2026: The "solar terms" of the AI industry have changed—how should entrepreneurs fine-tune their algorithms?
TechFlow Selected TechFlow Selected
2026: The "solar terms" of the AI industry have changed—how should entrepreneurs fine-tune their algorithms?
The recently passed 2025 will be the worst year for AI startups in the next five years.
Navigating Web3 tides with focused insightsAuthor: Zhang Peng
At the start of the new year, let’s talk about what lies ahead in 2026.
Recent developments—Meta's acquisition of Manus, and Zhipu and Minimax listing on the Hong Kong Stock Exchange—are undoubtedly strong signals to global AI entrepreneurs, validating with real capital the enormous opportunities of this era.
In my view, the long, anxious "struggling through spring" phase of AI entrepreneurship over the past three years is now over.
Why “struggle” through spring? As China’s traditional 24 solar terms suggest, different seasons call for planting different crops. If ChatGPT marked the Beginning of Spring, passing through 2024’s “Rain Water,” witnessing 2025’s “Awakening of Insects,” then 2026 may well be the year of “Spring Equinox.”
Looking back at the rhythm of founders and capital markets, this assessment is being validated. In 2023, only those building large models could secure funding in China’s AI startup scene. In 2024, people began experimenting with so-called “wrapper” applications; overall, capital remained cautious, and only a few founders achieved investor consensus. By 2025, applications that solve real problems started emerging, and domestic capital began to heat up.
Top-tier Chinese VCs, who were mostly conservative in 2024, have in 2025 begun deploying aggressively—one firm reportedly invested in dozens of startups within a single year. It feels like the accelerator pedal has been pressed down to 80%. And this “hard踩on the gas” is becoming widespread, with market consensus on deals forming much faster—such as leading investment firms now focusing intensely on AI hardware and acting swiftly.
By 2026, large model capabilities will continue accelerating. From a capital perspective, recent events like the Manus acquisition and the IPOs of Zhipu and Minimax are landmark triggers for decision-making chains. The willingness to “floor the accelerator” will rise significantly. Changes in technology and capital, coupled with shifts in market and user behavior, mean the ‘seasons’ are indeed changing.
So let’s ask a straightforward question: In 2026, which founders will find it easier to raise money and make big strides forward?
From “Interesting Exploration” to “Meaningful Delivery”
Two years ago, a cool demo might have excited users or even secured investment. But today, a “cool” product is no longer enough—it must be “truly useful,” even representing a crushing leap over previous generations, or redefining an entire product category.
This shift isn’t limited to AI software; it was clearly evident in this year’s CES hardware trends.
In past CES shows, any product labeled “AI” could generate massive buzz just by showcasing a few attractive features. But this year, the wind has clearly shifted—people no longer pay for pure “AI hype” or superficial “AI functions.” Products that treat AI merely as surface-level makeup—like adding basic chat functionality or simple AIGC tools—are unlikely to succeed.
The industry and users are becoming more rational: AI should not be superficial 'makeup'—it must be the 'skeleton' supporting the product.
What truly excites us now isn't whether a device qualifies as “AI hardware,” but whether it’s simply the best-in-class in its chosen scenario. This depends on embedding AI (even without relying on large models) as the engine or foundational capability within a well-defined use case, thereby delivering clearer, superior user experiences and value.
These deeper questions about value are becoming focal points for investors—and increasingly sophisticated users.
To achieve this shift from “cosmetic enhancement” to “structural support,” we must rethink how products integrate AI capabilities. In this current season, choosing 'specialized' over 'general-purpose' gives a clear advantage in achieving meaningful delivery.
General-purpose products offer users infinite possibilities, expecting them to explore and discover uses on their own. But without clear scenario assumptions, most users feel lost, and the product often fails at first contact.
In contrast, 'specialized' products allow you to focus sharply on solving specific problems for specific audiences. They align development and marketing efforts with actual user needs upfront, enabling concentrated resource allocation and achieving “meaningful delivery.” With rising AI capabilities, deep domain understanding, and effective orchestration of AI, you’re far more likely to deliver a stunning first impression.
This specialized approach helps build strong initial user stickiness and data loops, giving you a better chance than general platforms to gain early traction—laying the foundation for future horizontal expansion.
At this point, two questions become critical for founders: Does the problem you're solving actually exist? And does your solution offer a decisive, overwhelming advantage?
The Exciting “Extended Line”
Of course, narrower scenarios tend to be clearer—but they may also appear to have lower ceilings. If you’re aiming to be an independent developer, taking a solid first step may suffice. But if you want VC fuel to build a large-scale company with high upside potential, you must answer a second question: Where is your path of extension? What is your ultimate goal?
When I talk with investors—or evaluate startups myself—the founder’s ability to nail a small initial niche represents the “floor,” while the future potential and scalability extending from that niche represent the possible “ceiling.” At this stage, having a credible extended line becomes crucial.
This extended line isn’t built on storytelling—it’s embedded in your initial design. We can break it down into two key dimensions:
1. Don’t just look at ARR and sales volume—what data assets can grow inside your product “container”?
In the AI era, if your product’s core capability doesn’t get stronger with usage, how is it different from traditional software? While ARR—a legacy SaaS metric—is a good indicator of PMF, it doesn’t reflect the long-term value of AI-native products.
