Source: Market Information
Source: 01Founder
Yesterday, when I was swiping WeChat, a friend who does AI in the group threw a screenshot out: "It turns out that Quark is also on (deep thinking)." I casually browsed through the group messages, and I didn't pay much attention to this message, but I didn't pay much attention to it until I saw a sentence added by another friend below: "They didn't take DeepSeek, but used their own reasoning model." ”
This humble piece of information aroused my unquenchable curiosity.
In the past half month, the technology circle seems to have been swept by the DeepSeek model.
Almost every day, new products are announced to access this open source model, and the products of Tencent, Baidu, and Byte are scrambling to follow up, as if they will be ruthlessly abandoned by the times if they do not access. I remember that during the Chinese New Year, when Baidu Wenxin first connected to DeepSeek, a friend in the circle of friends couldn't help but sigh: "Baidu is too abstract, I actually took Deepseek!" Immediately afterwards, a number of products from Tencent ingots and bytes also followed suit, for fear of falling behind.
This wave of "DeepSeek dependence" has swept the entire Internet industry, forming a peculiar "herd effect".
It seems that if you don't pick up, you are falling behind
I can't help but wonder: in such a tide, does Quark's choice have a deep meaning?
Does relying on external models really mean giving up core competitiveness?
On the surface, the products connected to DeepSeek have indeed improved their inference capabilities in the short term, and the user experience has also been significantly improved.
But when you think about it calmly, the current situation of these products is somewhat embarrassing.
Where does differentiation come from when everyone uses the same model engine? What's worse is that these companies are unwittingly ceding technology dominance. Take a social app as an example, it does gain more intelligent conversational capabilities after connecting to DeepSeek, but when users say "this is an awesome answer", it is actually the ability of DeepSeek to praise the capabilities of DeepSeek, not the value of the application itself. When your core competencies come from someone else's model, you're effectively reduced to an application-layer tool, struggling to survive in an endless game of "feature copying".
Do we have to go through another model homogenization?
This brings me to a metaphor that may not be appropriate but is apt: freshly fried versus pre-made.
Restaurants that have given up their own kitchen research and development and gone straight to pre-made dishes. Ostensibly, they eliminate the cost of hiring senior chefs and developing recipes, and they can maintain a consistent level of production and serve food faster. But over time, all restaurants tend to taste similar, losing their unique flavor and soul.
What's even more troubling is that once pre-made food suppliers adjust or raise prices, these restaurants have no choice but to passively accept them. These restaurants that rely on prefabricated dishes seem to be "worry-free and labor-saving", but in fact, they give over the core competitiveness of the restaurant to others, and can only be reduced to "heating workers" in the end.
On the other hand, restaurants that insist on freshly stir-frying are more expensive and more complex, but can flexibly adjust to the freshness of the ingredients and the tastes of the guests, creating a unique flavor experience. In the fierce competition of the catering industry, this persistence often becomes a real moat.
Similarly, AI applications that rely entirely on external models will inevitably be subject to the iterative rhythm of the model, and the brand identity will be lost in the sea of homogenization. This seemingly "saving money and effort" shortcut is actually gambling on the convenience of the long-term vitality of the enterprise - this bet may be too heavy.
However, there are always people in the world who do not follow the crowd.
Those applications that have not yet been connected to DeepSeek, such as Zhipu, Kimi, and now Quark, are they really admitting defeat and waiting for death? I think it's the opposite, it's because they have enough technical confidence and strategic foresight that they dare to take a different path.
Looking at Alibaba's layout, Quark has become a precision test field for their AI strategy, verifying their self-developed capabilities through Alibaba's self-developed models, and quietly laying out a full-link AI tool ecosystem. Kimi chooses to deeply cultivate the fertile soil of long text processing and knowledge organization, and builds her own unique moat in vertical scenarios such as document speed reading and mind mapping. Zhipu is even more calm, focusing on polishing the automatic assistant experience of mobile phones and computers.
This reminds me of what a professor said in college: "The real innovators are often the ones who are not afraid of loneliness." ”
What these "alternatives" adhere to is their dedication and belief in technological sovereignty. They refuse to become "vassals of the model", believing that only through self-developed models can they truly grasp the right to define data, algorithms, and scenarios. They bet that the professional data and experience accumulated by Quark in the fields of education and health will eventually enable them to surpass the general model of open source in specific scenarios and needs.
In the final analysis, behind this seemingly simple "access or not", it is actually a game about survival and future.
I often wonder if I were a decision-maker in an AI company, what would I choose? In the short term, the access faction can indeed quickly reap market dividends, but at the cost of gradually losing the leading power of innovation; The self-developed school needs to bear high costs and uncertainties, and bet on the future of technology discourse. From the perspective of ecological competition, the DeepSeek ecosystem will prosper again, and its "model-as-a-service" model will sooner or later squeeze the profit margins of the application layer. In contrast, those players who insist on self-research are building a truly solid moat by building a closed loop of scene data.
Sometimes I think, looking back in a few years, who will have the last laugh in this AI war without gunpowder?
This will not be the last time such an exchange of intimidation
The prelude to industry differentiation seems to have begun.
Enterprises that rely on third-party models are gradually reduced to "functional shells", and profits are gradually eaten away by model providers. When the product functions of Baidu and Tencent converge to the point that it is difficult to distinguish, will users still care about your brand differences? And those "faith factions" who grit their teeth and insist on self-research may have truly impregnable technical barriers in the future, despite their huge initial investment.
When I talked to my friend about this, he made an interesting point: maybe there is a third way.
For those companies that are torn between the two options, they can try to integrate the common model in some scenarios, while insisting on self-development in their core business. For example, open source models are used to reduce costs in low-differentiation scenarios such as general Q&A and content recommendation, while self-developed models are insisted on in scenarios involving core data and professional fields (such as financial analysis and medical diagnosis) to build real technical barriers.
This compromise allows you to enjoy the short-term convenience of the open source model without completely giving up your technical autonomy.
In this era of accelerated AI development, the choice between "surrender" and "resistance" is essentially whether you want to be a participant in the ecology or a rule-maker.
There is no absolute right or wrong, and every company needs to find a balance between survival and ambition according to its own circumstances.
But one thing is certain: players who give up thinking will eventually be eliminated.
Those companies that dare to challenge the mainstream and insist on themselves may be able to open up a broader world in the future AI landscape.
The real winner in the age of AI is not whose model you use today, but whether you can be the rule-maker of tomorrow.
At least, Quark's choice gave me such a profound revelation.
Just as I was about to finish writing this article, my phone vibrated one after another: an app reminded me that I can now "click to generate a Deepseek summary", and a shopping software pop-up recommended "Deepseek recommends Mother's Day gifts".
I turned off the hustle and bustle one by one, and the scene reminded me of a scene from a classic sci-fi movie – "You have to decide for yourself whether to take the blue pill or the red pill."
In the choice of AI technology, why don't we face a similar choice?
Perhaps in this era when all apps are in a hurry to feed us blue pills, those "stupid companies" who insist on rubbing red pills by hand at least reserve the right and possibility to say "no" for us.