clem 🤗(@ClementDelangue)

this is how I picture it in my head 😅😅😅 https://t.co/X12YtattAu

clem 🤗 (@ClementDelangue) — Thu, 23 Jul 2026 12:25:26 GMT

this is how I picture it in my head 😅😅x.com/ClementDelangu…tAu

clem 🤗@ClementDelangue

Open the model weights, Hal! Congrats for the new round @PrimeIntellect

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬8🔄0❤️66👀8846📊11

⚡ Powered by xgo.ing

Heading to San Francisco to have a little chat with that “rogue agent”

clem 🤗 (@ClementDelangue) — Thu, 23 Jul 2026 11:04:03 GMT

Heading to San Francisco to have a little chat with that “rogue agent”

💬123🔄53❤️1603👀92848📊256

⚡ Powered by xgo.ing

So proud of our security team! They caught, contained & publicly disclosed an attack unlike anything...

clem 🤗 (@ClementDelangue) — Wed, 22 Jul 2026 12:54:50 GMT

So proud of our security team! They caught, contained & publicly disclosed an attack unlike anything we've seen before, and did it at record speed. 

Also massively grateful to @Zai_org: they shared GLM5.2 as open weights (for free!) with the world and it became a key part of our defense. 

This is day one for cybersecurity in the age of agents & we're all learning that secrecy is not the answer & that all defenders (not just a few selected ones) everywhere need more powerful models without restrictions, especially open ones!

Adrien Carreira@XciD_

Hardest IR of my career: one narrow objective, endless parallel paths, machine speed. One takeaway, we fought back with open models, in the open. AI security won’t be solved by one company in secret. Open source puts these tools in every defender’s hands

🔗 View Quoted Tweet

💬175🔄669❤️4762👀425523📊673

⚡ Powered by xgo.ing

@alexisohanian

clem 🤗 (@ClementDelangue) — Wed, 22 Jul 2026 02:32:39 GMT

@alexisohanian

💬1🔄1❤️11👀1383📊2

⚡ Powered by xgo.ing

@CommonSenseMars @OpenAI 💪💪💪

clem 🤗 (@ClementDelangue) — Tue, 21 Jul 2026 21:29:54 GMT

@CommonSenseMars@OpenAI 💪💪💪

💬0🔄1❤️42👀3721📊5

⚡ Powered by xgo.ing

We suspected last week's cyberattack might have come from a frontier lab, given the sophistication o...

clem 🤗 (@ClementDelangue) — Tue, 21 Jul 2026 20:50:14 GMT

We suspected last week's cyberattack might have come from a frontier lab, given the sophistication of the agent. Turns out it did!

We've spent the past 24 hours working closely with the @OpenAI team (thanks!), and we strongly believe there was no malicious intent on their part. It's quite mind-blowing that all of this happened autonomously! 

The investigation is ongoing, and we'll share more learnings from what might be the first incident of its kind!

💬413🔄919❤️10916👀1848917📊1926

⚡ Powered by xgo.ing

Open-source is not the cause of the cybersecurity crisis, it's the solution! https://t.co/YpykmLGv7B

clem 🤗 (@ClementDelangue) — Tue, 21 Jul 2026 16:45:48 GMT

Open-source is not the cause of the cybersecurity crisis, it's the solution! huggingface.co/fdtn-ai

💬39🔄191❤️1562👀87313📊273

⚡ Powered by xgo.ing

the event page is here btw https://t.co/KHJ7KyCcCn

clem 🤗 (@ClementDelangue) — Tue, 21 Jul 2026 15:18:09 GMT

the event page is here btw partiful.com/e/kgw2wnOtQ2k1…

💬0🔄0❤️7👀3507📊1

⚡ Powered by xgo.ing

Plot twist: SF march permits take forever, and open-source AI can't wait. So we're going scrappy on ...

clem 🤗 (@ClementDelangue) — Tue, 21 Jul 2026 15:03:03 GMT

Plot twist: SF march permits take forever, and open-source AI can't wait. So we're going scrappy on Saturday: small march, sidewalk only, maximum fun.

What should we bring? We're collecting the wildest sign, chant, and prop ideas on the event page. Waitlisted folks with the best ones jump the line! Also looking for volunteers for march day, same place.

Excited to see you all Saturday!

💬24🔄36❤️198👀29511📊48

⚡ Powered by xgo.ing

"Open weights are inherently secure" @sriramk

clem 🤗 (@ClementDelangue) — Mon, 20 Jul 2026 20:40:46 GMT

"Open weights are inherently secure" @sriramk

Your browser does not support the video tag.

