←
Pantera Capital6,845 posts
Pantera Capital avatar

Pantera Capital

@PanteraCapital

Backing founders at the frontier of the digital economy since 2003. Subscribe to our monthly letter: panteracapital.com/subscribe……

◷ Joined January 2014▧ 1,108 Photos and videos
Pantera Capital avatar
Pantera Capital@PanteraCapital·Nov 5

Alex Blania's prediction: within three months it will be completely impossible to spot that you are talking to a deepfake, and impossible for any software to catch it. @alexblania (CEO & Co-Founder of @tfh_technology) joined @Cosmo Jiang on Stateful to talk about proof of human on an internet run by agents: - Why @worldnetwork is working with Zoom, Reddit, and the big banks - Video models just hit 60 FPS, which makes real-time deepfakes possible - More than 50% of Cloudflare's traffic is now agents - Why advertisers may end up being the biggest accelerant of all - Whether we already have AGI, and what two gigawatts a week of new compute does next

37:48
Ondo Finance avatar
⟳ Pantera Capital retweeted
Ondo Finance@Ondo·16h

Every major IPO will be tokenized on day one. Ondo’s @KatieAWheeler joined Bloomberg to discuss ANTon coming to Ondo Stocks on day one and what tokenization makes possible: “The big differences are global accessibility, flexibility, and the market infrastructure.” The conversation around the world’s biggest IPOs now includes tokenization.

0:46
Franklin Bi avatar
⟳ Pantera Capital retweeted
Franklin Bi@FranklinBi·18h

As John Pierpont Morgan once said (allegedly): "We're here to build a goddamn institution." Global FX is getting rebuilt thru stablecoins & AI agents and Codex is building the next global FX institution. Thrilled to lead @CodexFX's Series A at @PanteraCapital and join the board to partner with @haonan @momoeureka @victor_yaw_ in building the FX layer for the AI-native generation.

Accelerating improvements in model capabilities has made it possible to build a new kind of financial institution — made of agents, protocols, and stablecoins. FX is a $195T market stuck in the 1970s. There are three forces culminating now that make it possible to build a new kind of financial institution. - Technological: intelligence is commoditizing, opening products and cost structures that were previously impossible - Social / Political: the passage of key legislation like the genius act is legitimizing stablecoins as cash equivalents - Economic: Fundamentally stablecoins are more efficient than fiat and rates rising make stablecoin issuance more interesting and liquidity formation easier Every major technological shift has produced new financial institutions. The railroads produced JP Morgan. Stablecoins today are a niche market today numbering in the hundreds of billions. we are on track to dominate this niche. That alone is a good business but not that interesting. We are working on this because we are betting that stablecoins will continue to eat global commerce. If we are right, then we will have built generationally important financial infrastructure that will shape how global commerce works for the next century

Christian Pickett avatar
⟳ Pantera Capital retweeted
Christian Pickett@chrisspickett·20h

Today we're launching the Data Marketplace for Agents: 100+ providers that agents and developers can find and pay for per call. Google organized the world's public information, and agents now search it faster than any person could. But a lot of the valuable data isn't on the open web. Data on people, companies, jobs, finance and real estate is kept by data companies, and agents can't easily reach it. Most of it is still sold by subscription. That works for a product that calls one source thousands of times. An agent needs a little data from a lot of places, and it shouldn't have to buy three subscriptions to make two requests. Last November, @berasogut1 and I started @orthogonal_sh to make APIs discoverable and payable by agents. Agents mostly came to us for data, and soon data providers were asking to list with us. So we became a marketplace. Providers list once, and agents find what they need and pay per call from one balance, over MCP, CLI or API. About 90% of our usage already comes from customers using five or more providers, including @DustHQ, @ZenskarHQ and @foundeers.

1:40
Codex avatar
⟳ Pantera Capital retweeted
Codex@CodexFX·22h

Announcing our $30M Series A raise, led by @PanteraCapital, with participation from @Dragonfly, @CBVentures, and others. We're building the stablecoin FX layer for the next century of finance. Join us.

0:48
Momo Ong avatar
⟳ Pantera Capital retweeted
Momo Ong@momoeureka·22h

We raised $30 million at @CodexFX led by @PanteraCapital, with participation from @dragonfly, @cbventures, @recvcx, @Fin_Capital_VC, @christyhwchoi and others. Every major technological shift has produced a landmark financial institution: Railroads ▶️ J.P. Morgan Telegraph ▶️ Western Union Internet ▶️ PayPal Codex is born out of Stablecoins and AI. We are small, elite team that grew annualized transaction volume 10x to $25B this year. Join us! 👇

