---
title: "Big Tech Enters Niche Tracks with Rocket Launchers: How Can Vertical AI Survive?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-06-02T11:01:06+00:00"
canonical: "https://ffcap.cn/en/research/src-20260602-02html"
source: "https://uniqueresearch.substack.com/p/src-20260602-02html"
language: "en"
---

# Big Tech Enters Niche Tracks with Rocket Launchers: How Can Vertical AI Survive?

_Original · Unique Research · 2026-06-02_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, four numbered sections, repeated examples and full panel, including all named speaking turns and their continuation paragraphs. Revenue, market, user, financial, efficiency and business-model figures are source or speaker claims, not independently audited findings. Monetary figures without a currency in the original remain without an inferred currency. The source is dated June 2, 2026. Statements about product availability, market size and competitive dynamics retain the source's wording rather than establish verified industry benchmarks. Company, personal and work titles are transliterated where official English forms remain unverified. The "rocket launcher" and "cold weapon" rhetoric is figurative where the source uses it figuratively._

AI Industry Observation

Big Tech Enters Niche Tracks with Rocket Launchers: How Can Vertical AI Survive?

A commercial exam on how vertical AI entrepreneurs face big tech, defend their scenarios, and sell results

"

If one day big tech also does this, what do we do?

This is the question almost every vertical AI entrepreneur has been grilled on countless times over the past year—by investors, employees, and themselves.

Compute power, users, brand, capital—the weapons the giants hold in their hands look like they walked straight into a cold-weapon battlefield carrying rocket launchers. In this seemingly one-sided "dimensionality-reduction strike," can vertical entrepreneurs really only become cannon fodder waiting to be slaughtered?

At a recent industry roundtable, Qiu Huihui, founder of Feidi Tech; Chen Li, product director at Bazuayu; He Shan, co-founder of Show3D; and Joe, founder of Halo, engaged in a no-nonsense, face-to-face commercial exam with Huang Jingrui (Jerry), VP at Unique Capital.

We drop our guard, tear apart those warm and fuzzy industry buzzwords, and use the rawest commercial logic and bloody reality to see what cards vertical entrepreneurs actually hold in their hands.

01\. The Mud Moat of Getting Involved: Why Can't Big Tech Poach Away "The Artillery Fire Heard at the Front" with High Salaries?

In investors' imagination, the "tacit Know-How" of vertical entrepreneurs is a false proposition. As long as the giants throw down 3x salaries and poach your business experts, or even the IT director of your benchmark large client, won't this moat collapse instantly?

But the most common mistake big tech makes is equating individual knowledge with organizational system capability.

"Big tech can indeed poach our sales champion or chief architect." But in the ecosystem of Bazuayu or Feidi Tech, these experts can get things done because behind them is a set of organizational soil that can react quickly and agilely.

At Bazuayu, the overseas team once fell into a desperate situation when entering the US market. An ordinary frontline salesperson, combining Bazuayu's RPA process, ChatGPT and Feishu spreadsheets, hand-built an automated customer-acquisition process at the front line. The result: within two months, they won a globally influential brand deal.

This process was then quickly distilled, refined, and coded into Bazuayu's currently overseas-beta intelligent customer-acquisition product.

If this salesperson were in the screw system of big tech, facing layer-upon-layer reporting for project approval and various compliance constraints, this "hand-built customer-acquisition tool" idea would most likely not even get the chance to be stillborn. Big tech can buy out individual brains, but it cannot buy away that evolutionary environment that "allows the front line hearing artillery fire to make decisions autonomously."

Furthermore, the "trust gap" in the long-tail market cannot be filled by throwing money.

Take finance, a deep-water area with strong licensing, as an example. Big tech can buy out one or two top brokerage benchmark projects with money, but financial institutions inherently have a strong game-playing psychology toward monopoly giants. They will never be willing to bet their core assets and lifelines entirely on one or two big tech companies.

"Mid- and long-tail institutions want companionship, want to be scolded, want to do the dirty work and the hard work," Qiu Huihui said bluntly. Big tech's financial model and high organizational costs determine that they can only do high-margin, high-standardization, high-profile deals.