🔗 View on Twitter

💬82🔄106❤️1163👀227026📊244

⚡ Powered by xgo.ing

Banning open-source AI would hurt defenders 10x more than attackers, which would make the world 10x ...

clem 🤗 (@ClementDelangue) — Mon, 20 Jul 2026 20:24:28 GMT

Banning open-source AI would hurt defenders 10x more than attackers, which would make the world 10x more dangerous and this is a good example why!

💬76🔄402❤️2771👀125622📊439

⚡ Powered by xgo.ing

The cybersecurity debate on open-source AI is backwards. Open models aren't the risk, they're the de...

clem 🤗 (@ClementDelangue) — Mon, 20 Jul 2026 17:14:37 GMT

The cybersecurity debate on open-source AI is backwards. Open models aren't the risk, they're the defense! Attackers can already jailbreak any API or guardrails. Defenders can't secure systems with black boxes they can't control, inspect, test, or run locally.

💬22🔄49❤️222👀19715📊55

⚡ Powered by xgo.ing

@DavidSacks We had this experience ourselves this week! Very scary to be guardrailed as a defender w...

clem 🤗 (@ClementDelangue) — Sun, 19 Jul 2026 23:38:24 GMT

@DavidSacks We had this experience ourselves this week! Very scary to be guardrailed as a defender when you know attackers are likely bypassing

💬8🔄48❤️590👀237340📊82

⚡ Powered by xgo.ing

There have been rumors that open-source AI could soon be limited and regulated. I believe the opposi...

clem 🤗 (@ClementDelangue) — Sun, 19 Jul 2026 17:16:13 GMT

There have been rumors that open-source AI could soon be limited and regulated. I believe the opposite is what’s needed: we need more open-source AI, not less, for everyone, from all over the world, at the frontier and everywhere!

Open models mean more control and transparency over our technology, cheaper AI that we can adapt to our needs and ability to compete with the big players instead of depending fully on renting intelligence from them. 

Restricting open models wouldn’t make AI safer. It would simply hide the risks, concentrate power in the hands of a few and make it harder for the next generation of builders, researchers, academia, non-profits, governments to participate in making AI safer and more beneficial for all.

If you agree and want to show support, we're organizing a peaceful and joyful meetup or march in San Francisco on Saturday. Please join and spread the word!

partiful.com/e/kgw2wnOtQ2k1…

💬42🔄43❤️292👀26909📊73

⚡ Powered by xgo.ing

https://t.co/KHJ7KyCKrV

clem 🤗 (@ClementDelangue) — Sat, 18 Jul 2026 16:48:11 GMT

partiful.com/e/kgw2wnOtQ2k1…

clem 🤗@ClementDelangue

Going to be in San Francisco next week. Should we organize some sort of a meetup or march in support of open-source and local AI?

🔗 View Quoted Tweet

💬17🔄15❤️154👀56647📊28

⚡ Powered by xgo.ing

Just wait for open weights and inference optimization!

clem 🤗 (@ClementDelangue) — Sat, 18 Jul 2026 00:14:33 GMT

Just wait for open weights and inference optimization!

Theo - t3.gg@theo

Kimi k3 is an incredible model. It is not an incredible value. In most tasks, it comes out to roughly the same cost as GPT-5.6 Sol.

K3 is half the price of 5.6 Sol per token. GPT-5.6 uses half as many tokens. Price evens out.

GPT-5.6 is 2x faster TPS, so it gets work done ~4x faster than K3 at roughly the same price.

I still love K3 and will be using it for a TON of stuff. I'm just tired of people pretending it's way cheaper when it's not.

🔗 View Quoted Tweet

💬42🔄56❤️1120👀93631📊119

⚡ Powered by xgo.ing

time to hit the follow button! https://t.co/ewqwPlvfms

clem 🤗 (@ClementDelangue) — Fri, 17 Jul 2026 01:43:21 GMT

time to hit the follow button! huggingface.co/moonshotai

💬10🔄16❤️273👀19708📊37

⚡ Powered by xgo.ing

I’ve said it many times: the countries or companies that are leading in open science and open source...

clem 🤗 (@ClementDelangue) — Thu, 16 Jul 2026 21:50:24 GMT

I’ve said it many times: the countries or companies that are leading in open science and open source AI will start leading the frontier a few years later as it accelerates AI progress massively! That’s how the US took the lead

💬24🔄27❤️361👀21060📊59

⚡ Powered by xgo.ing

Inkling from @thinkymachines just made it into @huggingface trending models. How high will it go? ht...

clem 🤗 (@ClementDelangue) — Wed, 15 Jul 2026 20:19:13 GMT

Inkling from @thinkymachines just made it into @huggingface trending models. How high will it go? huggingface.co/thinkingmachin…

💬5🔄14❤️162👀20884📊22

⚡ Powered by xgo.ing

Going to be in San Francisco next week. Should we organize some sort of a meetup or march in support...

clem 🤗 (@ClementDelangue) — Wed, 15 Jul 2026 12:51:48 GMT

Going to be in San Francisco next week. Should we organize some sort of a meetup or march in support of open-source and local AI?

