Behind the billion-dollar revenue of MaaS giants: How does Tec-Do navigate the "ROI closed loop" of AI marketing?

In early summer of 2026, the narrative logic of the Chinese AI industry is undergoing a dramatic change. On May 13th, Alibaba (09988) released a set of data during the financial report conference call that can be described as a watershed moment for the industry: the annual recurring revenue (ARR) of AI models and application services, including the Bailian MaaS platform, has exceeded 8 billion yuan and is expected to surpass 30 billion yuan by the end of this year. Alibaba CEO Wu Yongming stated, "Alibaba AI has officially crossed the early nurturing stage and entered the commercial return period." Coincidentally, according to exclusive reports from 36Kr, ByteDance's Volcano Engine significantly raised its annual revenue target for MaaS business to 15 billion yuan in April this year. Thanks to the revenue boost exceeding 1 billion yuan per month by the video model Seedance 2.0, this target achieved an astonishing 10-fold increase compared to 2025. The main theme of AI investment in the past two years has been dominated by the "Capex (capital expenditure) narrative" driven by NVIDIA's computing power and large model infrastructure. The release of these two sets of data has torn off the enthusiastic veil that has covered the big models in the past two years, sending a clear signal to the entire venture capital community: While big models are still marathon iterating, the tech giants are starting to charge for their electricity. However, as MaaS (Model as a Service) rushes forward as the underlying water, electricity, and coal, a cruel commercial paradox is emerging in the B-side market: the tech giants are making a fortune, but many terminal enterprises that have purchased APIs are still asking, "How exactly can AI help me make money?" Commercialization drives implementation: customers don't buy intelligence, they buy results On the B-side where MaaS giants are making leaps and bounds, a brutal contrast is playing out on the AI application layer. With the spotlight of the capital market shifting away from the underlying models, it is ruthlessly pushing out the early wave of AI tool companies that only built "shells" on top of the giant APIs and is triggering a round of valuation inversions and closures. On the other hand, another group of enterprises that can deeply embed AI into core business processes and directly deliver business results are experiencing valuation "double clicks". The root cause of this polarization lies in the equalization of capabilities of the underlying models, which is destroying the superficial moats of the application layer at an unprecedented speed. If a SaaS provider is only selling "smarter content generation" or "faster design tools" to clients, it is still essentially a fragile tool rental service, which can lead to customer churn at any time due to the price or iteration of another underlying model. Moreover, in the wave of globalization and stricter compliance requirements in this specific track of overseas expansion, the real pain points for enterprises are far more complex than simply buying an API from a tech giant. In actual overseas business scenarios, due to geopolitical policies, cross-platform ecosystems, and natural barriers of localized culture, enterprises entering overseas markets face a "heterogeneous underlying ecosystem" fragmented and intertwined by overseas giants like OpenAI, Google, Meta, etc. For the vast majority of enterprises going global, the depreciation of underlying models has not spontaneously evolved into lower customer acquisition barriers overseas. Instead, it has multiplied the trial and error costs for enterprises in multiple model selection, adaptation, and cross-platform compliance. A cold iron law in the B market is that enterprise customers never pay a premium for "technical parameters" and "AI concepts" themselves; they only pay for clear ROI (return on investment) and profit growth on their balance sheets. Tencent Technology recently expressed a similar view in its "Token Economics" series, saying, "Tokens are still important. They are like electricity, bandwidth, cloud resources, the foundation of everything. But what enterprises are willing to pay more for is not the foundation itself, but someone who organizes these foundational capabilities and hands them a job that can be checked, audited, and held accountable when something goes wrong." AI has shifted from the battle of capabilities to "paying for completion," a new price anchor has emerged. The hundreds of billions in scale of MaaS services provided by tech giants fundamentally sell general computing power and basic model capabilities, similar to the "original power" of the new industrial age. However, for terminal enterprises, what they lack is the industrial machinery to convert this power into business momentum. In this context, the competitive barriers of the AI industry have undergone a fundamental shift: the next wave of industrial dividends no longer leans heavily towards providing basic water and electricity, but towards those who understand how to "package" computing power into specific industry scenarios, reconstruct proprietary workflows, and directly deliver final growth results to customers - the "application layer catchers." As a leader at the intersection of AI, marketing, and enterprise services, Tec-Do is a typical practitioner of this "result delivery logic". Tec-Do's "First Principles" Facing the grand MaaS ecosystem laid out by tech giants, Tec-Do did not choose to compete in futile internal resources in generic capabilities but found a very clear ecosystem niche for itself: to do the most critical "last mile" of landing the giant MaaS ecosystem. Observing Tec-Do's commercialization path in AI, its most crucial barrier is not developing some revolutionary generic basic model but building a deep-entrenched industrial application architecture. Because