2024 AI Globalization Review: Linear's Zheng Can & Linkloud's Gao Ning

Summary

• In 2024, taking AI global shifted from a trial balloon into a major option for Chinese AI startups, with the revenue benchmark crossing $10M. Linear Capital invested in 17 companies during the year, 11 of them in AI; all but one of those 11 had been global from day one. Gao Ning relayed that Plaud AI’s monthly revenue has already surpassed $10M. Zheng Can’s fundamental view: “China’s engineer dividend should be transferred to a much larger market.”
• The moat around a $10M product usually lies not only in novel functionality, but also in launch speed, traffic operations, and the continued expansion of user value. Plaud AI first captured mindshare with call-recording hardware, then sold subscriptions to lawyers, dentists, construction workers, and insurance agents; Notta went from Japan to more than $10M in annual revenue; Higgsfield relies on founder Biao Ge’s end-to-end system, from product selection through traffic operations. Zheng said the same product can generate a difference in traffic measured in the “1,000x range” depending on whether its operator knows how to run it.
• As AI lowers the cost of entrepreneurial trial and error, fundraising is no longer a prerequisite for product validation, but investors still expect credible evidence at every stage. Many one- or two-person teams can now validate demand and generate revenue quickly, with Plaud AI and Higgsfield cited as examples of companies that grew without raising capital. Zheng believes the market “absolutely has no shortage of money,” but pre-product companies need an unusually good idea and prototype, while PMF-stage companies need impressive numbers.
• The next wave of revenue growth discussed in the episode is coming more from “new model capabilities × old demand” than from generalized tool-building. Image products are expanding into image-to-image, image-to-video, and vertical use cases such as fitness, restaurants, and SMB posters; the hosts noted that AI can simultaneously lower content-supply and SEO acquisition costs for religious products; voice AI can enter services with existing willingness to pay, such as negotiating bills and handling returns. “Industries where you already know people can make money” are the more reliable starting point.
• Japan is the clearest geographic opportunity discussed in the episode: the first step is exceptionally high, but once cleared, the path may become “an open road or paradise.” Local IT and AI supply is weak even as demand for digitization has accelerated over the past 2-3 years; entrants that demonstrate a revolutionary experience, localization, and a long-term commitment to serving the market may see acceptance rise sharply. Chinese teams also bring an engineer dividend and capital advantage; real-time translation is a concrete gap, while a To B strategy requires a Japanese office and a local country lead.
• AI product quality is shifting from “can the model do it?” to “how much control should a specific user retain, and how much work should the product take away?” Cursor preserves control for engineers, who can complete roughly 80%-90% of the work according to their intent; Bolt.new is better suited to non-programmers who simply want a complete result from natural language. The host said Devin “takes away all the joy and leaves me all the frustration,” yet found it comparable to a usable junior employee for research decomposition.
• The relatively certain development for 2025 is multimodality—especially voice—entering software and hardware commercialization; the real wildcard is the personal agent and the platform. Gao Ning hopes for a personal agent that can “see what I see,” act proactively, and provide emotional value, while acknowledging that current products remain far from human capability. Zheng instead asks whether a larger platform company will emerge after the proliferation of tools, and what role Chinese entrepreneurs can play.

Deep dive

1. Globalization Becomes a Major Option for Chinese AI Startups

• Gao Ning summed up 2024 with the word “unlock”: he hosted salons in the Bay Area and Tokyo, and watched companies that peers had taken global in 2023 finally establish a foothold in 2024—especially achieving zero-to-one traction in Japan.
• Zheng Can chose “discovery”: AI applications are like “an extremely open plain that grows larger every day,” and investors are searching for new species that have not yet appeared—the wind begins at the duckweed’s edge.
• Linear Capital invested in 17 companies during the year, 11 of them in AI; all but one of those 11 had targeted global markets from the outset. The portfolio is primarily made up of Chinese teams focused on education and companionship, productivity tools, and tech-enabled consumer products.