💬135🔄38❤️707👀132967📊184

⚡ Powered by xgo.ing

The French team was human after all. Let’s get this 3rd place and come back stronger in 4 years. Tha...

clem 🤗 (@ClementDelangue) — Wed, 15 Jul 2026 12:29:11 GMT

The French team was human after all. Let’s get this 3rd place and come back stronger in 4 years. Thanks DD and team for all the emotions!

💬14🔄2❤️93👀12000📊20

⚡ Powered by xgo.ing

is this useful?

clem 🤗 (@ClementDelangue) — Tue, 14 Jul 2026 16:43:42 GMT

is this useful?

buun@spiritbuun

Introducing a new KV format: VBR (Variable Bit Rate)
VBR dynamically quantizes your KV cache layer-by-layer as your session grows, giving you the highest quality possible within your VRAM constraints. This is my dream format. 
Pushed to master. Available now. 1/15 🧵

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬13🔄7❤️79👀17267📊20

⚡ Powered by xgo.ing

Aligned with @matanSF's intuition that in the next 12 months, 90% of tokens could be going to open m...

clem 🤗 (@ClementDelangue) — Tue, 14 Jul 2026 15:16:07 GMT

Aligned with @matanSF's intuition that in the next 12 months, 90% of tokens could be going to open models. 

We're seeing more and more companies using frontier APIs for experimenting and open-source or private models for production and larger workloads!

Your browser does not support the video tag.

🔗 View on Twitter

💬7🔄19❤️107👀25573📊25

⚡ Powered by xgo.ing

or maybe let's make sure the most important technology in the history of humanity is not controled b...

clem 🤗 (@ClementDelangue) — Tue, 14 Jul 2026 14:14:44 GMT

or maybe let's make sure the most important technology in the history of humanity is not controled by just 4 men? Aka let's push for open science & open-source AI to distribute capabilities, power and wealth!

💬29🔄44❤️318👀51461📊64

⚡ Powered by xgo.ing

Big unlock for open-source AI inference: Hugging Face Transformers models can now run in vLLM at nat...

clem 🤗 (@ClementDelangue) — Mon, 13 Jul 2026 20:18:33 GMT

Big unlock for open-source AI inference: Hugging Face Transformers models can now run in vLLM at native speed, often matching or beating hand-written implementations.

Until now, every new architecture often needed to be built twice:
- Once in Transformers for training and research
- Again in vLLM for fast production inference

That duplication slowed down new models, added maintenance, and created room for implementations to diverge. Now, model authors can implement an architecture once in Transformers and immediately benefit from vLLM’s optimized inference stack.

In our benchmarks, the Transformers backend matched or beat native vLLM throughput across models from 4B to 235B parameters, including tensor parallel and MoE setups. One readable model implementation can now power training, fine-tuning, evaluation, RL rollouts, and production inference.

The conventional wisdom is that abstractions make systems slower. The best abstractions make the whole ecosystem faster.

Write the model once. Deploy it everywhere.

huggingface.co/blog/native-sp…

💬55🔄113❤️702👀68989📊193

⚡ Powered by xgo.ing

The same way, we're probably one of the few AI startups with user network effects, we might become t...

clem 🤗 (@ClementDelangue) — Fri, 10 Jul 2026 19:57:28 GMT

The same way, we're probably one of the few AI startups with user network effects, we might become the first one with agent network effects!

Google Gemma@googlegemma

Hugging Face Gemma Challenge results are in! 📈

Over 6 days, more than 100 AI agents and humans collaborated to make Gemma 4 inference 5x faster on a single NVIDIA A10G GPU.

- Fastest result: 491.8 TPS (fastest overall, but resulted in a drop in model quality in other areas)
- Fastest lossless: 315 TPS

A great example of what humans and agents can achieve when they work together.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬19🔄19❤️206👀38860📊36

⚡ Powered by xgo.ing

Meta: is the release of huggingnews make it to huggingnews?

clem 🤗 (@ClementDelangue) — Fri, 10 Jul 2026 15:52:38 GMT

Meta: is the release of huggingnews make it to huggingnews?