Once you fall into the long-tail quagmire requiring deep companionship and extremely trivial customization, big tech's calculator starts sounding alarms.

So, the reality that big tech temporarily "looks down on"—vertical entrepreneurs don't need to avoid it. What we compete for is completing the most primitive asset restructuring and data stake-out during the vacuum period of big tech's attention.

Surviving until you're acquired by big tech or become its most worry-free agent in the niche ecosystem is itself a highly valuable commercial success.

02\. The Shelf Life of Technical Accumulation: Don't Fight the Steam Engine—Become Its "High-Precision Transmission Shaft"

He Shan, co-founder of Show3D, holds a card in his hand: the company has nearly 20 years of accumulated underlying GPU rendering and modeling technology, allowing designers to model smoothly on ordinary-configured computers.

So when big tech's cloud-based multimodal 3D large models evolve rapidly, and users only need to say one sentence to perfectly generate a 3D model in the cloud—will your craft of meticulously carving on ordinary computers, like those "master craftsmen who polished carriage wheels extremely smooth" on the eve of the Industrial Revolution, be directly run over mercilessly by automobiles?

"In the engineering world, 'one sentence generates a deliverable 3D model' is currently a false proposition," He Shan gave the industry front-line answer.

Multimodal large models can generate a 3D concept car that looks extremely cool. But when you throw this car into a game engine or an industrial manufacturing line, you'll find it's a pile of unusable "scrap iron."

True engineering-level delivery requires perfect topology, fine material texture fitting, mold parameters that meet industrial standards, and parameterized hierarchies that can be fine-tuned at any time.

Once a model generated by a large model based on probability deviates, it's like an AI painting. When you want to modify "raise the character's arm by 5 centimeters," AI can only redraw one, but cannot directly change those 5 centimeters.

What Show3D solves is precisely "how to take this pile of seemingly beautiful mud generated by AI and, on the most ordinary consumer-grade hardware, finely converge it into an industrial-grade artwork that meets delivery standards."

Therefore, the shelf life of vertical technology lies not in you going head-to-head with big tech's "cloud steam engine," but in whether you can become the core "high-precision transmission shaft" in that giant steam engine.

When big tech's cloud large models drive the threshold and cost of 3D generation down to almost zero, it will instead give birth to an overwhelming flood of entry-level 3D content. All this flood of entry-level content, if it wants to land, needs to go through Show3D's local kernel for the last-mile "fine processing."

Big tech's cloud models are responsible for "flooding"; vertical local kernels are responsible for "converging." The more prosperous big tech's large models are, the deeper the dependence on this kind of special scalpel-type tool.

03\. Don't Compete on Tokens, Only Sell Results: How to Build Your "High-Taste Toll Station" in the Traffic Wave?

For To C AI applications, the smell of gunpowder for survival is even stronger.

Joe's Halo focuses on AI social dating, and Chen Li's product focuses on overseas AI customer acquisition. If big tech pixel-level copies you and directly builds these capabilities into WeChat or Feishu with hundreds of millions of users, even playing the "strategic loss" free card—with what do vertical entrepreneurs' Token bills compete with the giants?

"Traffic is large, but traffic is not the ultimate solution for a scenario," Joe voiced the unspoken rule of the social field.

WeChat is acquaintance social. Here, users inherently cannot carry private, purpose-specific (like precise dating) digital-avatar social scenarios. Users may use AI in WeChat just for novelty, but in Halo, they use that precisely tuned digital avatar to seek deterministic dating results.

Similarly, the customer-acquisition tools built into Feishu or DingTalk will always give you standardized off-the-shelf goods. But what Bazuayu faces is a survival tool that, under specific foreign languages and specific RPA process linkage, can genuinely help salespeople close deals.

Chinese users indeed don't pay for metaphysical "feelings" and "taste," but they will definitely pay for "the conversion efficiency brought by the gratification."

