A Conversation with Tuya Co-Founder Alex Yang

A Conversation with Tuya Co-Founder Alex Yang: Breaking Through Boundaries as AI Inclusivity Drives Business Transformation

The book also pointed out that in the Internet era, many enterprises pursue user scale while ignoring profitability, and ultimately were eliminated due to the cash flow disruptions, leading to the famous Internet bubble.

In contrast, Google at that time achieved profitability through its innovative search algorithm and clear profit model based on advertising, becoming a survivor.

All of these confirm the ultimate essence of Schumpeter's theory of "creative destruction"—innovation must complete the value closed loop of "technological potential energy → business model → positive financial cycle."

In recent years, the wave of artificial intelligence has not been an exception either: Artificial intelligence has experienced three rises and two falls, and once fell into the strange cycle of "technologically stunning but commercially weak". The AI winter around 2020 is still vivid in our minds: The vast majority of the world's leading AI enterprises are still in the red, and algorithm companies with blurred commercialization paths have perished in large numbers.

Fortunately, there are also signs in the AI industry that the ice and snow are gradually melting, and the spring of the AI industry has arrived.

According to an article titled "Overview of DeepSeek-V3/R1 Reasoning System" published by DeepSeek on Zhihu, DeepSeek has achieved an astonishing breakthrough in theoretical cost-profit ratio of 545%; Tuya Inc. ("Tuya"), a global leading AI cloud platform service provider, has also reported its first annual profit in 2024.

These breakthroughs together outline a feasible path for the industrialization of artificial intelligence

Tuya Financial Summary:Fourth Quarter 2024 Financial Highlights

Total revenue was US$82.1 million, up approximately 27.4% year-over-year, significantly higher than Bloomberg's consensus expectation of US$80.4 million.

Overall gross margin was 47.8%, up 0.5 percentage point year-over-year and 1.8 percentage point month on month.

Non-GAAP net profits were US$22.1 million. Non-GAAP net margin was 26.9%, improved by 7.4 percentage points year-over-year.

Full Year 2024 Financial Highlights

Total revenue was US$298.6 million, up approximately 29.8% year-over-year.

Overall gross margin increased to 47.4%, up 1.0 percentage point year-over-year, and has continued to steadily improve for 6 consecutive years.

Non-GAAP net profits were US$75.3 million, up approximately 268.5% year-over-year, with a Non-GAAP net margin of 25%.

GAAP net profit: With good Non-GAAP operating performance, it effectively covered the accounting burden caused by historical equity incentives and other items unrelated to business models and operations. Net profits were US$5.0 million, marking the first fiscal year of net profitability on a GAAP basis.

1. Returning to Business Essence from Technology Worship

DeepSeek's impressive profit margin breaks the narrative of technology worship and pulls the industry focus back from the arms race of technical parameters to the essence of business.

For a long time, large model technology has been given an almost mystical aura by the media, but from a pragmatic perspective, the essence of large models is to generalize complex AI capabilities into callable tools.

The industry landscape revealed in the "MaaS Framework and Application Research Report (2024)" by the China Academy of Information and Communications Technology is also of great reference significance - the high similarity between artificial intelligence architecture and cloud computing architecture reveals that technological evolution is repeating historical patterns: AI, a disruptive technology, will return to the essence of infrastructure.

This judgment has been recognized by Alex Yang, Co-Founder, COO, and CFO of Tuya. Yang said in an interview with Guru Club: "The future shape of the large model should be the general Internet infrastructure, just as cloud computing has reconstructed the IT infrastructure".

In this context, transforming AI capabilities into callable toolsets and applications is undoubtedly a key step in the commercialization of AI. Based on this concept, Tuya proposed the "LLM Agnostic" concept.

“LLM Agnostic" continues the previous "Cloud Agnostic" and "Edge Agnostic".

In the early days of the wild growth of the Internet of Things, Tuya used "Cloud Agnostic" and "Edge Agnostic" to help customers cope with the dilemma of infrastructure fragmentation, freeing developers and enterprise users from the complexity of underlying technology to achieve the "Agnostic" feature.

Nowadays, Tuya continues to follow this approach, using its technological architecture and generative AI solutions to enable developers and enterprise users to focus more on business logic and core business innovation without having to pay attention to underlying technical details.

