Rising AI Token Costs No Problem for Toku, Says CEO as SGX-Listed Firm Charts Expansion

The narrative around enterprise AI costs has taken a sharp turn in recent months. After years of declining per-token prices, leading model providers like OpenAI and Anthropic have raised their rates, triggering widespread concerns among businesses that had bet big on AI-driven efficiency. Uber reportedly exhausted its entire AI budget for 2026 within the first four months. But for Toku, the Singapore-headquartered cloud communications and AI-powered customer experience platform, the token cost squeeze is simply not an issue — and CEO Thomas Laboulle says the company's architectural choices are the reason why.

"Short answer is 'No,' and I think it plays to our strategy," Laboulle said during Toku's 1HFY2026 earnings briefing on July 28. "We think that because there is an expectation of commoditisation in the AI field, working only with third-party AI providers is not sustainable for the enterprise segment." Toku, which listed on the Singapore Exchange in January 2026 as the first SGX debut of the year, builds and hosts its own fine-tuned AI models on local infrastructure and hyperscaler environments. This approach insulates the company from the rising API costs that are hitting enterprises reliant on frontier models. "We do not have any exposure to token costs as we are directly hosting either locally or with hyperscalers, and therefore our cost is on the compute itself," Laboulle explained.

The Toku strategy highlights a growing divide in the AI industry. Companies chasing the largest, most capable models — GPT-5, Claude 4, Gemini Ultra — face escalating per-token costs as frontier labs seek to recoup massive training and infrastructure investments. OpenAI CEO Sam Altman acknowledged the tension at a June enterprise event, noting that "my company spent my entire 2026 budget in Q1" has become a common refrain among customers. Toku's counter-approach: use carefully fine-tuned, smaller models that deliver sufficient accuracy for specific customer experience workflows without the overhead of frontier systems. "The outcomes do not need the biggest model possible," Laboulle said. "It's something that is very much fine-tuned in order to deliver cost efficiency and accuracy."

The distinction matters beyond cost. Toku reported 1HFY2026 revenue of US8.8 million, up 13% year-on-year, driven by a 19.6% increase in usage revenue to US3.3 million. The company has also accelerated its Middle East expansion, growing from initial deployments in Bahrain and Qatar to eight markets including Egypt, Iraq, Jordan, Kuwait, Oman, and the UAE. Laboulle noted that the Iran-US conflict, which began in February 2026, has paradoxically strengthened Toku's value proposition as enterprises and governments reassess cloud concentration risk and demand multi-provider resilience — an area where Toku's architecture, deployable across commercial cloud, local data centres, and hybrid environments, offers clear differentiation.

Why it matters for Singapore: Toku's contrarian position on AI token costs — and its ability to decouple its expenses from the pricing decisions of frontier labs — offers a blueprint for Singapore-based AI companies navigating a market where the cost of external AI services is rising. As a newly listed SGX company with a proven go-to-market strategy in the Middle East, Toku demonstrates that sustainable AI deployment does not require access to the largest models. For Singapore's tech ecosystem, the lesson is clear: architectural independence from third-party model providers is not just a technical decision — it is a strategic moat that protects margins and enables predictable scaling in an increasingly volatile AI pricing environment.