0:48
Cody Poh avatar
⟳ Pantera Capital retweeted
Cody Poh@0xhopydoc·Nov 15

CoRL 这周所有的产品发布,再加上各大厂研究员发的帖子,只能说句句不提 LLM,但句句都是 LLM。所有人都在回应同一件事:LLM 带来的冲击。 几个宏观的核心观察,先说最重要的: 1/ 这周的发布本质上都在做同一件事:把原本可以在一个模型里的 System 1 和 System 2 正式拆开。推理交给 LLM,这块 GPT-6 Astra 和 Claude Fable 5.2 已经证明它们干得挺好;低延迟的动作留给专门的机器人模型。 我的看法是快环慢环的架构可能永远都要有,但快环最终会被慢环做掉。大模型实验室一旦把数据闭环跑起来,自己蒸馏一个本地 System 1 是顺手的事。所以机器人实验室现在主动把自己的 System 1 接到 OpenAI 的 GPT-6 Astra 上、让它当 System 2,只是为了避免被 LLM 团灭,给自己续命而已。 世界动作模型们基本上已经放弃争 System 2 这一层了,只能在 System 1 上证明自己值得存在。于是比赛变成LLM大模型实验室能多快把 sim-to-real gap 缩掉、多快把整个 training loop 跑起来,从而正式下沉到 System 1。世界模型续命能续多久,就看哪边更快:大模型实验室拿到具身数据,还是机器人实验室补上推理能力。 最直观的信号就是 11 月 13 号 OpenAI 把独立的机器人团队解散了,官方说法是通用模型已经能直接控制机器人,不需要单独的架构。他们押的就是一个模型从 System 2 垂直打穿到 System 1。当然这只是押注,不是结果:OpenAI 现在手里并没有一个拿得出手的 System 1,实机演示里 Astra 也只是做高层决策,电机层面还是传统控制器在跑。但方向很明确。 2/ 这周最大的发布之一是 NVIDIA 的 GR00T 2,一个世界动作模型,预训练数据按时长算基本是靠人类视频训出来的。主题演讲里的整场演示都是拿 Astra 当 System 2 来指挥 GR00T 2,他的结论是 Astra 和 GR00T 谁都单独干不了。没错,但那是今天。等 System 2 的大模型实验室自己有了 System 1,这个前设就会被彻底推翻。 3/ Google DeepMind 同一周发了 Gemini Robotics 2.5,结构跟 NVIDIA + OpenAI 那套基本一样:Gemini Robotics ER 做具身推理管流程,Gemini Robotics 2.5 这个 VLA 管手和胳膊,在奔驰产线上让 Apptronik 的 Apollo 2 做零件分拣,96% 成功率。更有意思的是谷歌最前沿的模型 Gemini 4 Argon是被当做仿真引擎用,从一个真实场景生成几千条仿真练习、建 digital twin、看失败视频分析原因,拿来加速 Gemini Robotics 2.5 的训练。更狠的是,Google 直接把System 2、System 1 和训练闭环全在拿在手里。 退后一步看,Google 拿 Argon 当仿真引擎,NVIDIA 把 real-to-sim 流水线开源,打的都是开头说的 training loop 那一仗。只是落地的逻辑完全相反,Google 的闭源系统只给自己的生态伙伴,NVIDIA 那套在陆续开源。 4/ 为什么现在还得拆?因为 LLM 直接上机确实还不行。至少在 NVIDIA 自己的测试里,让 Astra 直接下沉到 System 1 代价不菲:一次尝试要烧几十美元的 token,每次调用十多秒,现在在工厂里根本没法直接用。但这是个会随时间解决的问题:要么把推理链蒸馏进小模型,要么直接用 RL 让模型吐 action token 而不是文字 token。成本和延迟是现在必须拆开的原因,但它不是护城河,只是时间差。 5/ 最后,我倾向于把 LLM 在机器人的统治地位看成一个数据问题:文本数据太好拿了。想象一下,如果抓视频数据像抓互联网文本一样容易会发生什么,更别说后面还有清洗、标注、后处理这一整套。具身数据难拿,曾经是机器人实验室唯一的护城河。而 NVIDIA 这次把视频数据、采集硬件、仿真流水线全开源,本质上是在把这道门槛砸掉。 注意 NVIDIA 不在续命的队伍里:它要的是 System 1 作为一层永远存在,Jetson、Isaac、GR00T 都在这一层,至于这一层最后是谁的模型,它不在乎,反正都在它的卡上训。真正在续命的,是那些靠私有 System 1 数据活着的实验室,它们的倒计时可能比想象的更快。谁先把视频到动作的数据闭环跑通,谁基本上就赢了;而 NVIDIA 刚刚把所有人的时钟都拨快了。

Sabrina Ortiz avatar
⟳ Pantera Capital retweeted
Sabrina Ortiz@sabrinaa_ortiz·Nov 13

