“I would guess that sometime in the next 12 to 18 months, we'll reach the point where most of the code that's going toward these efforts is written by AI. And I don't mean autocomplete... you give it a goal, it can run tests, it can improve things, it can find issues, it writes higher quality code than the average very good person on the team already.”
Founder and CEO, Meta Platforms
The 18-month window closes at the end of October 2026. Zuckerberg's claim is scoped to code going toward Meta's AI/Llama efforts, not all Meta code, and Meta has not published a verifiable figure for the AI-written share.
“We're calling the first one Prometheus and it's coming online in '26. We're also building Hyperion, which will be able to scale up to 5GW over several years. We're building multiple more titan clusters as well. Just one of these covers a significant part of the footprint of Manhattan.”
Prometheus is under construction at Meta's New Albany, Ohio campus (gas turbines plus nuclear supply deals signed with Vistra, TerraPower and Oklo in January 2026); Meta still says it expects the cluster online sometime in 2026, so the claim cannot be resolved before 31 December 2026.
“[Meta's US investment is] probably going to be something like, I don't know, at least $600 billion through 2028 in the U.S.”
Founder, Chairman and CEO of Meta Platforms
Zuckerberg later said on a hot mic that he had not been sure which figure Trump wanted; he subsequently confirmed Meta was planning $600B+ in US investment through 2028, and in November 2025 Meta formally committed to over $600 billion in the US by 2028 for AI technology, infrastructure and workforce.
“I think that it's extremely unlikely if you look out five years from now, for example — whatever period of time you want — that you don't have billions of people with a personal agent that understands your goals and that is just working on your behalf 24/7 to achieve your goals.”
Founder, Chairman and CEO of Meta Platforms
Made during the analyst Q&A, where Zuckerberg argued consumer personal agents will be an 'extremely important and massive market'; Meta said over a million businesses already use its agents weekly on WhatsApp and Messenger.
“We're training the Llama 4 models on a cluster that is bigger than 100,000 H100s or bigger than anything that I've seen reported for what others are doing. I expect that the smaller Llama 4 models will be ready first, and they'll be ready, we expect, sometime early next year. And I think that they're going to be a big deal on several fronts, new modalities, capabilities, stronger reasoning, and much faster.”
Founder, Chairman and CEO of Meta Platforms
The smaller Llama 4 models (Scout and Maverick) shipped on April 5, 2025 — essentially 'early 2025' but just past the first quarter — and were natively multimodal as promised. However, the promised leap in reasoning did not materialize: the flagship Behemoth teacher model was repeatedly delayed and effectively shelved, no Llama 4 reasoning model was released, and the launch was marred by a benchmarking controversy.
“The amount of compute needed to train Llama 4 will likely be almost 10x more than what we used to train Llama 3, and future models will continue to grow beyond that.”
Founder, Chairman and CEO of Meta Platforms
Llama 4 as actually trained fell far short of a 10x compute scale-up over Llama 3.1 405B (~3.8e25 FLOP): Epoch AI estimates Llama 4 Maverick at roughly 2.2e24 FLOP, and the largest model, Behemoth — reported as trained on about 32,000 H100s — at roughly 5e25 FLOP, only around 1.3x Llama 3.1 405B. Behemoth was delayed through 2025 and never publicly released.
“We're building massive compute infrastructure to support our future roadmap, including 350,000 H100s by the end of this year — and overall almost 600,000 H100 equivalents of compute if you include other GPUs. Our long term vision is to build general intelligence, open source it responsibly, and make it widely available.”
Founder and CEO, Meta Platforms
Meta does not publish audited GPU counts, but its buildout tracked the target: analysts put Meta among Nvidia's two largest H100 buyers in 2024, and Zuckerberg's January 2025 target of 'more than 1.3 million GPUs' by end-2025 was reported as roughly a doubling of the fleet, consistent with ~600,000 H100-equivalents at the end of 2024.
“Starting next year, we expect future Llama models to become the most advanced in the industry.”
Founder and CEO, Meta Platforms
Llama 4 launched in April 2025 to a poor reception and a benchmark-submission controversy; the flagship Llama 4 Behemoth was delayed from April to mid-2025 and then reportedly shelved over weak internal performance, and Meta reorganized into a superintelligence lab reportedly weighing closed-source models. Llama was not the most advanced family in the industry at any point in 2025.
“We had a strong quarter, and Meta AI is on track to be the most used AI assistant in the world by the end of the year.”
Founder and CEO, Meta Platforms
By Meta's own metric the target was met on paper: Meta AI reached nearly 500 million monthly actives by September 2024 and almost 600 million by December, more than any rival's disclosed user base at the time. But the figure counts incidental exposure inside Facebook, Instagram and WhatsApp rather than deliberate use, and ChatGPT remained the assistant people actively chose, so 'most used' is contested.
“Probably in 2025, we at Meta, as well as the other companies that are basically working on this, are going to have an AI that can effectively be a sort of midlevel engineer that you have at your company that can write code.”
Founder and CEO, Meta Platforms
Coding agents advanced sharply in 2025 and now handle substantial amounts of real engineering work, but no lab — Meta included — fielded a system that autonomously fills a mid-level engineer's role; Meta shipped no such 'AI engineer' product in 2025 and Zuckerberg's own follow-on framing pushed the milestone of AI writing most of Meta's AI code into 2026.
“Meta is building a 2GW+ datacenter that is so large it would cover a significant part of Manhattan. We'll bring online ~1GW of compute in '25 and we'll end the year with more than 1.3 million GPUs. We're planning to invest $60-65B in capex this year.”
Founder and CEO, Meta Platforms
Meta overshot rather than missed its spending plan: full-year 2025 capex came in above the guided range at roughly $72 billion (with 2026 guidance far higher), and the Louisiana Hyperion campus proceeded as the 2GW+ build described. The compute-online and 1.3 million GPU figures are not independently auditable, so only the capex leg can be scored — and it was exceeded, not met.
“This will be a defining year for AI. In 2025, I expect Meta AI will be the leading assistant serving more than 1 billion people.”
Founder and CEO, Meta Platforms
The headcount target was hit early — Zuckerberg announced at Meta's May 2025 shareholder meeting that Meta AI had passed 1 billion monthly active users across Meta's apps. 'Leading assistant' is the weaker half: ChatGPT continued to dominate deliberate assistant usage and engagement, and Meta AI's reach is largely a byproduct of its placement inside Facebook, Instagram and WhatsApp.
“In 2025, I expect... Llama 4 will become the leading state of the art model, and we'll build an AI engineer that will start contributing increasing amounts of code to our R&D efforts.”
Founder and CEO, Meta Platforms
Llama 4 shipped in April 2025 but was not state of the art: it drew a benchmark-submission controversy, the flagship Behemoth was pushed back and then reportedly shelved for weak internal performance, and Meta pivoted to a new superintelligence lab reportedly considering closed models. AI-written code did grow at Meta, but the central prediction failed.