That is AI for ALL, which aims to demystify AI. By continuously lowering the application threshold, it stimulates diverse creativity to be applied in practical scenarios and builds an AI equal - rights ecosystem where "everyone can participate and AI is everywhere".

Yang noted, "The essence of our advocacy for 'Agnostic' is to eliminate all barriers to entry, allowing more people to quickly enter this field and promote the rapid prosperity of new application scenarios

So, how does Tuya achieve this?

2. Being A Technology Integrator

Although it performs well in text reasoning, DeepSeek falls short in dealing with multimodal interactions in smart homes and spaces; Some Chinese models are constrained by high inference costs despite of strong content generation capabilities; Some models, though strong in Chinese content processing, features lower efficiency and higher costs, and may be more suitable for high value-added commercial content production; Some models may be more generalized, but perform average in terms of data analysis accuracy and rigor.

In addition, markets in different regions may have different requirements and preferences. European users may prefer local compliance models, Southeast Asian markets require multilingual support, and high-end commercial customers pursue private deployment.

In this situation, if a startup wants to develop AI devices and target the global market, it becomes very complex to choose the appropriate large model.

The LLM Agnostic concept of Tuya is to integrate the advantages of all models and provide developers with a universal platform to choose the most suitable model according to their needs.

In this way, developers only need to focus on their application scenarios, target groups, business models, as well as requirements for cost, accuracy, and usage preferences, in order to achieve better solutions.

Yang stated that through architecture level innovation, the platform enables developers to freely assemble models according to scene requirements, just like assembling Lego pieces. For example, it is possible to simultaneously integrate the semantic understanding of Tongyi Qwen with DeepSeek's code generation capabilities to create products that can be autonomously iterated.

But it is worth noting that in the highly fragmented business battlefield, being a technology integrator is far from enough. Tuya is well aware that breaking down barriers to entry requires a sharper ice axe.

Therefore, Tuya has chosen to build a global open and neutral AIoT ecosystem, providing developers with comprehensive empowerment in technology, supply chain, and ecological integration.

1. Cracking The Triple Barrier

Yang has repeatedly emphasized that Tuya must lower the entry threshold for downstream developers and users, and break down the barriers of large models.

In the consumer market, a simple "interface transfer" model may still be acceptable, but in the highly fragmented enterprise service field, customers face far more challenges than that. The technical challenges of data compliance and system adaptation are just the tip of the iceberg. More importantly, enterprises need to deeply integrate intelligent transformation into business logic. This multidimensional challenge creates a triple barrier of economy, technology, and market. And Tuya is breaking through these barriers one by one through the reconstruction of full chain capabilities.

At the technical level, Tuya has formed a full-chain end - to - end vertical system encompassing "data processing - vertical training - scenario optimization". Its dual ability of "understanding both code and workshop" not only masters cutting-edge technologies such as AI, but also understands the operating mechanisms of multiple fields such as energy, environment, entertainment and education. Moreover, relying on its in - depth engagement in the industries, it has accumulated a vast amount of high - quality industry experience and professional knowledge.

Thanks to these, Tuya enables developers to create applications and products in a short period rather than taking months, and to even directly achieve mass production of products.

When the industry was focused on the parameter race, Tuya had already evolved from a technical tool to an industrial base, truly dismantling the technological barriers that stand in front of enterprises by encapsulating complex algorithms into modules and products that can be embedded in business flows.

In terms of economic barriers, Yang stated that Tuya hopes to provide a very competitive price and has launched a "100 million yuan subsidy" plan to help AI developers accelerate innovation implementation.

In Yang's business blueprint, Tuya's "cost classification decoupling" strategy is quite forward-looking: one-time development costs are converted into fixed costs that can be digested through economies of scale, while focusing on reducing the key variable that determines the survival of the business model, which is the continuous use cost.

Just as users won't pay for toys with a monthly subscription fee of a thousand yuan, companies are more concerned about the long-term ROI of AI services," added Yang.

This shift in thinking from "cost shifting" to "value symbiosis" is restructuring the pricing system for AI services.

