Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days

ElevenLabs Buys Japan’s Kotoba for a Reported $100 Million-Plus to Rebuild Japanese Voice Agents From the Architecture Up

ElevenLabs Buys Japan’s Kotoba for a Reported $100 Million-Plus to Rebuild Japanese Voice Agents From the Architecture Up

ElevenLabs agreed to acquire Tokyo-based Kotoba Technologies. Nikkei Asia reported the price above 100 million dollars, while the companies did not disclose terms. Kotoba’s Koto model handles Japanese speech recognition, synthesis and speech-to-speech, and already runs inside Salesforce Agentforce Voice. All 15 employees transfer, closing is expected late in the fourth quarter.

ElevenLabs has agreed to acquire Kotoba Technologies, a Japanese speech-model developer, in what would be its first purchase in Japan. Both companies announced the deal on November 16 without disclosing a price. Nikkei Asia reported the consideration at more than 100 million dollars, roughly 15.9 billion yen. Closing is expected late in the fourth quarter of 2026, subject to regulatory approval.

Kotoba was founded in July 2023 by Noriyuki Kojima and Jungo Kasai, both computer science PhDs. Kojima, the chief executive, studied at Cornell, worked at Hugging Face and led language model projects on Japan's Fugaku supercomputer. Kasai, the chief technical officer, holds a doctorate from the University of Washington and was a research assistant professor at the Toyota Technological Institute at Chicago. Their Koto model family covers Japanese text-to-speech, speech recognition and speech-to-speech, with Korean and Chinese also in scope.

The commercial footprint matters more than the benchmark claims. Koto already powers Japanese-language conversations inside Salesforce's Agentforce Voice, which launched in August 2026, and supplies translation for SoftBank's daremo AI platform. It also runs the simultaneous interpretation behind a live performance by the artist YOSOBI and a consumer interpretation app of its own. Salesforce Ventures, an investor in Kotoba, is also an ElevenLabs investor, which made the two companies familiar to each other before talks began.

The team is small and the price per head is high. Kotoba grew from two people to 15 in three and a half years and raised more than 23 million dollars in total, including an 11.83 million dollar seed led by Globis Capital Partners in June 2025 and a 10 million dollar round led by Kindred Ventures with Salesforce Ventures and the Sony Innovation Fund. At a reported 100 million dollars, the deal works out to roughly 6.7 million dollars per employee, a rate that reflects how few researchers can build production-grade Japanese speech systems.

For ElevenLabs the sum is small. The company raised 500 million dollars in a Series D led by Sequoia at an 11 billion dollar valuation in February 2026, then closed a 300 million dollar employee tender offer at 22 billion dollars on September 30, led by Wellington and T. Rowe Price. Annual recurring revenue rose from 350 million dollars at the end of 2025 to more than 500 million dollars this year. The reported purchase price is under half a percent of the last valuation.

Japanese is not English with different characters, which is the strategic argument for buying rather than localizing. A single sentence mixes kanji, hiragana, katakana and Latin acronyms, which trips grapheme-to-phoneme models trained mostly on encyclopedia text. Mora timing and pitch accent drive perceived naturalness more than raw word error rates, and customer-service agents have to match honorific registers to industry norms. Chief executive Mati Staniszewski has argued that a user's language and social norms must be built into the architecture from the start rather than patched on afterward. ElevenLabs opened a Japanese subsidiary in April 2025, its first overseas base, and partnered with NTT Docomo Business on contact-center applications in June 2026.

The plan is to fold Koto into ElevenAgents, open a Tokyo research lab and hire local sales leadership, then extend the work to Korean and other East Asian languages. Integration is the open question. Kotoba's relationships with Salesforce and SoftBank predate the deal, and customers sometimes hesitate when a supplier moves under a larger vendor with competing products. Keeping 15 scarce researchers is the other variable, since every speech lab in the region recruits from the same small pool.

The wider read is that voice AI is fragmenting along language lines rather than consolidating around one global model. English-first stacks degrade quickly in markets where politeness, silence and context carry meaning, and the buyers are contact centers that cannot afford a stilted agent. Paying a premium for a regional team with live enterprise deployments is cheaper than rebuilding that capability, and it explains why the first acquisition for a 22 billion dollar company is a 15-person lab in Tokyo.

Comments (0)

Log in to join the discussion

Log In

No comments yet