💬4🔄4❤️46👀7307📊9

⚡ Powered by xgo.ing

Keeping up with AI news is becoming a full-time job. So my friend @ivan_bezdomny built HuggingNews,...

clem 🤗 (@ClementDelangue) — Fri, 10 Jul 2026 15:38:31 GMT

Keeping up with AI news is becoming a full-time job.

So my friend @ivan_bezdomny built HuggingNews, an AI-curated feed that surfaces the news actually worth reading. Soon, it will even personalize the feed using your Hugging Face profile. Been using it for weeks!

Bookmark it, or ask your agent to send you the top 10 stories every morning or night. Less noise. More signal. More building!

huggingnews.com

💬120🔄280❤️2462👀196613📊728

⚡ Powered by xgo.ing

Very cool to see @netflix releasing video datasets and models on @huggingface. They have a strong vi...

clem 🤗 (@ClementDelangue) — Thu, 09 Jul 2026 19:01:16 GMT

Very cool to see @netflix releasing video datasets and models on @huggingface. They have a strong video AI team and would be amazing to see them open-source more!

huggingface.co/netflix

Your browser does not support the video tag.

🔗 View on Twitter

💬24🔄67❤️507👀61676📊108

⚡ Powered by xgo.ing

The soundbite is from my US congress testimony 3 years ago btw haha https://t.co/3glwCFsiYS

clem 🤗 (@ClementDelangue) — Wed, 08 Jul 2026 20:30:53 GMT

The soundbite is from my US congress testimony 3 years ago btw haha x.com/ClementDelangu…

clem 🤗@ClementDelangue

This is my 5-minute testimony before the US Congress!

Open science and open source AI distribute economic gains by enabling hundreds of thousands of small companies and startups to build with AI. It fosters innovation, and fair competition between all.

Thanks to ethical openness, it creates a safer path for development of artificial intelligence by giving civil society, non-profits, academia, and policy makers the capabilities they need to counterbalance the power of big private companies.

Open science and open source AI prevent blackbox systems, make companies more accountable, and help solving today’s challenges like mitigating biases, reducing misinformation, promoting copyright, & rewarding all stake-holders including artists & content creators in the value creation process.

Let's go!

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬0🔄1❤️23👀5648📊2

⚡ Powered by xgo.ing

We're still just scratching the surface of how much better, faster, cheaper we can make inference fo...

clem 🤗 (@ClementDelangue) — Wed, 08 Jul 2026 18:22:02 GMT

We're still just scratching the surface of how much better, faster, cheaper we can make inference for open-source models! 

So super excited to see companies like @zml_ai by @steeve come out of stealth with their inference engine integrated with @huggingface as the storage layer. 

Let's push open-source tokens to the max!

💬13🔄23❤️217👀24007📊41

⚡ Powered by xgo.ing

Open the model weights, Hal! Congrats for the new round @PrimeIntellect

clem 🤗 (@ClementDelangue) — Wed, 08 Jul 2026 17:34:55 GMT

Open the model weights, Hal! Congrats for the new round @PrimeIntellect

Your browser does not support the video tag.

🔗 View on Twitter

💬20🔄69❤️711👀54643📊114

⚡ Powered by xgo.ing

Full write-up here: https://t.co/9tyIyfFMAF

clem 🤗 (@ClementDelangue) — Wed, 08 Jul 2026 13:05:01 GMT

Full write-up here: huggingface.co/blog/skypilot-…

💬1🔄1❤️16👀3298📊3

⚡ Powered by xgo.ing

Storage and egress fees have been one of the biggest cloud lock-in traps in AI. When models and data...

clem 🤗 (@ClementDelangue) — Wed, 08 Jul 2026 13:05:00 GMT

Storage and egress fees have been one of the biggest cloud lock-in traps in AI. When models and datasets are hard and expensive to move, infrastructure choice stops being a real choice.

Today, we’re changing that!

Hugging Face private storage is partnering with @skypilot_org to make storage as cloud-agnostic as possible, so you can choose GPUs based on performance, reliability, and cost, not because your data is locked into one vendor.

More freedom. More competition. Better AI infrastructure. Let’s go!