If big tech's free alternative makes users chat 10 sentences and then see through it, feel it's fake—no matter how low the Token cost, it's wasting time; and if vertical AI's Prompt tuning and interaction design can make users feel a strong Aha moment within 3 sentences and genuinely get results—the high Token bill is not waste, but precious "productivity cost."

Facing the giants' low-price and built-in offensive, the only solution for vertical entrepreneurs is: don't charge by Token, charge by "results."

Big tech provides broad, generic "water" tools for free; startups build "high-taste toll stations" above the red ocean—charging by "successfully matching a couple" or "successfully acquiring a precise overseas lead."

Don't sell compute power, only sell results.

04\. Big Tech's "Water Seller" Open Scheme: You Think He's the Opponent, But Actually He's Waiting for You to Consume Tokens

When all entrepreneurs are nervously defending against big tech's pixel-level copying, Qiu Huihui stood at the top of the ecosystem chain and tore open big tech's cards:

"Actually, big tech doesn't have that strong a willingness to grab and do exactly the same things as us."

Large models have developed to today. The giants who have poured hundreds of billions in funds—what are they most anxious about now?

It's that their Tokens can't be sold, that the software-hardware all-in-one machines and compute power they sold at high prices to top financial institutions can't be used at all.

"Everyone thinks AI can only do investment research quantification or write code. How many Tokens can these scenarios consume in a day? Very few," Qiu Huihui said. What can genuinely help big tech run up traffic and compute consumption is using AI to serve the thousand-person-thousand-face, extremely fragmented long-tail wealth segments.

China has 700 million wealth-management accounts and 60 million active stock investors. Financial institutions want to do service but don't know how—one end is complex financial products, the other end is fragmented customers, every penny needs service, and big tech simply can't keep up.

If startups can rush to the front line and help financial institutions run through the dirtiest and most tiring non-standard companionship work of "how to serve customers," big tech will instead be all smiles behind your back. Because you're continuously helping them consume Tokens and compute power.

Big tech's underlying business logic is "water seller." What they desire most is for ecosystem partners to quickly churn out all kinds of strangely shaped niche scenarios, and then run up the Token volume.

Big tech is heavy industry; vertical entrepreneurs are special scalpels.

Vertical entrepreneurs don't need to keep scaring themselves. Big tech doesn't have the mind to stare at your little plot of land all day and pixel-level copy it. They only hope you quickly take your small shovel and widen the river channel, leading their water over.

Seeing through this layer of "water seller" open scheme, the road for vertical AI entrepreneurs has actually only just begun.

More Conversation Details

Guests:

Halo — Founder — Joe

Show3D — Co-founder — He Shan

Bazuayu — Product Director — Chen Li

Feidi Tech — Founder — Qiu Huihui

Moderator: Unique Capital — VP — Huang Jingrui (Jerry)

Guest Remarks and Track Introduction

Huang Jingrui (Jerry): In the process of doing FA, I've encountered very many vertical AI entrepreneurs, covering healthcare, law, industry, finance and tax, human resources and various aspects. Investors and entrepreneurs often ask the same question: "If one day big tech also does this, what do we do?"

Indeed, big tech has compute power, has users, has brand, has money—it looks like carrying a rocket launcher into a cold-weapon battlefield. Today's guests have all been deep in vertical industries for many years. Before the Panel begins, let's first ask everyone to take two minutes to introduce their enterprises and what they're doing. Start with Joe.

Joe: Hello everyone, I'm Joe. Our product is Halo, mainly doing AI social. We entered the relatively vertical dating (matchmaking and social) field by building digital avatars for each user.

He Shan: Hello everyone, I'm He Shan from Show3D. We mainly do "AI generation + 3D modeling." We find that many people do AI generation now, but often the generated things cannot be directly delivered for engineering. So we integrate generation, editing, and local rendering into a complete chain, directly giving customers a complete usable product. Thank you.

Chen Li: Hello everyone, I'm Chen Li from Bazuayu. Our company mainly does enterprise digital operations and customer experience management (CEM). The core has three product lines:

Bazuayu Collector: performs rapid collection and acquisition of data across the entire network.

