As AI companies expand across increasingly complex markets, Toku’s approach suggests that global platforms may need to combine shared infrastructure with local models, data, partnerships and regulatory expertise.
For much of the technology industry, international expansion has traditionally meant taking an established product into a new market, adapting the interface, adding local language support and building a sales presence. AI is making that model harder to sustain. Language models, data residency, cloud infrastructure, telecommunications rules and AI governance can all influence how an enterprise system is deployed.
Toku’s expansion provides a useful case study of this shift. The Singapore-headquartered, AI-powered customer experience company generated US$18.8 million in revenue in the first half of 2026, up 13% year over year, while its order book reached approximately US$29.3 million. Its geographic footprint has also widened, with deployments in Europe and the Middle East alongside its established APAC operations.
The more significant story, however, is not simply where Toku is expanding. It is how the company is adapting its technology and operating model to each market.
Expansion Is Becoming an Architecture Question
Toku’s recent Middle East expansion illustrates the changing economics of AI localization. The company expanded its regional coverage from two markets to eight between December 2025 and September 2026, incorporated a UAE subsidiary, and partnered with Turkish conversational AI company SESTEK to bring Arabic-language capabilities into its platform. The UAE entity allows Toku to contract, invoice and hire locally and participate in government and semi-government tenders across the UAE and wider GCC.
The technology layer is being localized alongside the corporate structure. SESTEK brings Arabic conversational AI capabilities, while Toku offers locally hosted deployment options intended to address data residency requirements.
The Talabat deployment demonstrates why this matters. Toku is now live across all eight Middle Eastern markets in which Talabat operates, supporting more than 8,000 users across in-house and BPO teams. Its contact center integrates with different CRM environments and manages varying regulatory, infrastructure and language requirements. After voice was added alongside chat, Toku reported customer satisfaction of approximately 80%, representing an uplift of more than 20%.
This suggests that localization is moving beyond the user interface. For enterprise AI, it increasingly becomes part of the system architecture.
Keeping the Platform Global While Making the Stack Local
Toku’s approach is also visible in its AI infrastructure strategy. In July 2026, the company launched Kawa, its first publicly available sovereign conversational AI infrastructure under the Makimoto initiative. Hosted in Singapore, Kawa keeps customer audio and transcripts in-country and uses a composable orchestration layer that allows developers to replace components such as audio preprocessing and speech-to-text models without redesigning the surrounding system.
That architecture is important because AI models are evolving rapidly. A system built around a replaceable orchestration layer can potentially incorporate different models for different languages, jurisdictions or enterprise requirements without rebuilding the entire platform.
It creates a distinction between what can remain global and what needs to become local.
The global layer can provide communications infrastructure, orchestration, security and enterprise integration. Local layers can provide language expertise, models, data processing and deployment environments suited to individual markets.
South Korea Is a Particularly Relevant Test
South Korea provides an interesting environment for this model because it combines a sophisticated enterprise technology market with significant domestic AI capabilities and an increasingly defined AI governance framework. The country’s AI Basic Act took effect on January 22, 2026. The framework establishes requirements around transparency and creates a governance structure for high-impact AI, including responsibilities related to risk management, user protection and human oversight.
At the same time, Korea is developing AI capabilities that are deeply tied to its own technological environment. NAVER Cloud describes HyperCLOVA X as a Korean-language model designed to understand Korean language and culture. In September 2026, a NAVER Cloud-led consortium also began developing two cybersecurity AI models of approximately 700 billion parameters each, deploying 4,000 GPUs and bringing together 33 organizations.
For international AI companies, this changes the proposition. The question is not simply whether a global platform can operate in Korean. It is whether it can integrate with Korea’s existing AI, data, infrastructure and regulatory ecosystem.
Toku already has a South Korean subsidiary, according to its corporate information filed with SGX. However, its FY2025 annual report described the Korean entity as dormant. That distinction is important. Toku has not announced a Korean market launch in the material reviewed here. Instead, Korea offers a useful illustration of what its partner-led expansion model could look like in a sophisticated AI market.
Global Orchestration, Korean Intelligence
Thomas Laboulle, Founder and CEO of Toku, said the company’s experience across Asia has shaped its approach to market expansion. While conversing with KoreaTechToday, he explained that Toku treats regulatory complexity as part of the product, incorporating market-specific requirements around data handling, connectivity and customer communications into deployment rather than retrofitting them later.
He cited Taiwan as an example, where Toku established a branch in 2025 after new subscriber-verification requirements affected how VoIP services could be resold.
For Korea, Laboulle sees a similar principle, but with local AI capabilities playing a central role. Rather than attempting to build every component itself, Toku could work with Korean speech and language specialists and connect their capabilities to its wider orchestration architecture.
“We would enter the way we entered the Gulf: with a partner. Korea has strong speech and language specialists building their own models, and they would run behind our orchestration layer with data and inference kept in-country, as SESTEK does for Arabic. The nearer opportunity is Korean enterprises themselves; many already run customer operations in Vietnam, Malaysia and Singapore, where we are live today. Timing will be set by the right partner and the right first customer, not by a launch date.
The lesson for any sophisticated market is the one Asia taught us. Compliance is the product. Partner with the local language specialist instead of claiming to be best at everything. And prove it in production, on the customer’s own calls and under their own regulator’s rules.”
The New Competitive Layer Is Local Intelligence
Toku’s expansion points toward a broader change in enterprise AI. The emerging model is not necessarily global AI versus local AI. It can be global infrastructure combined with local intelligence. That intelligence can come from domestic foundation models, specialist speech technology, local cloud infrastructure, market-specific data practices or partners that understand regulatory and enterprise requirements.
For South Korea, the implications are particularly relevant. The country has both a strong domestic AI ecosystem and enterprises operating extensively across Asia. A global CX platform that can connect Korean AI capabilities with multinational communications and customer-service infrastructure could therefore address a different problem from simply bringing another foreign AI product into the market.
From Localization to Local Intelligence
The next phase of AI globalization may not produce identical systems deployed everywhere. Instead, enterprises may increasingly demand platforms that can preserve a common technological foundation while adapting critical components to the markets in which they operate.
Toku’s experience across APAC, the Middle East and Europe illustrates why that model is emerging. Its platform remains global, but language, data, infrastructure, partnerships and regulatory requirements can be localized around it.
For South Korea, this could become an important model for how international AI platforms engage with an already sophisticated domestic ecosystem. The competitive question may ultimately be less about who can build the biggest global AI stack and more about who can connect global scale with the right local intelligence.