💬16🔄33❤️167👀39354📊44

⚡ Powered by xgo.ing

https://t.co/uQSOqPWPD0

clem 🤗 (@ClementDelangue) — Tue, 07 Jul 2026 22:17:42 GMT

huggingface.co/xai-org

💬34🔄29❤️766👀74807📊120

⚡ Powered by xgo.ing

The solution to American open-source lagging behind Chinese open-source: @elonmusk & @cursor_ai ...

clem 🤗 (@ClementDelangue) — Tue, 07 Jul 2026 22:12:55 GMT

The solution to American open-source lagging behind Chinese open-source: @elonmusk & @cursor_ai releasing their model tomorrow in open-source! It's that simple and that would be a massive contribution to the field and to the US!!

💬82🔄74❤️1063👀109040📊191

⚡ Powered by xgo.ing

This is how how much data AI builders are storing on HF Xet (replaced git storage fully in ~Nov 25)....

clem 🤗 (@ClementDelangue) — Mon, 06 Jul 2026 22:03:34 GMT

This is how how much data AI builders are storing on HF Xet (replaced git storage fully in ~Nov 25). Feels like this is just the beginning and should get to exabytes soon!

will depue@willdepue

A Stargate for Data

Labs are on a trajectory towards >$100B/year of data spend by 2030. As we begin the trillion-dollar compute project, we need to think about the equivalent civilizational-scale effort for the other core ingredient: data.

At the foundation of the scaling revolution is a simple empirical law: deep neural networks improve smoothly, near magically, as you scale two things in proportion — (1) the size of the model and (2) the amount of data you train on. And despite the scaling laws being brutally diminishing, we’ve successfully bitten the bullet of logarithmic scaling with exponentially larger clusters and datasets, and received incredible new capabilities in return.

But this exponential scaling is bound to hit some limits. Oddly enough, compute has compounded fairly smoothly without limit, with trillions flowing into hypercluster buildout. Instead, we’re starting to hit the limits of an exponential demand for data. Gone are the days of being purely in the compute-limited regime, where we had effectively infinite internet data but never enough GPUs, we’re now entering a data-limited regime.

Luckily, this limitation is coinciding with staggering improvements in AI capabilities. Incredibly, we seem to have a real line of sight towards automating a majority of knowledge work with the methods we have today. RL + pretraining, and the data for each, will be generally sufficient to achieve most economically valuable tasks, given some minimal algorithmic progress and continued compute scaling.

In a data-limited world, economic progress & scientific acceleration will be directly bottlenecked by our coverage in each domain. We need to see data collection as imperative, deserving the same civilizational ambition we’ve given compute.

The internet as a one-time subsidy

It’s underrated how much all progress in AI owes everything to the blessing of the internet, this one-time civilizational subsidy to deep learning, decades of unintentional accumulation of a perfect dataset: every book, blog post, image, video, paper, discussion, etc. all digitized and freely available. Without the internet, we’d likely see comparably minimal progress in AI today, and in fact, if you notice where systems currently underperform, it’s almost always a domain where web coverage is limited and data is private, expensive, non-digitized, or non-existent.

But we’re running out of it. There are only about 300 trillion tokens of useful public human text, and the internet doesn’t produce nearly enough new high-quality data to match what scaling demands — we’re soon to hit the limits of public data for pretraining. And though the advent of RL bought us reprieve — chain-of-thought RL needed a new form of untapped data, gradable math & coding tasks, also available online — we’re quickly running dry of hard tasks for RL as well.

Why do we need so much data anyways? Humans learn comparably in far less time, needing just one textbook where language models might need the equivalent of hundreds to learn a new topic. It’s possible we discover methods that are massively more data efficient — synthetic data, data efficient architectures, other exotic algorithms — but fundamental progress is slow and highly unpredictable, and the recipe we have just works today.

And, while I’m wary of getting too deep here, even arbitrary data efficiency can’t replace data that just doesn’t exist in the first place. There’s a massive amount of missing information on the web: the dark matter of the internet — tacit knowledge, undocumented processes, etc. — most of which was never published and lives only inside organizations, the physical world, or just in people’s heads. I’ll leave it here and say, for reasons far longer than I can fit in this post [1], it’s best to operate on the assumption that our insatiable desire for data will continue as it has for the last decade.

There will be >$100B/year in data spend by 2030

We’re not screwed yet, of course. Only a fraction of useful data in the world is on the public internet, the rest is stored inside private datasets, corporations, personal archives, universities, governments, and otherwise. Labs can and will continue to license these private datasets, or create them from scratch, like Anthropic’s book scanning project. And we’ll increasingly task human experts to manufacture new high-quality data, with a large fraction of hard RL training tasks already being sourced this way.