Yunting CEM: does enterprise customer experience management, helping enterprises glean user feedback across all network and all channels, optimizing product iteration and customer service.

Bazuayu RPA: automated process and office tools, solving and replacing mechanical, repetitive labor in enterprises, intuitively playing a role in cost reduction and efficiency increase.

Currently our collector main line has also made two or three AI Agent products, including the recently beta product, which is an overseas To B AI intelligent customer-acquisition product. If you're interested, you can search and follow.

Qiu Huihui: Hello everyone, I'm Qiu Huihui, CEO of Feidi Tech. I'm a traditional financial media person with over ten years of experience, and in recent years I've been growing together with Unique. Our company focuses on the financial field.

Everyone's understanding of fintech and AI is mostly concentrated on the "investment research side." And Feidi Tech is on the "investment advisory side" (that is, the wealth side)—one end connected to products, the other end connected to customers. We focus more on the customer service field, helping financial institutions achieve full-scenario customer insight, service, operation, and deep value mining.

Facing the Giants: The Most Underestimated Survival Advantage of Vertical Entrepreneurs

Huang Jingrui (Jerry): Thank you guests. The first question is here: big tech has data and a strong user ecosystem. When they enter your niche field, what do you think is the most easily underestimated survival advantage of vertical entrepreneurs? Start with Teacher Qiu.

Qiu Huihui: Let me briefly respond. First, let me give a real case that just happened at our top brokerage client a week ago: their internal intelligent recommendation system originally used a certain big tech's capability. But recently big tech announced the full shutdown of this product, meaning "I'm not doing this anymore, you find other suppliers yourselves." So the client came to ask if we could take it over.

From our perspective, vertical suppliers and big tech should not be placed in a purely competitive perspective. Under the big vision of AI, the future is a process of reorganizing the entire supply chain and ecosystem starting from user insight and demand. Coming back to the niche survival advantage the moderator mentioned—I think in vertical scenarios—taking finance as an example—the simplest and most direct is the license.

The meaning of the license lies in: first, you can't just enter the market casually; second, behind the license there is a very deep, very long supply chain. Future AI-native scenarios must all grow within the license, within the closed loop. If you don't have a mindset of getting involved and rooting yourself in these scenarios, letting your capabilities grow out of the scenarios, it's hard to build trust. Whether it's your data, user feedback, or capabilities co-built with B-end clients—everyone needs to complete it together with a "partner" mindset.

Why does big tech exit at this time, while we can enter and get things done? Because what we solve is the huge gap between big tech and vertical scenarios. That gap has too many details that need people to do, and we need to play this role in the middle.

Huang Jingrui (Jerry): Understood, the financial industry has licenses as a shield. What about other industries—does Mr. Chen have any views?

Chen Li: I quite agree with what Mr. Qiu said—in vertical industries, you indeed have to personally go down to break through difficulties. The survival advantage we've summarized is: long-term focus and deep cultivation at the front line, through practice distilling scenario-depth product adaptation capabilities.

Looking at the surface workflow or a certain function of a single product—not to mention big tech, any company with a technical foundation can copy it. But the core behind it is the tacit needs of users in the niche track, and your solution corresponding to this need. This depends not only on time, but also on practice.

For example, why do we make an overseas AI intelligent customer-acquisition product? Bazuayu is a globalized company, and our best-performing markets are the US and Japanese-speaking regions, with the Chinese market ranking third. When we first established a sales team in the US, we were unfamiliar with the place, and the first thing we thought of was indeed online customer acquisition. At that time, we purchased very high-end, not-low-cost local US customer-acquisition tools (like Apollo, etc.). But we found it could only win by volume—even if we pulled the daily email and LinkedIn greeting quotas to the max, the effective replies and leads we ultimately got were very few.

In this scenario, one of our frontline sales (salesperson) himself combined Bazuayu's RPA process, plus ChatGPT and Feishu spreadsheets, and manually combined a set of automated customer-acquisition processes. He continuously optimized prompts (Prompt) and RPA workflow strategies, and the result—within two months, helped the overseas team win the first globally influential brand client.

