Divergent Approaches to AI Commercialization (Part 1): Comparing the U.S. and China
Whether or not AI is just hype will depend on its commercialization. The U.S. and China are taking fundamentally different approaches to this problem.
This piece is another collaboration with the ever-brilliant Grace Shao, who has her ear to the ground on AI developments in China. She writes the insightful newsletter AI Proem, and I highly recommend her recent piece on Hangzhou as a rising hub for the next wave of Chinese innovation.
AI is a solution in search of a problem—a general-purpose technology that has yet to carve out its place in the economy. With hundreds of billions of dollars in investment spent on its development, there is growing pressure for AI firms to turn a profit.
Over the past several months, divergent approaches to commercialization have emerged in the U.S. and China. In the U.S., AI has been largely diffused as a way to empower enterprises, with tech companies embedding the technology into their cloud services, business software, and productivity platforms. In China, by contrast, AI is increasingly woven into consumer-facing super-apps, enhancing e-commerce, social media, and digital payments. This divergence reveals how technological diffusion is shaped not just by the capabilities of AI itself but by the distinct market structures, business incentives, and even national institutions in each country.
In this post, we’ll try to articulate these differing approaches to AI commercialization. This is no easy task, as companies are continuously experimenting with new models, making it a moving target. On top of that, systematically quantifying how AI is monetized is tricky. AI, after all, isn’t a standalone industry but is embedded into existing applications and workflows, serving primarily as an enhancement rather than a distinct product. As a result, it is difficult to estimate what portion of a company’s revenue comes from AI.
There are a few exceptions to this. Some firms are entirely dedicated to AI. OpenAI, which generated $3.4 billion in 2024, and Anthropic, which earned $200 million, can obviously be fully counted as AI revenue. However, these firms represent only a sliver of the overall diffusion of the technology. Microsoft—the only of the giants to explicitly break out AI revenue in their income statement—posted $13 billion on AI alone. So, while firms like OpenAI and Anthropic are at the frontier of model development, it's the tech giants that are capturing the lion’s share of the financial rewards from commercialization.
This pattern is no different in China. Model development companies like DeepSeek remain unprofitable—nor is becoming profitable their goal, at least not in the short term. Instead, firms like Alibaba, Tencent, and ByteDance, with their vast digital infrastructure, are best positioned to monetize AI—and they are already among the most aggressive in doing so. With this in mind, this post will compare how AI is diffusing by focusing on the efforts of tech giants in China and the U.S.
(In the part one of this series, we lay out our analyses as to how the leading players from each country are positioning their businesses in this new AI sector, and in part two we’ll explore potential explanations as to why such a divergence has occurred.)
Commercialization Taxonomy
So far, we’ve discussed how AI commercialization can be either enterprise- or consumer-facing. However, within each of these categories, firms can pursue two distinct strategies: directly monetizing an LLM or monetizing downstream applications that leverage an LLM—a framework we refer to as "monetizing LLM value-add."
Keeping track of all this can be confusing, with the constant wave of new products hitting the market. To help structure the different approaches, we’ve created a table that serves as a taxonomy of how firms are commercializing AI. This should serve as a helpful reference, as all the products discussed in the next two sections will fall within this structure.
The U.S.’s Proven Enterprise Cloud Advantage
In the U.S., the top backers of AI—Microsoft, Google, and Amazon—are also the cloud giants, and so it isn’t a surprise that these firms are monetizing AI through the cloud. The way they’re doing this is by including various large language models (LLMs) into their suite of cloud offerings and making it available to the millions of enterprise customers that already rely on their cloud services. Giving customers access to LLMs is just the tip of the iceberg. On top of that, cloud providers also allow customers to customize (e.g., fine-tune), deploy (e.g., inferencing), and manage the life-cycle of these models—all of this runs on top of existing cloud infrastructure.
Microsoft’s $13 billion in AI revenue is primarily driven by this approach—offering access to foundational models such as OpenAI’s, providing customers with the ability to fine-tune these models, and providing the infrastructure for managing inference workloads at scale. Interestingly, Microsoft’s AI growth is largely fueled by inference workloads, as enterprises increasingly rely on its cloud services to deploy and operationalize AI models in real-world applications.
What’s striking about this is that the focus isn’t on AI as a direct consumer product but rather on enabling other businesses to integrate AI into their own workflows and operations. Microsoft CEO Satya Nadella emphasized this point, stating that AI’s “real demand” is in “the enterprise space or our own products like GitHub Copilot or Microsoft 365 Copilot.”