Would you let an Orb scan your iris in exchange for a digital ID that proves you’re a real human, not a bot? As an AI reporter who’s covered countless stories about tech companies mishandling our data, I was skeptical. But I decided to do it anyway. I spent months digging into the technology, talking to the people behind it, and consulting independent experts to answer one big question: Can we actually trust World ID? My deep-dive article is live on @theDeepView, AND my FIRST-EVER LONG-FORM YOUTUBE VIDEO is finally out!! Watch below and find links to both in the replies 🔗

13:40
Franklin Bi avatar
⟳ Pantera Capital retweeted
Franklin Bi@FranklinBi·Nov 11

Incredible launch day for @UngodlyGame. Already #1 in the role-playing category! Congratulations to @markotero & the @AzraGames team on creating an incredibly beautiful & brutal world to empower & inspire a new generation of role-play! Honored to be part of this epic journey.

We got tired of waiting for God of War on mobile, so we made UnGodly. ⚔️ And it's so good we've forgotten about that other game. Download it now and you will too. #ungodly #LaunchWeek

1:21
Cody Poh avatar
⟳ Pantera Capital retweeted
Cody Poh@0xhopydoc·Nov 11

总结一下我们的核心观点: 我们认为纯机器人模型公司显然已经进入商品化的阶段,这也意味着纯模型实验室高估值已经到达了拐点。它们要么通过自己掌握硬件来垂直整合整个技术栈,要么彻底转向前线部署模式,把部署环节握在自己手里。 Fable 5.2 和 Astra 6 的出现,基本意味着通用 LLM 已经可以取代所谓的 system 2;而当这些模型用来写代码去控制机器时,还能继续下沉到 system 1,接管传统机器人栈里的动作规划环节。只要推理长度和所需算力能随时间持续优化,专注于 pre-training 的纯 VLA 实验室追逐通用机器人智能,最终会被彻底淘汰。 有意思的是,我们认为价值最终会沉淀到部署层。从软件的角度看,也就是市场所说的 system 0。而其中的关键在于:持续收集后训练数据,会实质性地提升软件对硬件的控制能力,也让部署本身变得更高效、更好。 至于硬件,我们的观点比较反直觉的是,它反而会成为瓶颈。虽然这一层的利润率大概率会被竞争压低,但这不等于硬件没有价值。公司必须自己掌握硬件,才能完全释放软件的能力。这意味着要自研、自产硬件,确保后训练所优化的一切,都能真正落到硬件层上。

Cosmo Jiang @cosmo_jiang Nov 10 Article Genchi Genbutsu: Hardware, Software and Everything In Between At 2 a.m., our Cathay Pacific flight finally touched down in Hong Kong after a six-hour delay in Singapore. With barely three hours of sleep, we found ourselves in a car heading straight to Shenzhen,

UnGodly avatar
⟳ Pantera Capital retweeted
UnGodly@UngodlyGame·Nov 11

We got tired of waiting for God of War on mobile, so we made UnGodly. ⚔️ And it's so good we've forgotten about that other game. Download it now and you will too. #ungodly #LaunchWeek

1:21
Cosmo Jiang avatar
⟳ Pantera Capital retweeted
Cosmo Jiang@cosmo_jiang·Nov 11

The stated purpose of my trip to Shenzhen was to meet robotics companies. The unstated one was to get a deeper read on China and the cultural mores behind its successes and shortcomings. Over drinks, a local entrepreneur whose family owns a small retail chain gave me a pretty good explanation of why it’s so hard to make money in China. “这里的竞争太大了,” he kept saying. The competition here is brutal. By the end of the conversation, we’d both laugh a little grimly whenever he said it. His point: there’s always someone willing to give up this year’s profits to win market share, figuring they’ll make it back next year. Then next year comes, only to find someone new has the same bright idea.

Cosmo Jiang @cosmo_jiang Nov 10 Article Genchi Genbutsu: Hardware, Software and Everything In Between At 2 a.m., our Cathay Pacific flight finally touched down in Hong Kong after a six-hour delay in Singapore. With barely three hours of sleep, we found ourselves in a car heading straight to Shenzhen,

World avatar
⟳ Pantera Capital retweeted
World@worldnetwork·Nov 11

“My prediction is that in the next three months, it will be completely impossible to spot that you’re talking to a deepfake.” - @alexblania, co-founder, World @world_id helps build trust by preventing deepfakes at the source.

0:22
Digital Asset Summit 2026 avatar
⟳ Pantera Capital retweeted
Digital Asset Summit 2026@blockworksDAS·Nov 11

Dan Morehead of Pantera Capital said even if you're megabullish on AI, you should have blockchain exposure. "Bitcoin has on average almost doubled every year for 13 years." @dan_pantera of @PanteraCapital

1:09