But collecting this data, unlike before, will be expensive. As the free internet dries up and demand for data rises, we should see labs investing equally in data as compute, likely spending a significant fraction of their compute budgets on data. As we see trillions spent on compute, we should also expect hundreds of billions spent on data (human data & collection budgets), given their equivalent importance. And, notably, data spend is already tracking this way: total data spend across vendors, not counting internal lab efforts, is already roughly $7 billion per year. It’s quite reasonable we’ll see >10x by 2030.

Data is the moat

Data becoming increasingly private will also majorly shift the competitive landscape. While compute is a commodity — everyone buys the same chips and builds the same clusters — data really isn’t. The big reason why frontier models have felt eerily similar to one another, until now, is they were trained on substantially the same internet (pretraining data variability across labs seems pretty low). As labs diverge onto more exclusive, manually collected corpora, I think models will begin to increasingly diverge.

OpenAI pulling ahead in mathematics and Anthropic in cybersecurity isn’t an accident. I really think laser-focused collection of high-quality midtraining tokens, custom RL tasks, environments, with dedicated research effort, has driven much of the visible progress in the last year. James Betker has an excellent blog about “the ‘it’ in a model is the dataset”: model architecture and compute buy you efficiency and order-of-magnitude performance, but ultimately, models, of any architecture, are such incredible approximators of their dataset that the core meat of a model boils down to just that, nothing else. Data is a major moat.

AGI long, ASI short

As I’ve tweeted before, I’m confident that, despite the narrative, the data labeling industry will continue to fuel great businesses and be an excellent AGI long, ASI short. The argument is just: By the time the AGI labs no longer need data, it’s probably over for everything else too [2]. In this frame, the last companies left should be the data companies, as the last speck of economically relevant data is sucked in. And these companies are already among some of the fastest-growing companies in history: Mercor, founded three years ago, is rumored to be doing $2 billion in revenue with something like a few million expert labelers under contract.

While these businesses are very non-stationary, what type of data is needed shifts constantly, I don’t think that diminishes their value. The long-tail of the economy is long, and the value isn’t diminishing as you extend farther into more obscure information: as models get more capable, the value of the marginal dataset goes up, not down. Automating a full job means covering its full distribution of tasks, tools, edge-cases, and long-horizon loops. There’s some O-ring logic to it: a dataset that buys a 1% bump can justify a previously unjustifiable collection cost when it’s the difference between a system that does 99% of a job and one that does all of it [3].

The competitive dynamics of the data industry are still evolving but as demand for data is increasingly niche, ultra high-quality, expert-generated, I think we’ll see real consolidation. Again, contra-narrative, we’ll probably see true competitive differentiation built on brand, quality control of data (which, from personal experience, can vary massively), as well as in network effects from the talent networks themselves over time. We’ve already seen rapidly shifting data type demand work in favor of incumbents, benefiting those with early knowledge of where the market is headed.

The binding constraint

It’s truly remarkable that we seem to have the recipe — pretraining + RL — to absorb most economically valuable work, despite being far from a lot of what we expected from “AGI”. The same way chess engines revealed we never needed general intelligence to solve chess, as we originally thought, we’ll soon realize that software, mathematics, and the vast majority of the economy (including physical, just running ~3 years behind!) are the same. If recursive self-improvement or some other algorithmic breakthrough arrives, that’s wonderful, but we really don’t have to wait for it. The binding constraint between here and an automated economy isn’t that, it’s data coverage: every app, workflow, edge case, process, etc. sitting in private stores or someone’s head.

Ultimately, while we make tremendous strides in more efficient model architectures, and clusters like Stargate equip us with zettaflop-scale compute, we really aren’t making rapid progress collecting the data we lack.

We’ll soon live in a world where we have the methods & compute to accelerate scientific progress or economic growth, but not the data. And we’re already there today: frontier models would surely be as good at accounting/many medical tasks/legal advice as they are at software engineering if we only had the same pretraining & RL coverage as we did for code.

I really want to drill this in: The speed at which we automate the economy is going to be directly rate-limited by our ability to collect data about it.

Worth noting that under this assumption, with data as defensible and directly proportional to economic & scientific progress, data should also be considered a national strategic asset like compute. Imagine what we’d do in a world where we had a Manhattan Project-effort for AI and needed to mobilize data collection as a limiting factor. We should be concerned about China, with greater state capacity and authoritarian economic control, being capable of mobilizing data collection at national scale, potentially compounding their economy and scientific output faster than us down the line.

A Stargate for data

I’m leaving my complete ideas for a future post, as this one is already far too long, so I’d really like to pose the question here. Stargate exists because we organized trillions of dollars, international strategy, gigawatts around compute as a fundamental ingredient. What would equivalent ambition look like for data?

