The Scoop
AI-based language translation startup DeepL has spent a good chunk of the millions it has raised on hiring, growing its headcount nearly seven-fold in three years as it tries to maintain its edge over deep-pocketed big tech companies.
The German startup, which recently raised $300 million on a $2 billion valuation, pioneered the AI technology that will help create a world where language barriers hardly matter. On Wednesday, it announced it’s adding 165 new markets for its translation product for businesses in Asia, Africa, Europe, and the Americas.
But Apple, Amazon, Google, Microsoft, and Meta are all investing in similar technology. They also have an edge in the consumer market. Phones, earbuds, smart speakers, and other products could be the translation interfaces of the future. As hardware and software improves, real-time translation on those devices could soon become a reality.
In an interview with Semafor, DeepL co-founder and CEO Jarek Kutylowski said the only answer is to continually get better, faster.
“It’s all about the pace of innovation,” he said. “We’ve been able to convince our customers of our superior quality.”
Kutylowski said the company devotes up to 100 employees to research, tasked with improving and expanding DeepL’s machine translation capabilities, out of around 1,000 total staffers. “It’s a pretty significant part of a company of our size, maturity and revenue,” he said. Its clients include Nikkei, Deutsche Bahn, and Zendesk.
Katharina Wilhelm, a partner at Index Ventures, which led the recent investment round in DeepL, said consumers love DeepL, but the real value is in enterprise. “You probably won’t be willing to pay $50 a month for your private translations,” she said. “But you might be as a journalist or as a business because accuracy and security are so important to you.”
Kutylowski declined to say whether big tech companies have approached DeepL about partnering or acquiring its technology. He also said DeepL is not planning on building its own hardware device to compete with those firms.
Index Ventures is betting on a home run, and not a sale to a major tech company. “Obviously if there’s an insanely attractive acquisition offer, you’ll have to look at it,” Wilhelm said. “We would not have invested if it were a near-term acquisition by a big platform. This will be a generational company.”
Kutylowski said the best translation models are still too large to run on-device and must access the cloud, meaning hardware devices that conduct real-time translation are still a ways off.
Rather than focus on the consumer market, DeepL has turned to enterprise, selling seven-figure accounts to large corporations like international law firms, or to those that need diplomatic translations.
For example, one customer, a Japanese carmaker that he declined to name, has its research and development in Japan and its commercial operations in the US. “Communication between those parts of the business are extremely important, and so much can go wrong if it doesn’t work out,” he said.
The company is also moving into the literature space, with its products available to publishers. “We have a shelf of books in our office that have been translated using DeepL,” he said. “There’s even a machine learning neural network book.”
In this article:
Know More
When DeepL was founded in 2017, the major language translation tools on the market used machine learning techniques. At the same time, neural networks, the cutting edge technology in AI, were beginning to sweep through every industry.
DeepL took advantage of that sea change by starting from scratch with a neural network-based approach, and then made it widely available on the market.
It instantly caught on in Europe, where language barriers are a bigger issue than they are in the US. That gave DeepL an immediate edge and the ability to disrupt the incumbents.
As generative AI blew up in 2022, investor interest in DeepL soared. In January 2023, it announced a $100 million investment that valued the company at over $1 billion, making it one of Europe’s only AI “unicorns.”
Then, last month, it announced another $300 million fundraising round that doubled its valuation.
Its headcount has skyrocketed and so has the compute power it uses to train models. It also uses more than 1,000 human translators, most of them contractors, to help train its models, the company said.
Those investments create a significant barrier to entry for new startups looking to compete with DeepL. It does not, however, deter the big tech companies.
Reed’s view
DeepL may have the best translation in the industry, but the best doesn’t always win in tech. Take music, for example. Consumers almost instantly ditched compact discs in favor of an inferior-sounding technology — the MP3 — simply because it was more convenient (and, in some cases, pirated for free).
It could be that DeepL retains a lead in enterprise, where companies are willing to spend on quality.
But the big tech companies won’t want to pay DeepL for using its technology unless it can acquire it for a bargain. They’d rather settle for “good enough.”
As tech companies incorporate AI translation into devices, they should make DeepL and other translators an option for consumers. Even if tech companies don’t pay to license DeepL technology, consumers should have the option of paying extra to use third-party translation tools as a default in operating systems like iOS. If they don’t, regulators could step in and force them to open it up in the name of innovation.
I thought it was telling that Kutylowski emphatically dismissed the idea of the company building its own hardware. He is probably right that it would not be a good business decision. But it’s a shame that companies won’t consider that as an option because of the competitive landscape.