Amazon and Google’s approach to AI commercialization is similarly built around cloud-based AI services. And like Microsoft, their strategy prioritizes enabling businesses rather than developing standalone consumer AI products, focusing on building AI into their own business processes. What’s noteworthy about this approach to commercialization is that these firms provide access to a marketplace of over a hundred pre-trained models for customers to select from. So, while Microsoft’s preferred cloud model is OpenAI’s ChatGPT and Amazon’s is Anthropic’s Claude, cloud providers aren’t locked into a single model. Notably, DeepSeek’s model is available on both Amazon’s and Microsoft’s cloud.
At the core of this strategy is the direct monetization of the AI models themselves, whether through licensing access, charging for customization (fine-tuning), or deployment (inferencing). Fundamentally, this approach is about selling AI to businesses, enabling them to integrate and customize models to enhance productivity, automate workflows, and drive efficiency across their operations.
(We’ve compared the U.S. and Chinese cloud computing sectors here.)
Among the giants, Meta is the exception. While it rivals the cloud giants in capex for AI infrastructure, it itself isn’t a public cloud provider and can’t rely on enterprise adoption for growth. Instead, its AI commercialization strategy is tied to AI-enhanced advertising—the company’s primary business line. Meta hasn’t reported AI-generated revenue separately. However, its recent boom in advertising revenue is attributed to AI-driven optimizations. This includes generating countless variations of the same advert, analyzing and testing them in small batches to determine which works best with the target audience, and then flooding the market with the highest-performing ones.
Unlike the other giants, Meta also has the best shot at popularizing a consumer-facing AI app with its control over WhatsApp, Facebook, and Instagram. Mark Zuckerberg has been one of the loudest in advocating for a consumer-facing AI agent, even stating that Meta will be behind a 1-billion user-serving application.
So far, consumer-facing AI apps are largely dominated by firms like OpenAI and Anthropic. However, as we've discussed, this represents only a small fraction of AI’s overall commercialization. The real financial gains remain firmly in the hands of big tech, where AI is primarily monetized through enterprise services that rely heavily on cloud computing and the vast compute infrastructure required for AI’s diffusion.
China's Hopeful Consumer Strategy
In contrast, China’s tech giants have focused on monetizing AI through applications rather than directly commercializing the model layer. This reflects a broader challenge in China’s economy: individual users and business customers are far more reluctant to pay just for access to AI models. Even in the consumer chatbot market, where OpenAI and Anthropic have had decent traction in attracting paid users, their Chinese counterparts, such as Moonshot’s Kimi or Baidu’s Ernie, have struggled to convert users into paying customers.
As a result, the prevailing strategy in China has been to leverage AI to deepen engagement within existing consumer platforms. This includes AI-powered recommendations, content generation for e-commerce, and interactive features designed to enhance user retention across existing apps. Numerous consumer-facing AI applications are also being rolled out for entertainment purposes, from Meituxiuxiu's AI-powered image editing to Talkie's AI-generated avatars to role-play character companions. Such efforts, however, have resulted in little financial payoff, with tech leaders in China even saying they aren’t sure what monetization will look like.
Fundamentally, there is more of an expectation for AI to be free in China, making monetization through subscriptions more challenging. Part of this could be because there was simply no outstanding LLM until DeepSeek (maybe except for Alibaba’s proprietary Qwen). Then, when DeepSeek was launched, it was offered for free, setting the expectation that LLM access should be free anyway.
DeepSeek’s breakthrough in model performance has also served as a wake-up call for China’s tech giants, forcing Tencent and Baidu to reassess their approach. Both companies have come to recognize that their heavy emphasis on application-driven engagement has come at the expense of advancing their own foundational models. In other words, they’ve rushed to commercialize subpar AI models, integrating them across China’s internet ecosystem before achieving true model competitiveness.
Ironically, DeepSeek’s success has also triggered a new race to commercialize AI. As Paul Trilio, Partner and SVP for China and Tech Policy Lead at DGA Group, notes: “Chinese AI firms have struggled to come up with ways to monetize AI model deployments, and the emergence of DeepSeek [...] means that the application side is critical: quickly leveraging an advanced model that is deployed across multiple platforms and drives down costs is now seen as the goal.”
Tencent—lacking both a leading model and a working monetization strategy—is capitalizing on this shift. It has integrated DeepSeek into its consumer-facing applications, notably embedding it directly into its super app WeChat, where it can optimize user experience across the super app’s over 1.5 billion monthly active users. Similarly, Baidu has also dropped its closed-sourced approach to integrate DeepSeek into its search app—a more lackluster product but backed by similar thinking.