In addition to economic costs and technological barriers, Tuya also pays great attention to market barriers. The company's differentiation advantage lies in its ecosystem and global network. In 2024, Tuya served 5,800 brand customers, covering over 120,000 channel stores, and established its own cooperative ecosystem and brand influence in major countries and regions.

Yang particularly emphasizes the leverage value of ecology: "Developers can quickly achieve cross-border deployment with our channel matrix, and enterprise users can quickly go global. This may require years of accumulation in traditional models.

The ability to convert channel resources into market momentum significantly reduces the time window cost and trial and error risk for enterprises going global. We hope to leverage these resources to enable excellent ideas to quickly land, rather than being buried," said Yang.

4. Building A Systematic AI Empowerment System

When the triple barrier is systematically broken down, the AIoT revolution finally found a concrete carrier. The Tuya AI Agent development platform was officially launched in the fourth quarter of last year, marking the advancement of its technology from the conceptual level to the physical world.

At present, through the collaboration between T3/T5 SOC series modules and AI Agent development platforms, developers can flexibly choose to integrate different large models, including DeepSeek, Doubao, OpenAI, etc., and achieve unified scheduling of multimodal capabilities such as audio, video, image, and text.

The core advantages of the AI Agent development platform are mainly reflected in three aspects:

Firstly, the AI Agent platform adopts a low code development mode, which significantly reduces the complexity of algorithm adaptation, allowing developers to complete product development and mass production without having to deeply understand the underlying details of the model;

Secondly, the platform has strong cross-platform compatibility and supports deployment in multiple hardware environments ranging from embedded devices to cloud servers;

Finally, the platform has developed standardized solutions for eight major scenarios, covering areas such as energy management, environmental control, home living, health monitoring, entertainment and education, convenient travel, communication and collaboration, and audio and video analysis.

It is worth noting that for customers, this is not about launching a new business, but a seamless upgrade of existing business. Customers can switch and stack applicable AI functions at a low cost, and the delivery method is consistent with other services of Tuya, truly achieving "LLM Agnostic".

However, this is just the beginning.

Tuya’s vision is to systematically provide AI capabilities to all customers and seamlessly integrate AI capabilities into the three major business models of PaaS, SaaS, and Smart Solution.

In the PaaS (Platform as a Service) business, Tuya is actively promoting the productization of AI capabilities. At present, some categories of PaaS already support AI capability development, and developers can directly call these capabilities when developing products.

Some features have been opened to developers, while the rest are still under continuous development. Tuya is striving to fully open up AI capabilities for all categories on PaaS, promoting a comprehensive transformation from hardware to AI hardware.

In the field of SaaS (Software as a Service), Tuya is developing AI based SaaS services. Taking the pet track as an example, its bird and other animal species recognition algorithm and video automatic editing service help pet industry enterprises build software capabilities with "outputting emotional value and conveying companionship value".

These services allow developers to directly call AI capabilities on existing devices, thereby expanding application scenarios.

To meet the needs of different customers, Tuya also provides diversified delivery methods. For novice developers, there offers a zero code delivery solution, where users do not need to write code and can simply click the switch in the background to enable AI functionality.

For ordinary developers, API and SDK capabilities will be provided to support the development of applications based on AI capabilities. For hardcore developers, customized solutions are conducive to meeting their needs for high performance and flexibility.

It can be seen that Tuya is committed to building a comprehensive and layered AI empowerment system for users - from the full openness of PaaS to the innovative development of SaaS services, and then to the delivery of Smart Solution solutions.

This gradient empowerment strategy stimulates long-tail innovation by lowering the threshold for ecological participation, ultimately achieving the symbiosis and prosperity of platform value and ecological partners.

5. Conclusion

This transformation that starts with the reconstruction of the technological architecture will ultimately evolve into the reshaping of the business ecosystem.

When the threshold of development falls to a critical point, the network effect of the ecosystem will give rise to a qualitative change of "innovation density" - just as Android stimulates the mobile Internet revolution through an open architecture, participants represented by Tuya are creating similar chain reactions.

This symbiotic relationship between technological inclusiveness and business sustainability may herald that the intelligent hardware industry is about to enter a new stage where "marginal costs approach zero and the value of the ecosystem multiplies".