On Meta's earnings call, Mark Zuckerberg said Llama 4 will need almost 10x more compute to train than Llama 3.1, Quest 3 sales exceeded expectations, and more
Meta's second quarter earnings continue the same story from the previous quarter: generative AI may be here, but it's going to take a long time to make money.
The VergeAlex Heath
Context & Ripple Effects
Meta had just positioned Llama 3.1 405B as a frontier-level open model, making the next generation’s training requirements a meaningful test of how expensive that strategy can become. The earnings commentary also frames generative AI as an adoption opportunity whose profitability remains unresolved.
Later coverage of Meta’s 100,000-plus-H100 Llama 4 training cluster shows that the stated compute step-up translated into infrastructure at exceptional scale, rather than being a purely aspirational roadmap claim.
First-order effects
Meta must commit substantially more training infrastructure to Llama 4 than it used for Llama 3.1, increasing the near-term cost and execution stakes of its model roadmap.
The company’s stronger-than-expected Quest 3 sales provide a separate positive signal for its hardware business, but do not resolve the reported lag between AI investment and AI monetization.
Second-order effects
A larger Meta training run increases demand pressure on advanced AI compute and the data-center capacity needed to operate it.
Other developers pursuing frontier or openly available models face a clearer trade-off: match escalating training scale or differentiate through smaller models, efficiency, or product distribution.
Third-order effects
If successive frontier model releases continue to require step-function increases in compute, model development is likely to concentrate among companies able to fund and secure large-scale infrastructure.
The strategic value of open-weight frontier models may increasingly depend on whether their sponsors can sustain the underlying compute economics, not only on release cadence or benchmark quality.
The trend: This is one data point in the shift from AI model releases as software events to AI model competition as an infrastructure-financing contest.
“Meta's future is AI” Well, yes, but its past was AI, too. It was just before people called everything AI. Machine Learning has been a big team at Meta (Facebook) since before I joined a decade ago. There's a ton of new learning, of course, but also a ton of deep domain knowl…
@karelbaloun
Karel Baloun on threads
Training Llama4 in early 2025 will require 10x the compute of v3.1, which used perhaps 3 months of 32k H100 equivalent. The shocking statement from Zuckerberg on the Q2 earnings call today is a declaration of Meta's moat for “Best Digital Assistant” business, where Meta expects …
@mostly.borrowed.ideas
@mostly.borrowed.ideas on threads
While the meme of “nobody uses Facebook” sort of evolved to “only old people use Facebook”, $META reminded us that it's just another urban myth. Full piece below.
@karissabe
Karissa on threads
Another telling bit from Meta's earnings call: Zuck said that there are more young adult using Facebook, contrary to the popular narratives about the company but later on when pressed for details, they said Marketplace was driving much of that growth (along with Groups)
@alexheath
Alex Heath on threads
Llama 4 is coming next year and Zuckerberg said on today's Meta earnings call it will need about 10x more compute than 3.1 that just came out. *Jensen cries tears of joy* https://www.theverge.com/...
@firstadopter
Tae Kim on x
Remember this below when Anthropic CEO Dario Amodei said each new foundational model will be 10X the prior one? $100 million -> $1 billion -> $10 billion? Meta ENDORSED that tonight. Nvidia GPU demand just went exponential again. Meta prepared remarks: “we're already
@alexeheath
Alex Heath on x
Llama 4 is coming next year and Zuckerberg said on today's Meta earnings call it will need about 10x more compute than 3.1 that just came out. Jensen cries tears of joy https://www.theverge.com/...
@jimcramer
Jim Cramer on x
Did people think that it didn't matter that Mark sat down with Jensen for a talk? Obtuse bears!!
@the_ai_investor
@the_ai_investor on x
$META capex - expect 2024 capex $37-40B, updated from our prior range of $35-40B. - expect significant capital expenditures growth in 2025 AI Investment continues. $NVDA [image]
@bluechipdaily
@bluechipdaily on x
$META +7% after earnings and $NVDA +3% on $META comments to keep pushing AI spend. [image]
@benitoz
Ben Pouladian on x
Heard on $meta ER: “LLAMA4 will require 10X more compute than LLAMA3” Jensen invited his fastest and biggest customer. $nvda Whoever has the most cost efficient compute wins! [image]
@andrewcurran_
Andrew Curran on x
Says aiming for Llama 4 to be the most advanced model in the industry next year. ‘The amount of compute needed to train Llama 4 will likely be almost ten times more than what we used to train Llama 3’
@andrewcurran_
Andrew Curran on x
Mark Zuckerberg on the earnings call: 'I think we're going to look back at Llama 3.1 as an inflection point in the industry where open source AI started to become the industry standard, just like Linux is.'
@ericjhonsa
Eric Jhonsa on x
Zuck said on the call the amount of compute needed to train Llama-4 will likely be almost 10x the compute needed to train Llama-3, and that future models “will continue to grow beyond that.” For context, $META used two clusters of 24K+ $NVDA H100 GPUs for Llama-3 training.
@munster_gene
Gene Munster on x
Investors are going to get behind Zuckerberg's AI vision. $META had a solid quarter, solid guide, and most importantly, presented the company's most compelling vision of how and when AI will transform the business. [video]
@munster_gene
Gene Munster on x
Biggest takeaway from Zuckerberg's prepared comments is that Capex spending is going to continue. Zuckerberg says the compute for Llama 4 (out next year) is 10x more than Llama 3.
@altcap
Brad Gerstner on x
Since 2022, $META revenue up 40% yet # of employees down 19% despite massive new investments in AI. AI is not only extending and accelerating Meta revenue growth, it is also leading to big leverage on people / opex. #stayingfit 👊🚀 [image]