- Embedding VC Newsletter
- Posts
- Embedding VC Newsletter - June 2025
Embedding VC Newsletter - June 2025
Hi friend,
June flew by in the best way — sunny days, inspiring convos, and a steady rhythm of building and learning 🧱. As we turn the page to July, we’re excited to share some of the progress, perspective, and momentum we’ve seen in the last month as we close out the first half of 2025 🔍.
Welcome to the June edition of the Embedding VC Newsletter, where we provide insights into the latest developments in our portfolio and the AI ecosystem.
👋 Refresh about Embedding VC:
If it is the first time you received our newsletter. As a fresher, we invest $100K to $1M checks into early-stage generative AI startups. If you or your founder friends are building something exciting, feel free to reach out to us at [email protected].
✨ Insight
Escalation in the AI Talent Arms Race
In this past month, Meta made headlines with unprecedented moves to acquire AI talent at almost any cost. Its most dramatic play was a $14.3 billion investment for a 49% stake in Scale AI, a data-labeling startup. It was an unusual deal aimed less at Scale’s business and more at securing its CEO, Alexandr Wang, as Meta’s new AI leader. This partial acquisition, Meta’s largest since WhatsApp, valued Scale at $29 billion. In other words, Meta essentially spent billions in a modern “acquihire” to jump-start an internal “superintelligence” lab under Wang’s leadership.
The deal structure was intentional: by taking a non-controlling stake, Meta avoided the months-long regulatory scrutiny that would come with a full acquisition (Wang and a few key Scale executives started working immediately). Scale's prior relationships with major enterprise customers gave it rich insights into model performance and infrastructure, but they’re also losing customers due to this tie with Meta, raising questions about future revenue. This means Meta might be paying a premium for a company whose valuation could soon decline. Still, the message was clear: in today’s AI arms race, buying the team (even at peak pricing) is sometimes faster than building one.
Meta didn’t stop there. The company had reportedly approached multiple top AI startups as acquisition targets throughout the first half of 2025. Targets included Runway and Perplexity among others. Most notably, Meta tried to buy Safe Superintelligence (SSI), a secretive one-year-old AI venture co-founded by former OpenAI scientist Ilya Sutskever, at a whopping $32 billion valuation. SSI had fewer than 30 employees and no public product, yet had raised $2 billion by April at that lofty $32 billion valuation on the strength of its elite team and mission (building a safe, “aligned” superintelligent AI). Sutskever rebuffed Meta’s acquisition offer, preferring to keep SSI independent.
Thwarted in buying certain companies outright, Meta shifted to poaching individuals. In late June, Meta successfully hired SSI’s co-founder Daniel Gross (along with former GitHub CEO Nat Friedman), effectively bringing in SSI’s CEO even after failing to buy his company. Sutskever has just announced that he’ll be the new CEO of SSI. Gross and Friedman’s venture fund will also partially come under Meta, and both will now work under Alexandr Wang on Meta’s superintelligence project. Mark Zuckerberg also personally reached out to AI researchers at competitors like OpenAI and Google DeepMind with staggering compensation packages, i.e. signing bonuses up to $100 million, plus massive pay packages.

Up to this point, Meta has brought on 11 new hires in the AI field, including researchers from OpenAI, Anthropic and Google.

At the end of the day, the scarcest piece of the AI puzzle seems to be people, not compute or data. You can not acquire OpenAI for $400B, yet with $10-20B, someone may acquire most of OpenAI's core talent — if superintelligence is the goal.
Meta’s aggressive moves in June highlight just how scarce and valuable top AI minds have become. As competition heats up, we believe it’s becoming increasingly important to map AI talent early, especially those driven by their own mission. Spotting them before everyone else does might just be the next frontier of competitive advantage.
🗞️ Vectors In The News
Updates from Embedding VC portfolio companies:
The leading generative AI platform in education
New Features: MagicSchool just launched new accessibility features in Student Rooms, including text-to-speech, translation, readable fonts, adjustable text size, and more, to ensure every learner has the support they need.
Featured in Inc.: MagicSchool was recognized in Inc. Magazine for its impactful work supporting educators with AI. A proud moment spotlighting their mission to empower every classroom.
Cursor for creatives
More Updates!