the "first principles" of marketing in this industry are always about low-cost, high certainty business growth. Take their flagship marketing Multi-Agent Navos, for example, it is a complete revamp of the full chain of overseas marketing. Behind it is a vast amount of data and sophisticated large models built on top of this data. Data is the foundation of AI marketing technology architecture. The company captures a wide range of business signals, including advertising execution data, creative attributes, and advertising conversion paths. Currently, the company has accumulated over 400 million advertising strategies and over 14 million SPUs managed for clients, building a vast and diverse marketing data foundation. On top of this data, their 25-year-old Tic engine constitutes the intelligent engine of the company's solutions. Designed specifically for cross-border marketing, Tic deeply integrates global market dynamics research, cross-border growth strategies, and creative intelligence. With human-computer interaction (HCI) and modular architecture at its core, Tic can flexibly adapt to multi-lingual, multi-modal, and multi-scenario business needs, precisely connecting "audience-media-creative-budget," thus enhancing cross-regional, cross-channel delivery efficiency, commercial conversion rates, and stable performance. In January 2026, the Tic Q&A reasoning model (Tec-Chi-Think-1.0) ranked first in the global comprehensive ability in the SuperCLUE-Mkt advertising marketing professional large model evaluation benchmark list, reaching industry-leading SOTA levels in core dimensions such as market insights and text creative production. Data provides empirical evidence, the model transforms signals into decisions, and the product coordinates and oversees end-to-end marketing campaign execution. As the campaign progresses, effectiveness data flows back to data weaving, constantly optimizing model decisions, and improving subsequent decisions. The traditional marketing enterprise service model, with its fragmented and heavily individual dependent work chain - analysts looking at data, creative teams creating material, and optimization teams manually managing campaigns in the backend. In this traditional context, early AI applications only played the role of a basic intern - human gives the order, AI generates an image or text. This single-point entry cannot change the high marginal cost of labor, nor can it solidify true technological barriers. Under the Navos architecture, the capabilities of large models are broken down and "embedded" into every core node of the growth process, shifting from "single-point AI tool" to "Agent-native workflow". AI is no longer a one-time plugin that requires human fine-tuning but has transformed into multiple collaborating, industry-specific professional Agents with specific know-how. From global market intelligence insights in the early stage to generating multi-modality video materials for fine-tuning in multiple countries, achieving cross-platform automated monitoring and real-time bidding strategy adjustments. This reconstruction means that what Tec-Do delivers to customers is no longer just a cold software account, but a "digital external brain" that can operate in a closed-loop. In this workflow, media platform real-time consumption and conversion data (ROAS) form a closed-loop feedback. In full compliance with global data regulations, it continuously feeds back and optimizes the algorithm model's delivery decisions. This "more accurate as you run" deterministic growth loop is the irreversible moat that Tec-Do is building for B-side enterprises in the second half of the AI application. The "Marginal Cost Curse" breached by AI If the reconstruction of business processes is Tec-Do's moat on the customer side, then the resonance from investors evaluating the business elasticity of these types of enterprises brought about by this Agent-native workflow is the fundamental subversion of traditional financial models. For a long time, the core pain point of overseas marketing has been the "linear growth curse of human resources." In the traditional model, the expansion of business scale relies heavily on a proportional increase in headcounts (optimizers, designers, data analysts). This "heavy asset, heavy manpower" delivery mechanism leads to high marginal delivery costs, preventing companies from achieving the scale effects characteristic of technology companies. However, the implementation of the Navos architecture has broken this industry ceiling. When Tec-Do transitions bulk material production, cross-platform monitoring, real-time bidding adjustments, and other discrete nodes which were heavily dependent on human experience to an automated, collaborative Multi-Agent system, the logic of their business expansion has made a critical leap: from "human-driven" to "computing-driven." It must be pointed out that AI is not without costs - token calls of underlying models and cloud computing power also constitute new variable expenses. However, the essence of the Tec-Do model lies in utilizing standardized proprietary workflows to transform expensive and discrete human resource fluctuations into predictable, optimizable, and highly marginal-effective computing assets. With continuous iteration of algorithms and optimization of model call efficiency, the per-unit computing cost of a business exhibits a decreasing trend, while the labor efficiency and productivity it drives increase exponentially. This story is not just about the implementation of AI but a profound structural reversal. Tec-Do, stripped of the technological hype, is using this brand new operational model breached by AI to deliver a certainty to the market. By January 2025, through Navos and the large Tic model, the company has served over 100,000 advertisers, managed over 400 million advertising strategies, and covered industries such as e-commerce, gaming, entertainment, local life, and other globally-facing industries.