Meta’s New AI Strategy as Industry Moves Toward Open Standards
181Meta is taking a different path in the AI world, especially after the recent launch of a new industry group called the Agentic AI Foundation (AAIF). This group, created by The Linux Foundation, aims to develop shared tools, standards, and foster community-driven innovation for AI agents used by businesses. It includes big names like AWS, OpenAI, Google, Microsoft, IBM, and Cisco, but Meta is notably absent.
According to a Bloomberg report, Meta isn’t part of AAIF because it’s working on its own proprietary AI model, called Avocado. This model is designed to generate revenue for the company. Industry experts say Meta’s approach reflects a broader strategy to keep control over its AI technology and data, rather than sharing openly with the community.
Why Meta Avoids Open Source Models
Brian Jackson, a research director at Info-Tech, explains that Meta has never been fully committed to open source models. Instead, they prefer open weights, which are just adjustable parts of a neural network, without sharing the training data itself. This allows Meta to protect its competitive edge because training data is a key asset that differentiates its AI from others.
Jackson notes that Meta wants to maintain control over how its models are used and integrated with other systems. As the Linux Foundation works toward clearer standards for truly open source models, Meta seems to realize it can’t control the distribution and use of its models in the way it originally planned.
Open Source Models and Revenue Challenges
At industry events, like AWS re:Invent, leaders discussed the challenges of funding open weights models. AWS CEO Matt Garman highlighted that open source software benefits from community contributions, but open weights models only have the provider contributing. This makes such models expensive to develop and maintain without a clear revenue stream.
Jackson points out that Meta’s move suggests they’re looking for ways to monetize their AI work. Instead of giving away models for free, companies often put APIs behind paywalls, charging per use or token. This approach helps cover the huge costs of training and maintaining large AI models.
Meta’s strategy appears to be shifting toward creating a proprietary model that can be monetized directly, rather than sharing their data and training techniques openly. This move may also be aimed at undermining competitors by commoditizing large language models and making it harder for others to replicate Meta’s AI advancements freely.
What do you think?
It is nice to know your opinion. Leave a comment.