I’ve always been a bit puzzled by Meta’s LLaMA (Large Language Model Meta AI) strategy.

If you’re not familiar, LLaMA is a free series of high quality large language models published by Meta.  They’re not entirely free in the foam-mouthed-Richard-Stallman sense, but free enough in the sense that you can download them and run them for no charge.  You can even use them in commercial projects, as long as you don’t hit a 700-million-user benchmark (in other words, everyone except big tech can use them).

Creating these models is not cheap.  Meta used thousands of Nvidia GPUs to train these models – in fact, over 100,000 A100s to create the latest, LLaMA 4.

The benefit to Meta to spending millions on AI models that individuals and users can use for free is supposedly:

• Democratization of AI
• Compete indirectly with OpenAI, Google, etc.
• Goodwill for Meta
• Establishes Meta as an AI leader
• Real-world testing on alignment, performance, security weakness, etc.

The problem with these benefits is that they’re pretty marginal.

Democratization of AI: this is a social goal.  It’s like if Apple decided to spending tens of millions of dollars on a vaccination program.  Sure, that’s good for society, but there’s no commercial benefit for Apple.

Compete indirectly with OpenAI, Google, etc.: In other words, release free AI models so OpenAI’s competitors use them instead of giving money to OpenAI.  But the problem here is that Meta doesn’t sell AI services, so even if they torpedo OpenAI revenue, there’s little benefit to Meta.  It’s like if Meta started a chain of free hamburger stands.  McDonald’s bottom line would suffer, but…so what?

Goodwill for Meta: They could sure use some, but this is pretty abstract.

Establishes Meta as an AI leader: Valuable if they’re moving into the AI space, but are they?  Giving away AI models is not a business.

Real-world testing on alignment, performance, security weakness, etc.: This is a valid way to make the LLaMA models better, but again, they’re free.  Even if they are polished to very high quality, how does this benefit Meta?  Meta does use LLaMA internally and so they derive some benefit, but multiple other big tech companies have developed high quality models without needing to crowd source them.  Meta saves some cost on what is essentially QA here, but does this lone tangible benefit justify giving models away?  If Meta had a successful commercial AI project (e.g., competed directly with OpenAI and Anthropic) they would derive the same benefit.

Meta Agrees With Me

Perhaps the financial realities of spending tens of millions (if not more) dollars on giving away free AI models has finally caught up with Meta or at least turned on a lightbulb.

According to today’s news reports, Meta is considering ‘de-investing’ in LLaMA:

In another extraordinary move, Mr. Zuckerberg and his lieutenants discussed “de-investing” in Meta’s A.I. model, Llama, two people familiar with the discussions said. Llama is an “open source” model, with its underlying technology publicly shared for others to build on. Mr. Zuckerberg and Meta executives instead discussed embracing A.I. models from competitors like OpenAI and Anthropic, which have “closed” code bases. No final decisions have been made on the matter.

This is on the heels of Meta’s semi-desperation on AI recently.  Earlier this year, Mark Zuckerberg decided Meta was falling behind in AI and began poaching talent from other companies at a frenetic pace, dangling $10 million paydays in front of top AI talent.

Perhaps Meta has finally realized that all the money it spent on giving away AI would be better spent on building true AI advantages.

raindog308

Raindog308 is a longtime LowEndTalk community administrator, technical writer, and self-described techno polymath. With deep roots in the *nix world, he has a passion for systems both modern and vintage, ranging from Unix, Perl, Python, and Golang to shell scripting and mainframe-era operating systems like MVS. He’s equally comfortable with relational database systems, having spent years working with Oracle, PostgreSQL, and MySQL.

As an avid user of LowEndBox providers, Raindog runs an empire of LEBs, from tiny boxes for VPNs, to mid-sized instances for application hosting, and heavyweight servers for data storage and complex databases. He brings both technical rigor and real-world experience to every piece he writes.

Beyond the command line, Raindog is a lover of German Shepherds, high-quality knives, target shooting, theology, tabletop RPGs, and hiking in deep, quiet forests.

His goal with every article is to help users, from beginners to seasoned sysadmins, get more value, performance, and enjoyment out of their infrastructure.

You can find him daily in the forums at LowEndTalk under the handle @raindog308.