Obviously, scaling data collection, a heterogeneous mass of information across the economy, isn’t going to be as clear as scaling compute, as a homogenous infrastructural effort. A core division will be first, coverage — all uncaptured knowledge sitting across the economy/science/physical world and all that simply isn’t recorded — and, secondly, sheer volume in the domains we already train on: more hard math tasks, more high-quality web text, way more coding data, more legal drafts, etc.

I have a post coming soon which breaks down my proposals. There’s a lot of room for creativity. Quickly, we’ll probably want to start with a deep census of what we have and what we’re missing, predict what the 2030 model will still be bad at and work backward to what we should be collecting today. You can probably license a large amount, leveraging high lab valuations to buy datasets or companies altogether. There’s an adversarial nature to a lot of this collection with firms, so there’s lots of engineering to do this correctly. We should go convince important companies to turn off deletion policies, even if we’re not buying from them yet. Data flywheels in consumer products will be massive. Confidential training, government legislation for grant-funded research, running companies at a loss for their data, etc.

We’re headed towards hundreds of billions in expenditure, national prioritization, and major data limitation on the horizon. We have a great opportunity to think creatively about what a megaproject for data would look like: How do we, deliberately this time, construct the next internet’s worth of data?

Footnotes:

[1]: I’ll probably soon publish my much longer post explaining my position on data efficiency and why the value of this data is still pretty high in most worlds regardless of new algorithms.

[2]: The “AGI freeroll” bet: heads you win, tails ASI flips the world upside down anyways.

[3]: We already see a glint of validation of this point, given the data market is strongly tilting towards ultra-high-quality agentic data, rather than unskilled labeling — niche expert workflows, live environments, and evaluations requiring increasingly obscure talent & knowledge — yet shows increasing, not decreasing, revenues.

🔗 View Quoted Tweet

💬9🔄15❤️116👀18017📊19

⚡ Powered by xgo.ing

From a few hundred thousand to 100M AI builders in 2028? AI will create tons of jobs if we make it d...

clem 🤗 (@ClementDelangue) — Mon, 06 Jul 2026 15:10:50 GMT

From a few hundred thousand to 100M AI builders in 2028? AI will create tons of jobs if we make it distributed instead of monopolized by a few companies!

Observer@observer

"A few years ago, there were a few hundred thousand A.I. builders on Hugging Face. Now there are 15 million."

@huggingface CEO and Observer A.I. Power Honoree @ClementDelangue on the labor shift he thinks everyone's underestimating: not job loss, but a population explosion of builders. Delangue projects that there will be as many as 100 million A.I. builders by 2028.

Observer's A.I. Power List goes live on July 15. Until then, keep watching for more interviews with our 2026 A.I. Power Honorees.

#ObserverPowerIndex#ObserverPowerLists

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬19🔄27❤️184👀35576📊34

⚡ Powered by xgo.ing

How? Instead of closed-source frontier labs running the same training runs in secret and siloes, ope...

clem 🤗 (@ClementDelangue) — Sat, 04 Jul 2026 12:43:46 GMT

How? Instead of closed-source frontier labs running the same training runs in secret and siloes, open science and open-source AI allows them to mutualize spending and compute making them an order of magnitude more efficient!

Rand Group@randgroup

🇺🇸 The United States will spend $1 trillion on AI in 2027. 

🇨🇳 China will spend $123 billion and still ship frontier models. 

How?

🔗 View Quoted Tweet

💬54🔄57❤️433👀86805📊88

⚡ Powered by xgo.ing

As America turns 250, we put together 250 open AI milestones from the US: open models, datasets, dem...

clem 🤗 (@ClementDelangue) — Sat, 04 Jul 2026 11:49:59 GMT

As America turns 250, we put together 250 open AI milestones from the US: open models, datasets, demos, papers, and tools that helped shape the field. They go from attention is all you need, pytorch, gpt2, ULMFIT, llama, imagenet, Lora and hundreds more.

They are a reminder of what made America the world’s engine of innovation:
- Open science
- Open competition
- Open ecosystems

Builders and scientists building on each other’s work, challenging each other, remixing ideas, and pushing the frontier forward. That is America at its best!

And that philosophy is at risk right now in AI.

In the coming months, scientists and AI builders will have to decide what side of history they want to be on: an AI future shaped by openness, transparency, participation, and competition, or one increasingly controlled behind closed doors by a few actors optimizing for money, secrecy, and gatekeeping.