ByteDance has thus far refrained from integrating DeepSeek’s model into its applications, opting instead to stick with its proprietary LLM. Before DeepSeek, ByteDance’s Doubao was China’s leading LLM chat app, but its success was driven more by “cool features” than by the strength of its foundational model. These features included love therapy chatbots, English tutoring, and even a bot that mimicked Elon Musk—all designed to generate revenue through embedded pop-up advertisements rather than through direct monetization of the model itself. (See more about Doubao here.)
In just two months of its launch, DeepSeek surged to 22.2 million daily active users, surpassing Doubao's 17 million users. This highlights that foundational model capabilities remain the key driver of competitiveness.
Among the big tech companies in China, Alibaba and, to a lesser extent, Huawei are the only ones to adopt an enterprise-first strategy similar to the U.S. cloud giants. Alibaba has a consumer-facing app, Kuake, which has tried to challenge Doubao’s dominance, but it remains a second priority for the company. Instead, Alibaba is focused on commercializing AI through cloud sales with internal goals of allegedly 10 percent year-over-year revenue gains for next year, hich it achieved and even exceeded during this past quarter.
Mimicking the U.S. cloud giants, Alibaba CEO Eddie Wu made the analogy during the most recent earnings call that the company will be the grid provider (cloud infrastructure) for the electricity demand (AI). It has offered its open-source Qwen model on top of or bundled with its cloud service, posting a 13 percent year-over-year revenue growth for its cloud business during its most recent earnings announcement. Like Google and Amazon, however, revenue directly from its cloud AI services is not reported separately and can only be inferred at best.
Huawei, also more focused on AI commercialization through enterprise cloud, has trailed behind Alibaba, finding success mainly among its existing SOE client base, largely due to its lauded safety features. (Grace has written about Alibaba and Huawei and their AI strategies)
What’s clear is that there is no consensus over how best to monetize AI, but everyone is choosing a lane that makes the most sense with their existing business to race in.
Implications
We’re still in the early days of the commercialization of generative AI, and a breakthrough “killer” application from one side could quickly push the other to follow suit. It's also possible that no such application ever emerges, and the excitement around AI remains just that—hype.
However, it is becoming increasingly clear that the top firms in the U.S. and China are thinking quite differently about their plans to monetize AI—one focused on enterprise adoption and white-collar productivity, the other on consumer engagement and digital consumption. These divergent approaches will shape not just who profits from AI but also how AI reshapes the economy itself—whether as a tool for workplace transformation or as a driver of consumer spending and engagement.
In the U.S., AI commercialization is largely enterprise-driven, with a strong focus on enterprise adoption. This approach takes the pressure off big tech firms to drive AI innovation alone, instead allowing the millions of businesses that rely on cloud infrastructure to integrate AI and explore new ways to use it on their own terms. This likely means more emphasis on enhancing worker productivity as AI tools become embedded in enterprise software.
By contrast, China’s AI strategy is heavily consumer-focused, prioritizing consumer applications that directly shape how users interact with digital platforms. This means that AI diffusion is largely driven by its impact on user engagement and retention, with companies racing to embed AI into super-apps, e-commerce platforms, and content ecosystems. As a result, the effects of AI in China are more likely to be concentrated on stimulating domestic consumption rather than transforming workplace productivity.
Regardless of whether AI is monetized through enterprise adoption or consumer engagement, from where we’re looking at it now, it is the tech incumbents who are best positioned to reap the financial rewards. Unlike the internet era—where new consumer apps could scale without owning the underlying fiber infrastructure—the AI era is far more capital-intensive. This gives a significant advantage to the firms that can afford to build or rent vast amounts of compute infrastructure and even secure dedicated energy sources to power these data centers. Whether through enterprise cloud services in the U.S. or AI-powered consumer platforms in China, the firms that own and control such infrastructure will likely dominate its commercialization, leaving smaller players dependent on their ecosystems.
Nevertheless, this divergence in commercialization strategies reflects a deeper divide in the tech ecosystems of the U.S. and China, shaping not just how AI is monetized but how innovation diffuses in each political economy. These differences are not accidental—they are rooted in structural factors that have long influenced how technology is adopted and scaled in both economies.
In Part 2, we’ll explore the deeper factors driving this divergence. We’ll examine why enterprise software (e.g., SaaS) never gained traction in China and how this created a path dependency that continues to shape AI adoption in enterprise products. We’ll also take a look at how the economic and labor structures of these two countries have influenced their distinct AI commercialization strategies.
 | A guest post by
|
interesting read; gets me thinking if a viable solution would be the convergence of these 2 approaches ?