2. $10M Companies Win an Opening Wedge, Then Turn One-Off Features into Subscription Value

• Gao Ning cited Plaud AI, which entered through iPhone-compatible call-recording hardware that generates a summary after each call. He relayed word from his network that the company’s monthly revenue has surpassed $10M.
• The host’s objection was straightforward: recording and transcription are not technically complex, and the product faces cheap copycats from Huaqiangbei. Gao’s answer was that hardware is only the entry point; the real value lies in what the product can do for different professions after converting calls into text.
• Lawyers, dentists, construction workers, and insurance agents are already using it as a work tool and continuing to pay for subscriptions. Moving first also compounds long-tail advantages in keyword searches, video views, and user mindshare.
• Higgsfield’s dozens of products represent another path. Zheng believes founder Biao Ge has mastered the complete playbook from product selection to traffic operations; he said the same thing can generate a difference in traffic of as much as 1,000x in the hands of an operator who knows how to run it versus one who does not.

3. Notta Shows How a Non-English Market Can Build a Moat Before Going Global

• Linear-backed Notta started in Japan with monthly revenue of a few thousand dollars and later surpassed $10M in annual revenue. Zheng said more than 75% of its consumer users are in Japan, while roughly 25% of revenue now comes from outside Japan, perhaps 20% from the U.S.
• Japan’s market barriers leave many competitors “voluntarily shut out,” while local competition remains limited. In the early To C phase, teams could use Japanese-language talent in China to do SEO; after moving into To B, they need a Japanese office and a local country lead.

4. Lower Experimentation Costs Have Changed Fundraising Tempo Without Eliminating Stage Gates

• Zheng has formally engaged with and logged roughly 100-plus projects, but in 2024 he “didn’t need to fight for deals.” The reason was not simply a cooling in capital markets; technology was changing so quickly that founders and investors needed more time to bring their judgments into alignment.
• Because AI materially lowers the cost of trial and error and accelerates validation, investors expect teams to produce credible evidence first: pre-product companies need an unusually strong idea and prototype, while PMF-stage companies need impressive numbers.
• Gao Ning is seeing more one- or two-person teams and small groups run indie-hacker-style validation first, generate revenue, and only then decide whether to raise capital. Zheng’s view remains that the market “absolutely has no shortage of money”; teams that deliver strong results at the relevant stage do not need to worry excessively about fundraising.

5. The Image Boom Is Not Over; It Is Splintering Across Modalities and Verticals

• Gao Ning observed that text-to-image companies built on capabilities such as Stable Diffusion in 2023 had extended into image-to-image and image-to-video by 2024. OpenArt, PixVerse, and others continue to grow revenue and break through among specific user groups, countries, and regions.
• Effective expansion can target fitness users, photo-editing needs across countries and age groups, SMB posters, restaurant assets, and other niches; the resulting revenue gains have been “far beyond what anyone imagined.”
• Zheng also cautioned that image, video, and Character.AI-like companion chat are already large and crowded markets. Chinese teams’ edge in video comes from years of accumulated content and tooling, not from any inherent ease of entering the category.
• AI-native hardware is another branch. Gao Ning expects emotional-companion devices designed for specific user groups and countries; the host cited “呼呼,” a $25 device seen at CES that perches on the rim of a cup and blows on hot drinks, arguing that it mainly sells “joy” and a story.

6. Religious and Wellness Products Show How AI Can Reshape Both Supply and Acquisition

• Bless, developed by Chinese founders, serves Christians through scripture recitation and guided prayer; its impact comes not only from functionality but also from the spiritual experience. The XMind team has built a successful Buddhist product and later launched an AI tarot application.
• The tarot product’s unexpected moat was not a sophisticated model but more polished color and character design, attracting users willing to pay for aesthetics. “测测” uses AI for interactive psychological tests, MBTI, and interpretation.
• The host summarized the mechanism: religious products traditionally require costly, large-scale content production. AI lowers the cost of supply while also assisting SEO; one technology that solves both content and traffic creates a sustainable opportunity.

7. Free Web Traffic Remains a Tailwind, While Mobile Is Better for Copying Validated Demand

• StressWatch is not a typical AI product, yet publicly visible Sensor Tower data suggests monthly revenue of $600K-$700K. It is a reminder not to overlook Chinese developers’ ability to monetize globally in categories beyond AI.
• The host argued that after more than a decade of mobile internet development, traffic competition on PC has become less intense than on mobile. Web products can still acquire free traffic efficiently through SEO; mobile has no comparable entry point, but can transplant Web-validated use cases to similar users.