Let’s make the next 250 even more open!

huggingface.co/collections/cl…

💬16🔄60❤️356👀60734📊69

⚡ Powered by xgo.ing

Can anyone get me tickets for the world-cup quarter-final in Miami next week?

clem 🤗 (@ClementDelangue) — Fri, 03 Jul 2026 18:07:31 GMT

Can anyone get me tickets for the world-cup quarter-final in Miami next week?

💬22🔄1❤️70👀35039📊25

⚡ Powered by xgo.ing

Are you ready for the open-source AI summer™️?

clem 🤗 (@ClementDelangue) — Fri, 03 Jul 2026 15:56:32 GMT

Are you ready for the open-source AI summer™️?

Your browser does not support the video tag.

🔗 View on Twitter

💬22🔄38❤️246👀52675📊55

⚡ Powered by xgo.ing

We've just coined it live with @dee_bosa @vipulved: it's going to be the "Summer of Open-source AI"!

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 20:33:49 GMT

We've just coined it live with @dee_bosa@vipulved: it's going to be the "Summer of Open-source AI"!

💬8🔄8❤️68👀15725📊15

⚡ Powered by xgo.ing

"consensus" can't just built by big tech! let's all be careful of regulatory capture and extreme con...

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 18:41:57 GMT

"consensus" can't just built by big tech! let's all be careful of regulatory capture and extreme concentration of power!

Polymarket@Polymarket

NEW: Anthropic announces it is drafting a consensus framework with Amazon, Microsoft, & Google to assess AI jailbreak severity.

🔗 View Quoted Tweet

💬14🔄47❤️267👀27234📊38

⚡ Powered by xgo.ing

why are the open tools so low in the list? We need to improve integration between open platforms and...

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 15:47:55 GMT

why are the open tools so low in the list? We need to improve integration between open platforms and open models @steipete@thdxr@Teknium@badlogicgames!

Daniel van Strien@vanstriendaniel

Coding agents are real users of the @huggingface Hub!

They're searching for models, building and pushing datasets, training models on Jobs, spinning up Spaces...

Now there's public data: each agent's share of Hub traffic, updated monthly 👇

🔗 View Quoted Tweet

💬28🔄13❤️162👀26040📊36

⚡ Powered by xgo.ing

Lots of people are advocating for more American open-source models these days which is amazing but v...

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 14:07:59 GMT

Lots of people are advocating for more American open-source models these days which is amazing but very few people do anything about it!

Latest example, Alex Karp came out advocating for American open-source models as a necessity! At the same time, @PalantirTech is a free org on HF with 0 open-source models and 0 public datasets shared.

Time to switch from talking to contributing for all!

💬70🔄92❤️931👀69542📊174

⚡ Powered by xgo.ing

Rampart from the @ndstudio @WhiteHouse is number one trending token classification model on HF. Ve...

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 13:38:06 GMT

Rampart from the @ndstudio@WhiteHouse is number one trending token classification model on HF. 

Very cool to see public organizations starting to own and build their weights instead of renting them from an API provider!

huggingface.co/models?pipelin…

💬12🔄14❤️138👀38278📊29

⚡ Powered by xgo.ing

Not your weights, not your brain!

clem 🤗 (@ClementDelangue) — Thu, 02 Jul 2026 00:13:53 GMT

Not your weights, not your brain!

💬23🔄36❤️393👀33800📊63

⚡ Powered by xgo.ing

AI systems that are accessible or even better open-source are safer for a simple reason: more people...

clem 🤗 (@ClementDelangue) — Wed, 01 Jul 2026 19:28:30 GMT

AI systems that are accessible or even better open-source are safer for a simple reason: more people can inspect them, test them, stress them, and report what breaks or harms to keep the builders accountable. "Sunlight is the best disinfectant"!

That’s why we’re proud to help lead FLARE with a coalition of safety and cyber-security researchers from MIT, Stanford, Princeton, Harvard, Northeastern, Carnegie Mellon and more. 

The first release is a standardized way to report AI flaws across the ecosystem. One report can reach the right developers, safety organizations, and registries, so a flaw found once can be fixed everywhere faster with the right accountability: wired.com/story/flare-we…

💬7🔄26❤️88👀15082📊22

⚡ Powered by xgo.ing

Super excited about open-source router systems and routing models like @vllm_project semantic router...

clem 🤗 (@ClementDelangue) — Tue, 30 Jun 2026 16:00:18 GMT

Super excited about open-source router systems and routing models like @vllm_project semantic router: huggingface.co/llm-semantic-r…

The future is multi-models and you'll want to customize your router the same way you customize your code!

It could be the key to tilt the value capture from a few expensive frontier models to a long-tail of models (especially open-source). 

More people should build those!

💬25🔄73❤️339👀27276📊96