe will be how much better are the closed source models, because they have these massive war chests of investment going into them,” he says. “People are going to really care about quality here.”

So, while the pursuit of building the best foundation models in generative AI might be somewhat out of Europe’s hands, Jaffe still thinks European companies — even incumbents — can compete by building products that leverage the technology. 

It will now be up to Europe’s entrepreneurs working with generative AI to execute strategies that create value for the consumer in smart ways — something Jaffe says European’s have shown no shortage of talent in.

Tim Smith

Tim Smith is a senior reporter at Sifted. He covers deeptech and all things taboo, and produces Startup Europe — The Sifted Podcast . Follow him on Twitter and LinkedIn