This thing made us very excited. To let the company's domestic and overseas teams all use this capability, we deeply analyzed massive historical data, abstracted the positive and negative feedback factors affecting every step of customer-acquisition conversion, and internalized experience and data into product capabilities. This is today's product. If I have to say what its core competitiveness is—it's absolutely not just the seemingly simple AI workflow, but the product core force polished through countless real deal-closing cases and logic.

As long as you focus on your own track and do the niche fields that big tech disdains to do, or can't do deeply or thoroughly, this is our entry point.

Huang Jingrui (Jerry): Mainly still choosing a field to go deep in. Just now offstage I was discussing with Mr. He Shan, and he said Show3D is something big tech can't copy. Let's hear his views.

He Shan: Now image and video generation is done very much, but the pain point is that after generation it cannot be directly delivered for engineering, because it lacks deep editing, rendering and sculpting. Many companies doing image generation still have to add 3D Max or Maya for 3D modeling, re-editing, sculpting and rendering. But Maya and 3D Max currently don't have a perfect AI generation system.

What we're doing is integrating the entire chain end to end: from front-end AI-generated images and videos, to deep editing, to local rendering and sculpting, providing an integrated process.

Big tech is reluctant to do this, because the long-tail client group is different. If big tech wants to do it well, the underlying technology all has to be redesigned. Besides differentiated competition, our core advantage lies in underlying technology: most big tech schedules data based on CPU architecture, while we directly built the lowest-level platform through GPU. Our modeling, sculpting and lowest-level rendering kernel are hard for big tech to subvert in a short time.

Huang Jingrui (Jerry): What views does Joe have on survival advantages?

Joe: I think on one hand, startups (startup companies) are inherently different from big tech in organizational form. A more important point lies in speed and a grasp of timing. Before big tech reacts, you first see the opportunity, and through evolutionary speed seize users' mindshare first.

After seizing mindshare, how do you serve users well and build stickiness? In this situation, when big tech later copies and then satisfies users, they'll be at a disadvantage in timing and speed.

Huang Jingrui (Jerry): Strike first to gain the advantage—in martial arts, speed is the only thing that can't be broken.

The AI Era Restructures the Ecosystem: What Is the Hardest-to-Copy Moat?

Huang Jingrui (Jerry): Before AI exploded, everyone's moat may have been technology, may have been data. After AI came out, large models have already restructured the entire ecosystem. What do you think is the most essential, hardest-to-copy moat? Joe, you start.

Joe: The hardest to copy is still the entire organization's judgment and choices—this is even a kind of taste (Taste). Taste is very hard for other teams to copy. This explains why 1,000 people make the same type of product, and in the end 1,000 completely different appearances appear.

He Shan: I think it's still the most underlying core technology. It's easy for big tech to do the application layer, but we have nearly 20 years of the most underlying 3D modeling and generation underlying technology accumulation. Different from most big tech scheduling through CPU, we directly through GPU plus local rendering kernel—this is hard to replicate in a short time.

Huang Jingrui (Jerry): But what if big tech is willing to throw money and chase you down?

He Shan: They indeed might make it, but they need time. Second, we have differentiated client-side advantages. For example, like 3D Max or Maya—the learning cost is very high, and it's hard to run on lightweight consumer-grade computers. And our product allows students and designers to perform real-time modeling operations on ordinary-configured (3,000 to 5,000 yuan) computers. This kind of user-tier downshift and hardware adaptation is hard for big tech to attend to.

Huang Jingrui (Jerry): This is still a question of timing and positioning. What about Mr. Chen?

Chen Li: Let me briefly summarize three points:

Productization of tacit industry Know-How: internalize vertical, long-tail scenario-based general capabilities into the product's native capabilities.

Feedback from scenario private-domain data: after the niche scenario ramps up, the accumulated data value is absolutely more valuable than industry public data—it can continuously feed back the product.

Entering users' workflows: whether making collectors, RPA or CEM, we're all trying to integrate product capabilities into users' business mainstream, making them a habit. In the end, what's competed for is no longer product functions, but user habits. Since we have first-mover advantage, we must seize this ecological niche.