Many founders I speak with recognize that products in the AI age are essentially self-reinforcing “growth containers.” There’s little debate anymore over whether product and model should be integrated—a strong product company will inevitably become a model company too. But before that, one core mission is to first become a “data company,” building defensible moats using fresh, live data derived from real user needs. This self-enhancing feedback loop is vital for moving from thin to thick scenarios, low LTV (lifetime value) to high LTV, and sustained growth in commercial value.
2. Your “supply chain” shouldn’t be flat
Whether software or hardware, supply chain strategy deserves serious attention today. Simply calling a large model API or leveraging China’s manufacturing base to create demos works—for starters. But for a real product, your supply chain should be “simple but not simplistic.” From day one, think about how to meaningfully build your supply chain—by creating unique engineering assets (domain-specific labeled data and models, workflows, accumulated data), or, once scaled, customizing and strengthening generic supply chains (like DJI and Unitree did with motors). The longer your supply chain moat, the less pressure you face from copycats, and the greater your lead in time and space.
On a completely “flat” supply chain, you control very few links. Eventually, your differentiation might boil down to UI and ID (industrial design), plus a sliver of first-mover advantage. But when you personally extend that chain, you create strategic “high ground.” For example, using accumulated user data to enhance personalization, abstracting and reusing specific workflows to reduce inference costs, or tightly integrating compute, sensors, optics with local models… A product is the tip of an iceberg; the supply chain you build yourself is the massive submerged part that determines its true scale. The bigger the underwater portion, the higher the visible peak.
From today to tomorrow, from first step to final vision, the clearer your thinking and choices, the greater your local pressure—and the more freedom you gain in strategic direction.
In 2023, founders could say, “Just start and figure it out!” But in 2026, “How should we do it?” becomes a critical prerequisite.
GeekPark published a deep dive into DJI’s core technical DNA (“Deep Dive: How DJI Became a New Giant in Imaging”). Back then, Wang Tao defined the drone’s core value as a “flying camera”—a definition that became both starting point and guiding star, leading to the development of deep, non-flat capability stacks beyond flight control: motors, gimbals, imaging systems. It was this imaging-centered “extended line” that shaped today’s DJI—not just the undisputed leader in drones, but also a major player in the modern imaging industry, thanks to hit products like Pocket 3, whose unit sales already surpass consumer drones.
Great Ventures Are Often Like a “Three-Step Layup”
If entrepreneurship were basketball, the “basket” would be the ultimate user value you aim to create.
You can choose the dramatic “half-court three-pointer,” or the agile “three-step layup”—both are valid strategies toward the same basket.
The “half-court three-pointer” means targeting a grand vision from day one, betting everything on making the shot. Founders who take this route often come with elite pedigrees, raising $100M+ right out the gate, surrounded by a “basket full of balls.” If one shot misses, capital and resources allow them to try again. Don’t envy them too much—higher expectations bring greater pressure. Everyone struggles in their own way.
But for most “regular” founders, you might only have “one ball.” Without hundreds of millions in runway, your most rational strategy is probably the “three-step layup.”
A successful layup requires identifying a rock-solid need and entry point under current tech and resource constraints, building a “simple but not simplistic” supply chain, crafting products that seem easy to replicate but are actually deeply defensible—each with seemingly limited ceilings, yet part of a continuous journey of connecting dots into a line.
Many saw Manus’ acquisition as a “half-court three-pointer”—a complete misunderstanding. Viewed over time, Manus’ success was a textbook “three-step layup.”
Founder Xiao Hong didn’t start with vast capital to build a universal Agent. His first step was Monica, a lightweight browser extension that quickly validated PMF at low cost and gathered insights into real user needs in the AI era. When he realized the growth ceiling of browser extensions, he decisively took the second step—leveraging prior learnings to launch Manus, a broader platform for general-purpose AI agents. He invested heavily in engineering to stretch and deepen the “supply chain,” bridging countless gaps between models and complex real-world applications, becoming the first to deliver tangible agent value to users.
Then, thanks to this “first-mover advantage,” user behavior itself (e.g., most frequent task requests) helped refine the general architecture into more focused, effectively deliverable vertical tasks. This focus drove better delivery, which brought more users and revenue, culminating in acquisition at peak momentum—scoring the basket. That’s a classic “three-step layup.”
Could a founder who naturally commands a $100M+ valuation choose *not* to shoot from half court, but instead adopt the three-step layup—reducing funding pressure and exploration risk, while increasing innovation success through incremental progress and shortening the distance to the basket?
Absolutely! And such founders may well be the toughest competitors.
There’s no objectively perfect strategy for founders—only subjectively optimal ones. Your choice depends on your personal resources, technological context, and market environment. Learn from history, but history never repeats itself exactly. The Manus story may not replay in 2026, but exciting new stories certainly will emerge.
I believe that 2025, just passed, will turn out to be the worst year of the next five for AI startups. In different seasons, plant different seeds, write different stories.
Wishing everyone in the new year finds their own “basket” and starts moving toward it.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