Kling 2.1 Master is live, delivering cinema-grade video generation with expressive motion, real physics, and 1080p output in under 60 seconds
FLUX.1 Kontext for in-context image generation is live, preserving character consistency, enabling local edits, and maintaining visual style across scenes
Veo 3 by Google DeepMind is also now live, providing 4K cinematic video with sound, physics-aware motion, and unmatched prompt precision (FLORA is the first to ship it!)
The world's first open-source AI wearable device
Privacy-First Update: Omi AI is now the world’s most private AI wearable! With newly released end-to-end encryption, users generate a unique key to fully secure their data, and only you can decrypt your conversations.
Omi Glass is Here: Omi launches AI smart glasses that see, hear, and think alongside you, offering full-day recording, real-time answers, and brain-controlled AI. From counting calories to finding lost items, it’s a bold leap toward becoming 1000x smarter with AI.
First AI “software engineering mentor” to support engineers throughout the entire software lifecycle
New Extension: Entelligence.AI now supports VS Code, Cursor, and Windsurf! Review code and get instant feedback directly in your IDE before merging. It’s built for seamless, in-the-flow coding.
Cursor helps you move fast. Entelligence makes sure what you ship actually scales.
Fastest Growing Launch: Entelligence.AI in IDE code reviews is #2 on product hunt right now just above Perplexity Tasks!
💡 New Embedding Vectors
Embedding VC joins Anvil’s (YC X25) pre-seed round
AI-native interfaces like ChatGPT, Gemini, and Claude are rapidly changing how people discover information. Vercel, which powers millions of modern websites, now sees up to 40% of its traffic coming from GPT-powered sources. In this new era of Generative Engine Optimization (GEO), Anvil is building the platform to help companies monitor and improve their reference rate across leading models.
Founded by Daniel Siryakov (former AI PM and MLE) and Barak Ben Noon (ex-Google and SEO veteran), Anvil delivers clear visibility into brand presence, competitor benchmarks, and actionable steps to win in AI search. We're excited to back the Anvil team as they bring GEO for the LLM era.
🎤 Builder Spotlight
Dhravya Shah, Founder and CEO of Supermemory
Meet the Founder: Originally from India, Dhravya Shah was granted an O-1 visa (visa for individuals with an extraordinary ability in the sciences, education, business, or athletics) at just 19 years old. He dropped out of university to dedicate himself fully to building the future of AI memory infrastructure. A full-stack developer with a builder’s spirit, he has created tools that have served hundreds of thousands, including high-scale web apps and bots with millions of impressions. Now, with Supermemory, Dhravya is on a mission to solve one of the most fundamental challenges of AGI: memory. As he puts it, “If there’s ever a time in human history to build, it is right now.” We couldn’t agree more.
Founder Fireside Chat - Q&A with Dhravya
Q: What inspired you to start Supermemory? | A: When I first started supermemory, it was called AnyContext - back then, there were no ‘second brain’ or knowledge garden tools out there that utilise AI properly, so I set out to build my own. Quickly, I realised the infrastructure challenges related to giving ‘memory’ to AI agents, in my case it was telling AI about what the user liked and disliked, and offering a natural search engine over the knowledge garden. I became a consultant to many companies because of this (especially because of the fact that anycontext, which was renamed to supermemory, was open source). Being a consultant, I realised that the problem is so much deeper and complex, and that many companies are not doing the right things when building memory systems in their applications and agents. So, I set out to build supermemory’s memory engine, which was a deep research and infrastructure problem, and productized it. |
Q: What are you most excited about for the next three years? | A: I’m excited about extremely personal LLMs, and how that changes how we interact with them over time. I’m also extremely excited about how robotics and robots walking around will have and manage their memories (can they have a negative impression of someone?) and spaceflight. I’ve been interested in spaceflight for a decade now and I believe that there’s a lot of inflection points to come - AGI, Robotics, Genomics, Energy but the main big thing will be when life is multi-planetary. |
Q. What have you enjoyed about working with Roger and Embedding VC? | A: Roger is extremely thoughtful, about everything from what to do and what not to do. This is a different and unique experience than other VCs because everyone tells you exactly what to do but Roger understands the founder’s shoes and trusts with the decisions that the founder makes, but gives good pieces of advice on what has worked in the past. It has been a pleasure to work with him so far, even though it’s not been very long. |
👋 Happy 4th! 🎆🦅
See you soon! We are Embedding VC.
Embedding VC invests in early-stage Generative AI startups.