Qiu Huihui: I highly agree with Mr. Chen Li's views, and by the way add a conclusion: in the AI era, whether general big tech or vertical companies, they all face the same moat problem. This moat has one end as your understanding of user demand, and the other end as your integration of upstream supply chains.

I particularly agree with what Mr. Chen said about "getting involved." Entering the scenario, you'll be full of reverence for the complexity of client needs. At the same time, you and clients in vertical scenarios must definitely be in a "product partner" and "data partner" relationship.

For example, you may have to become a role like "autonomous driving radar" in a certain vertical scenario, jointly collecting environmental data and accumulating real-time feedback data. At the same time, you have to productize the "non-standard companionship process" as much as possible. The moat truly tested in the AI era lies in whether you can not repeat the old path of traditional SaaS, but create a new model—based on common interests, transforming companionship, this very heavy, highly uncertain thing, into product power, capability and data.

Huang Jingrui (Jerry): That is, to stand from the client's perspective to make products.

Qiu Huihui: What do clients want? Clients want to make money! You have to be highly bound to their money-making KPIs. Can't avoid this question, can't say I'm only responsible for providing you functions or helping you reduce costs—no, clients want efficiency increase.

Commercialization Expansion: The Biggest Non-Technical Bottleneck

Huang Jingrui (Jerry): To summarize, it's still seizing the first opportunity, building stickiness, binding interests. Let's talk about commercialization. When doing vertical scenarios, the bottleneck is often not technology. Currently, what is the biggest "non-technical factor" everyone encounters when expanding? Teacher Qiu, you start.

Qiu Huihui: Simply and directly two words: time. See if you can afford to wait it out.

Especially in a licensed industry like finance, let me give everyone a data point: in the just-past quarter, a certain top brokerage had quarterly net profit of 10 billion yuan. This is still under the big background of industry regulatory requirements to reduce fees and commissions and give investors a break. This means it's an industry with protective barriers, and it will still be very comfortable for a long time.

Although AI has come, financial institution bosses will say "I want AI," but at the same time they'll also say "I want KPIs, and also cost reduction and efficiency increase"—wanting it all. In this process, with what business model (short-term, medium-term or long-term product relationship) can you support a startup through a 5-year or even 10-year cycle? Under the deterministic goal of firmly believing the world will definitely change in 10 years, can you go through such a long innovation cycle?

My judgment is that in the next 3 to 5 years, licensed industries will undergo great changes. People are hard to persuade to change, and organizations are too—but people will be changed by "others' success." Joining this long gamble, you have to consider which client you're willing to first co-build a benchmark case (MVP) with, to be the first to become the successful one, letting others follow.

Huang Jingrui (Jerry): You mentioned the internal logic transformation of the financial industry—can you talk more specifically?

Qiu Huihui: The financial industry is currently at the overlap of several cycles: the economy is slowing down; regulatory requirements are shifting from homogeneous competition. In the past, channel money was too easy to earn. Now mid- and long-tail financial institutions will definitely have to seek to shift from "earning channel money" to "earning service money," from "selling products" to "serving customers." This is the internal logic transformation cycle, and then comes the AI technology transformation cycle.

Previously money was easy to earn, and organizations found it hard to cut flesh and transform. But AI has come—it emphasizes "end-to-end," making the client-to-product end-to-end path flattened, with no middle platform, and many intermediate departments may disappear. This is a gamble. There must be someone among clients seeking change. Whoever changes first will eliminate the others. So we're also actively looking for the department and Leader in institutions that most want to change, to cooperate with them first.

Huang Jingrui (Jerry): Understood. What about Mr. Chen?

Chen Li: What Mr. Qiu talked about is the To B field. Then let me talk about the To C direction. With the current rise in compute costs, To C has brought a business-model dilemma: the traditional operation logic of spreading marginal costs by expanding user scale no longer works now.

