In the traditional internet era, the core logic of product growth was “reaching more people.” Companies assumed that product value was certain; as long as marketing and channels made more people aware, user acquisition and retention would follow. But Sirius points out that in the generative AI era, this approach is failing.
Increasingly, cases show that the essence of AI product growth is not user acquisition but the diffusion of belief. People download or use an AI not because they compare features, but because everyone is talking about it, showcasing it, or even afraid of missing out. AI products have no users, only believers.
From the traffic funnel to concentric circles of belief
Traditional SaaS or consumer products follow the AARRR funnel: acquisition, activation, retention, monetization, referral. But the value of AI has three characteristics:
Uncertain (each output is different)
Emergent (new capabilities constantly appear)
Needs to be understood to feel its value
Therefore, growth is no longer a matter of reach but a matter of cognitive transmission.
The diffusion of AI products is more like five concentric circles:
Appreciators (tech community, developers)
Evangelists (KOLs, media, creators)
Practical users (work or daily life users)
Followers (social experimenters)
Masses (atmosphere-driven)
The key rule is that the farther outward, the lower the belief density, but the more people there are.
Source: Sirius
Between each layer, translation is necessary; technological breakthroughs must be translated into industry narratives, transforming usage scenarios into social atmospheres. If any layer’s translation fails, diffusion stops. Many technically powerful AI products cannot break out of the circle because their stories cannot be simplified.
The true growth engine: mimetic desire
Most people think AI growth comes from network effects, but in reality, what drives explosive growth is a more primitive force: mimetic desire. Typical cases include:
ChatGPT and Ghibli-style images going viral
Suno AI music spreading on TikTok
DeepSeek’s popularity and the “if you don’t try, you fall behind” sentiment
Users don’t join because “more users make the product better,” but because they see others doing it and want to do the same. Therefore, the AI market won’t see Facebook-style monopolies (desire shifts). Growth strategies are not about optimizing features but about creating imitable behaviors. If outputs cannot be shared, showcased, or copied, diffusion won’t happen. This is also why some technically impressive but use-case-lacking products struggle to popularize.
Why traditional growth methods are collapsing
AI products are overturning five core assumptions of the past twenty years:
Marginal costs approaching zero: each inference costs real computational power; you can’t “subsidize growth first and monetize later.”
Feature stacking as a moat: AI competes on output quality, which is quickly leveled by model updates.
Network effects as the core barrier: most AI are single-user products; heavy usage doesn’t improve the experience for others.
CAC / LTV can be precisely predicted: model upgrades can cause users to churn overnight, making historical data unreliable.
Funnels are linear: in AI, sharing may happen before the first use; revenue may come before retention; activation depends on an “超预期体验” (超预期体验, an unexpectedly excellent experience). This means companies need to manage not just funnels but belief systems.
Deep structural similarities between AI and crypto
The growth dynamics of AI are highly similar to the crypto market:
Crypto AI
Meme-driven price Meme-driven user
Airdrops attract users Free trials create believers
Token release curve Free quotas and rate limits
Fork culture Open-source model competition
Consensus value, narrative value
Both share that users buy not just current features but future possibilities. But the difference is that crypto belief can sustain itself, while AI belief must be continuously validated with each output. Belief is rented, not owned.
The real battleground after growth: retention
Mimetic desire can generate traffic but doesn’t guarantee retention. AI retention is a race:
Short-term: supply-side innovation, continuously launching new capabilities to prolong curiosity.
Long-term: establishing new scarcity, accumulating personalized data, embedding into workflows, developing interaction habits, building trust assets.
If dependency is established within the mimetic window, the product can evolve from a popular tool to an infrastructure. Cursor, Bolt, and similar development tools are typical examples.
Five modes of AI explosion
Success cases from 2024–2025 can be summarized into five categories:
Meme explosion (Ghibli images, Suno)
Narrative shorting (DeepSeek: low-cost disruption of consensus)
Layered release (invite-only, waitlist)
Output as marketing (user creations become ads)
Workflow colonization (embedding into daily routines)
Among these, the last is the most difficult but also the most valuable in the long run. The only two questions that determine the success of an AI product are:
Does your product break a widely held misconception?
Is the translation between each layer—from core circles to the masses—smooth?
In a world without strong network effects, the AI market will naturally be fragmented. Traffic is no longer a moat; relationships and trust are.
