How Mark Zuckerberg made Llama a cornerstone of AI ambitions at Meta, whose smartphone-era services and products have been constrained by Apple and Google
Zuckerberg has pushed Meta to build Llama into all its products and services, including its Orion augmented reality glasses now in prototype stage.
FortuneSharon Goldman
Context & Ripple Effects
Meta's Llama push followed its effort to use open sourcing to build developer standing, reflected in Llama 2's early developer adoption. The strategy gives Meta an AI platform it can carry across its existing services rather than rely solely on mobile gatekeepers.
Later coverage reported 1 billion Llama downloads, while Zuckerberg discussed a Llama API and Meta's broader AI opportunity at LlamaCon. That arc makes the product-wide integration important: distribution inside Meta's products and external model adoption can reinforce one another.
First-order effects
Meta makes Llama a common AI layer for its services and brings that strategy into Orion, tying the prototype hardware effort more closely to its in-house model roadmap.
Zuckerberg's AI agenda becomes less about a stand-alone research release and more about deploying Llama wherever Meta controls the user experience.
Second-order effects
Embedding Llama across Meta surfaces increases the value of its distribution relative to Apple- and Google-controlled mobile access, while putting more pressure on Meta to make its AI experiences distinctive enough to retain attention.
Orion becomes a more consequential test of whether Meta can pair proprietary hardware with its own AI stack, rather than treating AR hardware as a separate bet.
Third-order effects
If Meta can turn model releases, developer uptake, and product deployment into a feedback loop, AI distribution may become a way for platform companies to reduce dependence on the mobile operating-system layer.
The move points toward a split between firms that mainly distribute third-party models and those that combine models, consumer surfaces, and emerging hardware; the durability of that split depends on product adoption and model competitiveness.
The trend: AI platforms are increasingly treating foundation models as a distribution and product-control layer spanning apps, developer ecosystems, and new hardware.
🚨 New @fortunemag story alert! 🚨 I've spent 6 wks digging into how @zuck is betting Meta's future on open-source AI with the Llama platform. From chatbots to AR glasses, Llama is redefining AI at Meta—and taking on giants like OpenAI with an open-source twist (yes, there's deba…
@yannlecun
Yann LeCun on threads
Fortune has an excellent article on Meta's AI strategy and its focus on open source. — I'm quoted multiple times. — Free version: https://finance.yahoo.com/... Paywalled Fortune version: https://fortune.com/...
@michae.lv
Michael Veale on bluesky
Useful reminder in this long read on Meta and LLaMA that its ‘open source’ play is about creating a platform, reflecting its insecurity around gatekeepers Chrome, Safari, Play & App Store. Android, after all, is ostensibly open.
@jeremyakahn
Jeremy Kahn on x
Great story on how @Meta and @finkd decided to go all in on open-source AI from my @FortuneMagazine colleague @sharongoldman (yes, ok, open weights, as Sharon says.)
@sharongoldman
Sharon Goldman on x
In January, Zuckerberg moved FAIR, the AI research group, into the same part of the company as the team deploying generative AI products across Meta's apps. “I think it makes a lot of sense to put [FAIR] close to the family of app products,” says @jpineau1
@abhishekn
Abhishek Nagaraj on x
Worth a read including some juicy new details on the behind the scenes decision to open source LLaMa ...
@sharongoldman
Sharon Goldman on x
“We got incoming requests from people who said, 'You have to open-source that stuff. It's so valuable that you could create an entire industry, like a new internet.'” - @ylecun, describing reactions to the 2023 leak of Meta's Llama https://fortune.com/...
@sharongoldman
Sharon Goldman on x
“I had a big remit,” @Ahmad_Al_Dahle says: “Develop state-of-theart models; put them in product at record speed.”
@sharongoldman
Sharon Goldman on x
Open-source coders accept that Zuckerberg “has some corporate realities that will distort his messaging,” @natolambert says. “At the end of the day, the community needs Llama models.”
@sharongoldman
Sharon Goldman on x
I really, really, REALLY had hoped to speak to Zuckerberg directly but I think he had a jam-packed rapping schedule 🎙️🎵
@sharongoldman
Sharon Goldman on x
“It felt a bit like Swiss cheese,” @nickclegg , Meta's president of global affairs, says of the failed attempt to keep Llama behind closed doors.
@sharongoldman
Sharon Goldman on x
Why open-sourcing Llama is good for Meta is “the big puzzle,” says Abhishek Nagaraj, associate professor at the University of California at Berkeley's Haas School of Business, adding that it's “hard to justify” from a purely economic standpoint. @abhishekn
@sharongoldman
Sharon Goldman on x
NEW: Excited (ok, stoked!) to share this @FortuneMagazine online/print deep-dive into how Mark Zuckerberg went all-in on open-source AI (yes, come at me, really open weights) and made Meta's Llama the cornerstone of the company's generative AI ambitions. https://fortune.com/...
@arian_ghashghai
Arian Ghashghai on x
A routine reminder that OpenAI uses Meta's OSS AI framework (PyTorch) (the framing that Zuck went “all-in” on oss as some kind of recent company-wide epiphany is... misrepresentative) [image]