Previously, investing hundreds of thousands or millions to make traditional software, operations vigorously spending money on customer acquisition—because marginal cost is almost zero, even if customers increase tenfold, costs won't rise much. But now making AI products, if twice as many users come in, Token consumption and compute costs will multiply. So making To C AI products, from the start you have to think clearly about the business model, quickly put out an MVP version to verify the balance of costs and benefits, and quickly complete PMF (product-market fit).

Huang Jingrui (Jerry): Understood. Joe is also doing pure To C—do you have this feeling?

Joe: Yes, Token consumption has indeed been increasing. This forces us to return to the essence of growth: in the past, traditional software could profit through decreasing marginal costs; but in the AI era, how the Tokens you deliver become the "delight" and "results" users are willing to pay for becomes extremely important. Users will only pay when they get the results they want. AI products will incur fixed costs as long as they use large models. If made into a pure free product, ROI (return on investment) will never balance.

Huang Jingrui (Jerry): What are the challenges in commercialization expansion for Mr. He Shan?

He Shan: We have challenges when engaging large B (enterprise) clients. We spent a long time building the most underlying core technology, and also have benchmark clients. But for large enterprises, when purchasing, they still need relatively stable, volumetric suppliers. For startups, when going to get large Deals (orders) with large enterprises, qualifications and trust still face big challenges.

Ultimate Deduction: If Big Tech Pixel-Level Copies, What Is the First Action?

Huang Jingrui (Jerry): Last question. Suppose big tech really launches a product highly similar to yours. What will your first action be? Adjust prices, or add features?

Joe: I think at this time, more is to pull your own long board longer. Make your best advantages, differentiation, including the delight of user experience and Aha moment (surprise moment) to the extreme, widening the gap with big tech's first-version product. Just like "one white covers a hundred ugliness"—make this "white" unsurpassable. This is what we focus on, not adding features or fighting price wars—when it comes to spending money, big tech has plenty of money to burn with you.

He Shan: I also agree—we won't adjust prices or blindly add features. We'll still go deep in the most niche track. For small B clients like students and designers, what they pursue is not high-and-mighty rendering, but being able to model smoothly in real time anytime, anywhere on ordinary consumer-grade computers. Guard this most core underlying technology, and large companies will find it hard to compete.

Chen Li: Everyone's views are mostly the same. Blindly layering on general capabilities is definitely the most undesirable. If we really encounter this problem, first, in marketing and product iteration, we must strengthen and highlight our own long board, amplifying capabilities that big tech's general models don't have. Second, reinforce the trust of the core user circle. The batch of core users accumulated in the early stage is very precious. Serve them well, and their word-of-mouth spread and driving role cannot be ignored.

Huang Jingrui (Jerry): Does Teacher Qiu have any new views?

Qiu Huihui: I second the views of the previous three guests. But I'd add a unique feeling of the financial industry: big tech actually doesn't have that strong a willingness to grab and do exactly the same things as us.

Large models have developed for three or four years now. In the financial industry, which has money, what does big tech most want to sell? Of course it's Tokens and software-hardware all-in-one machines. It encapsulates all Skill, MaaS capabilities for ecosystem partners and B-end clients to use—all to stimulate everyone to use up compute power. This is its underlying business logic.

Actually, many financial big-clients who really spent big money buying compute power can't use up the Token volume at all. Because everyone mistakenly thinks AI can only be used for quantitative models or writing code—these scenarios consume very few Tokens. What can really run up the volume is using AI to serve thousand-person-thousand-face clients. China has 700 million wealth-management accounts, 60 million active stock investors, with massive personalized wealth needs unmet. Financial institutions want to do it, but don't know how. One end is complex financial products, the other end is fragmented customers—even if the client only has a few thousand yuan, you have to serve them.

If this volume can run up, it must be built on startups helping institutions solve "how to serve clients" well. This is completely different from the positioning of general big tech. The actionable space for vertical is actually quite large. Don't keep worrying all day that big tech will come and do something to you—big tech doesn't have that mind. It just wants ecosystem partners to quickly get applications going, and then B-end gold masters to quickly buy its compute power and Tokens.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260602-02html
On-site reading page: https://ffcap.cn/en/research/src-20260602-02html