What’s 🔥 in AI/Infra/VC #507
Don't Mistake Investor Interest for Company Readiness
Made a small but important change to the newsletter. When I started writing What’s 🔥 over 7 years ago, AI was barely part of the conversation. Today, it’s at the center of almost everything I spend my time thinking about.
So to better reflect that, What’s 🔥 in Enterprise IT/VC is now What’s 🔥 in AI/Infra/VC. Thanks for sticking with me.
On a different note, something has been bothering me lately. I’m seeing too many founders mistake investor interest for company readiness. Those aren’t the same thing.
So here’s my advice. Founders: Avoid the "Forever" Fundraising Round
I'm writing this because I'm seeing it happen far too often. Great founders get pulled into fundraising before they're ready, and it becomes one of the biggest distractions a young company can face.
As always, 🙏🏼 for reading. Please share with your friends and colleagues!
Scaling Startups
#🎯
#so true
#for those neo labs
#💯
#🤔
Enterprise Tech
#🤯 Kimi - “open frontier intelligence” sums it up - huge new release
#regarding China - this is a must read!
- reaffirmed commitment to open source to promote AI “openness and win-win”
- warns against “over stretching” the concept of national security as applied to AI where one country’s national security is prioritised over others
- China opposes emergence of “new historical injustices” in AI (one of the most strongly worded parts of the speech)
#this is a huge deal from Mira’s Thinking Machines - the best US open weight model, not as good as Kimi but better than Nemotron - need more of this!
#hosting open weight models a huge part of this 📈
#with the 💰 being spent on data centers and AI infra, we need to rethink what security looks like - check out this report from Lava
#Satya continues to deliver on his messaging for control and enterprise learning loops - must read
#must read memo from GLM 5.2 founder found…believes AGI coming in 2 years???
#turns out kubernetes is still best software to build, manage, orchestrate and scale AI workloads - congrats Spectro Cloud
#RL matters just as much if not more than raw compute
To build their latest model, Meta executed an emergency pivot, pulling thousands of employees off their normal duties to generate expert traces, which are hand-guided coding examples required to post-train the model. This confirms that the bottleneck for frontier models is not just raw compute, but high-fidelity, human-verified reinforcement data.
which is why so many companies focused on data and RL
#speaking of the 99th percentile Series A at $1B, this is one example - Super Intelligence just raised a $130M Series A - in under a year, demand has scaled to over $100M in annualized revenue, alongside a large open-source community of 500+ environment creators and 100k+ downloads.
Pre-training concentrated the frontier of AI inside a handful of closed labs.
RL breaks that open: teams can now own their model-optimization loop — train directly on their product, optimize for their workflows, and ship agents that continuously learn in production.
Own that loop and you build a compounding moat in the agentic era. The only missing piece was the infrastructure.So we built the Open Superintelligence Stack — training, inference, and compute:
› Training — hosted training, prime-rl (large-scale agentic RL), environments, sandboxes, evals
› Inference — dedicated, serverless serving and continual learning
› Compute — frontier liquid clusters
We train our own open frontier models on it, and ship the same stack to our customers
#competition is healthy for users
#Welp
Markets
#IBM - dollars going for more compute and memory, taking away from software but bonus for cybersecurity
#extreme concentration risk
#$1 Trillion of market value gone…
#just sharing - let’s do work on positive messaging so the fringe does not become more than just that
