Meta is quietly changing its AI strategy.
In the past, Meta has adhered to the "open source" route - making its self-developed AI models free for everyone to use and improve. But now, under the leadership of CEO Zuckerberg, the company has launched a new AI project codenamed "Avocado", and the goal has changed: no longer just to be a "technical contributor", but to create closed-source AI products that can make money directly.
It is worth noting that Meta used China's Alibaba's Qwen to help optimize performance when developing this new model. This shows that Meta is actively integrating global resources and accelerating its catch-up.
The reason behind it is realistic: in the face of the rapid commercialization of opponents such as Google and OpenAI in the field of AI, Meta has invested tens of billions of dollars, but has not yet seen a significant return. Now, it hopes to earn back its investment as soon as possible through closed-source, fee-for-service AI services.
Meta is no longer "giving away" AI for free, but is about to start "monetizing".
From "free and open source" to "closed source to make money"
According to reports, Zuckerberg has formed a new team called TBD Lab to develop a new generation of AI models codenamed "Avocado". When training this model, the team not only used its own technology, but also referred to several external open source models, including:
Google's Gemma OpenAI's GPT-oss (Open Source Version) Alibaba's Qwen
This shows that Meta is widely learning from global excellent achievements and making every effort to accelerate research and development.
"Avocado" is expected to debut next spring and is likely to be released in closed source form - that is, no longer free public code, but strict control over usage rights. According to people familiar with the matter, Meta is even considering licensing fees for some AI models.
If it does, it would be a major turning point in Meta's AI strategy. Over the past few years, Meta has been adhering to "open source" in a high-profile manner, making large models free of charge to developers and researchers. But now, it is moving closer to rivals such as Google and OpenAI - first make good products, and then find ways to make money.
The pressure behind it is also realistic: Meta has invested heavily in AI but has not yet brought in significant revenue. Wall Street is beginning to worry about its profitability. Zuckerberg must prove as soon as possible that the "superintelligence" route he bet on will work.
And the key to all this depends on whether "Avocado" can really make a comeback.
Zuckerberg personally grasped the project and changed the general
Meta is undergoing a profound AI strategy turnaround, followed by large-scale organizational adjustments and personnel changes.
The turning point came earlier this year: Meta released the open source model Llama 4, but the results fell short of expectations, and even Zuckerberg himself was disappointed. This setback prompted the company to make a complete shift - from insisting on "open source sharing" to focusing on "closed source monetization".
In order to promote the new direction, Zuckerberg personally ended up: he reorganized internal resources and marginalized some core members of the original Llama project; at the same time, he personally led the establishment of a new team called TBD Lab to report directly to him; this team was even arranged to work centrally near his headquarters office, so that he could follow up on progress at any time
More critical moves: Meta recently acquired AI data company Scale AI for about $14.3 billion and appointed its 28-year-old founder, Alexandr Wang, as the company's chief AI officer. Wang is a staunch supporter of the closed-source route and will directly steer the research and development of Meta's future core AI products.
On the other hand, Yann LeCun, once known as one of the "godfathers of AI", chose to leave Meta. According to reports, he was partly dissatisfied with the company's neglect of the open source strategy and the tilt of resources towards closed source.
The company's executives also believe that his philosophy is no longer in line with Meta's current AI direction, and even before his departure, his voice was gradually restricted.
| Meta is betting on AI, but the market is not completely buying it
AI is now Meta's number one strategic priority. The company is pumping a lot of resources away from old directions such as the metaverse and virtual reality and betting on AI - especially AI glasses and related hardware.
Zuckerberg even boasted that $60 billion will be invested in the United States in the next three years, mainly to build the computing infrastructure needed for AI.
But Wall Street is cautious about this. Although Meta said that AI has helped improve the effectiveness of the advertising business, and Zuckerberg also emphasized that "we must build computing power quickly, otherwise we will fall behind", investors are still worried: Can such a large investment really be exchanged for enough profits before 2026?
The pressure is also reflected in the product.
In order to get ahead of OpenAI's release of Sora 2, Meta hurriedly launched a video generation tool called "Vibes" - the technology is actually licensed from Midjourney. Although the internal user usage is "okay", the product was completely overshadowed by the popularity of Sora 2 just a week after its release, and almost no one paid attention.
For now, Meta is pinning all its hopes on the core AI project "Avocado".
The company is trying to prove in highly regulated Europe and the giant-ridden United States that its "superintelligence" route is not only viable, but also winning.
Jingtai View|Pay attention to the success or failure of Meta's transformation, lay out AI infrastructure and domestic substitution
Once, Zuckerberg said: "AI should belong to all mankind." Today, he said, "AI must create value for shareholders." "This is not a betrayal of ideals, but in the brutal AI war, only by surviving is you qualified to talk about ideals.
Meta's turn marks that the global AI race has officially entered a new stage of "closed-source-led, business-oriented". For investors, stop asking "who is the most open", but ask: whose technology can be implemented? Whose model can collect money? Whose computing power can withstand the next storm? The wind has changed.
For Meta Stock (META):
Short-term pressure: high capital expenditure + profit uncertainty, difficult to increase valuation significantly; Key observation points: whether "Avocado" has differentiated capabilities (such as multimodality, inference efficiency, cost advantage) in 2026; if it is successfully commercialized, it may start a new round of valuation upward revision; If it fails, it may fall into the "AI investment trap".
For the AI industry chain:
Good for AI infrastructure: servers, optical modules, liquid cooling, AI chips (especially solutions that support closed-source deployment); Pay attention to the potential of domestic large models to go overseas: Qwen was adopted by Meta to verify its technical strength, and Alibaba, Baidu, MiniMax, etc. may be revalued; Beware of pure concept AI stocks: the "PPT model" with no real customers and no landing scenarios will be cleared at an accelerated pace.