8. The Commercial Opportunity in Voice AI May Lie in Phone Calls Nobody Wants to Handle Personally

• Zheng called this roughly the first year in which AI can “speak like a human.” Overseas businesses have long outsourced phone calls to negotiate bills, process returns, or resolve various fees; AI can now perform these tasks where customers already have a clear willingness to pay.
• He cited Pine from the 声网 team, which performed well on Product Hunt. Compared with creating demand from scratch, businesses modeled on DoNotPay and BillShark operate in “industries where you already know people can make money.”

9. Coding Products Have No Single Winner; The Key Is Assigning the Right Amount of Control to Each User

• Zheng defines product quality as “making the product just right.” Cursor preserves the control engineers need, allowing roughly 80%-90% of the work to be completed according to their intent while reducing interaction to the minimum necessary.
• Bolt.new targets people who cannot code, delivering a complete result directly from natural language. Engineers may feel it is always “a little too far left, a little too far right,” but for users who do not want to participate in the process, that may be exactly right.
• The host used Cursor to automatically check Gmail invoices and save them to Google Drive. For clear, small execution tasks like this, he prefers an interactive tool over an agent that operates autonomously for long periods.
• Devin’s coding tasks on a shared team account “almost all failed,” but for research such as selecting a language market, it could break down a problem like McKinsey. Its value may lie less in the answer than in showing the path of reasoning.

10. NotebookLM’s Lead Comes from Product Details, Not Just TTS Capability

• NotebookLM’s Audio Overview can turn PDFs and other materials into podcasts several minutes long. Zheng tried multiple ways to replicate it and found very few that could match its level; the host said other replicas had not even reached 60% of its capability.
• The gap lies in conversational details: one speaker interrupts when the other becomes animated, the listener offers verbal cues, and speaking speed changes with emotion. A newer feature lets users ask questions mid-conversation, after which the two hosts respond naturally before answering.
• Tracing the team, the host found a bestselling author among its core members who participated throughout the project. Zheng added: “This is a product. The other replicas we’ve seen are, in my view, mostly replicas of the technology and code.”

11. Good Products Are Hiding the Prompt and Making Contextual Judgments for the User

• Gao Ning described YouMind as a product he uses almost every day: he first Snips the key points from articles, images, and videos into a creative asset library, then records thoughts on the excerpts or discusses anything he does not understand with AI.
• What impressed him most about Gamma was not AI Slides but the fact that beginners “don’t need to write a prompt.” It automatically breaks a rough description into entities, style, and environment; users can usually select a suitable generated image within two attempts.
• The host’s AI blog workflow automatically selects 3 of the most visually compelling illustration positions for each post, calls the Recraft API with the surrounding context, and generates tables and a table of contents at the same time. After hundreds of finished posts, “every one of them looks like a breath of fresh air.”

12. Japan’s High Barrier Creates a Supply Gap, Rising Acceptance, and Room for Exits

• Zheng described Japan as a supply-constrained market: local IT and AI capabilities are weak, while demand for digitization has been extremely strong over the past 2-3 years. “The first step is exceptionally high,” causing many companies to abandon the effort automatically, but beyond it may lie “an open road or paradise.”
• Gao added that once a product delivers a revolutionary experience and proves it will localize for Japanese users over the long term, market acceptance can “rise in a straight line.” Real-time translation remains a concrete gap because Japanese’s inverted sentence structure requires the system to reorganize earlier translated text based on context.
• On the exit side, options include acquisitions by major conglomerates and CVCs, as well as different listing channels corresponding to revenue ranging from $10M to tens of millions of dollars. Relatively weaker early-stage capital adds another advantage for teams entering Japan with Chinese funding.
• The two speakers’ shared expectation for 2025 is that voice and other modalities will commercialize across software and hardware. The larger unknown is whether personal agents can move from efficiency tools into emotional companionship, and whether platform-scale opportunities will emerge after the current tool boom.