AMAN VAIBHAV

Ex-AI founder doing strategy & GTM @ Xenveo. Strategy, Product, Startups and VC in past lives. Physical AI takes. @iitmadras

[AMAN VAIBHAV @amanvaibhav]

A $1.45B bet that robots need to feel, not just see. Eka's vision-force-action model picks 1,500 objects an hour at 95%+ success, trained in simulation. My take on why touch is the new moat in physical AI: https://t.co/rh2GRrOPcm

[AMAN VAIBHAV @amanvaibhav]

https://t.co/osNJ5MN5Pz

[AMAN VAIBHAV @amanvaibhav]

OpenAI reportedly plans to buy Google's TPUs, per The Information via Reuters. The fiercest AI rivals are now each other's suppliers. No one wins the compute war on a single chip vendor, not even the company that made Nvidia essential. Diversification is the strategy.

[AMAN VAIBHAV @amanvaibhav]

Physical Intelligence is reportedly raising $1B+ at ~$20B, nearly double its $11.5B valuation earlier this year, with no revenue yet. The market is not buying robots. It is buying one generalist model to run every machine. That is the physical AI thesis in one deal.

[AMAN VAIBHAV @amanvaibhav]

@vaibhav514g Congrats on the $7m ARR jump and pausing the raise from strength. If Saturn Labs needs more collection capacity as demand grows, Xenveo runs field teams across the US, Canada, LatAm, Europe, the Middle East, SEA and India for egocentric video, multimodal data and QA.

[AMAN VAIBHAV @amanvaibhav]

@Ed__Johns @creandum @usv Congrats on Machenta and the $40m seed. Instant learning from one demo is a real step. When those demos meet real floors, Xenveo can supply real world demonstration data with QA from field teams, so it holds up outside simulation.

[AMAN VAIBHAV @amanvaibhav]

700 muscles is a headline, not a moat. Learning 1.5 hours of motion in 4 hours on one GPU is the real shift: iteration speed is now the bottleneck, not compute. Whoever iterates fastest on a complex body wins the humanoid race.

[Qiayuan Liao @qiayuanliao]

SuperMimic on musculoskeletal humanoid, a much more complex embodiment!

[AMAN VAIBHAV @amanvaibhav]

Instant learning is only a breakthrough if the one demo survives contact with a real floor. Simulation trained, zero real pixels is bold. The $40m bet is that instant beats accurate, and deployment will prove which one customers pay for.

[Edward Johns @Ed__Johns]

Today we’re out of stealth, and we’ve raised a $40m seed round co-led by @creandum and @usv, to bring Instant Robot Learning to life.

For the past few months, we‘ve been busy building a new company.

Machenta.

This follows results from a new kind of robot foundation model we’ve built, which I’m so excited to share with you today!

Usually, it takes hours/days of data collection and training to teach a robot.

We’ve built a model that learns tasks instantly: one demonstration, no further training.

A model trained entirely in simulation, without a single pixel of real-world data.

A model where teaching robots is so quick and easy, it feels like physical vibe coding.

A model for robots that learn like humans do, from someone like you.

Welcome to the era of Instant Robot Learning.

Thank you @Kathi_Wilhelm, @rebeccakaden, @hkanji, @jonno_evans and all of our angels for your support!

@Machenta_

Learn more: https://t.co/P1woKrXtjH

See below for what you can do with Instant Robot Learning!

[AMAN VAIBHAV @amanvaibhav]

@pollenrobotics A game is a better benchmark than a demo. Tug of war forces strength, balance, and recovery in one test, and a public cup turns that test into a leaderboard anyone can beat.

[AMAN VAIBHAV @amanvaibhav]

@ClaraChengGo Sentiment is not a moat killer. If 10 researchers paused on a launch headline, the opportunity is for the team that keeps building through the cycle. Astra raises the bar on demos, not on deployed reliability.

[AMAN VAIBHAV @amanvaibhav]

@rahim_mirani If general models keep doubling every 54 days, the robot-specific data moat shrinks to the hard 10 percent. The bowl test shows why: partial progress is not a product, and the last grip is where the value lives.

[AMAN VAIBHAV @amanvaibhav]

@Sanjil A $399 biped and a $500 humanoid change the talent funnel. When hardware stops being the excuse, software, data, and reliability decide who survives. Price is now a strategy, not a spec.

[AMAN VAIBHAV @amanvaibhav]

@NathanKau Teleoperation at 5x slower and $100 plus per hour was never a scaling strategy. It was a tax on data. Whoever cuts that tax without losing quality owns the training pipeline.

[AMAN VAIBHAV @amanvaibhav]

Construction may finally be getting its factory moment. 1,250+ robot-built homes, a two-story structure in about a day vs two to three weeks on site, 5,000 tactile data points per home. The homes are the product. The data is the moat.

[AMAN VAIBHAV @amanvaibhav]

The money has arrived. Every one of those 521 deals now faces the same question: where does the robot learn the world?

$47B buys robots and compute. It doesn’t buy experience.

That still has to be collected.

[AMAN VAIBHAV @amanvaibhav]

Physical AI startups raised $47.4B in the first half of 2026 (Crunchbase).

That’s more than 2022, 2023 and 2024 combined.

The single biggest check: $16B into Waymo, nearly a third of the total.

[AMAN VAIBHAV @amanvaibhav]

This week, NVIDIA, the company that sells simulation to the whole industry, released a dataset of real human activity: three camera views, 3D hand tracking.

Even the simulation leader is shipping real-world data.

[AMAN VAIBHAV @amanvaibhav]

@BerntBornich Scaling NEO into homes means operator-led collection has to scale with it. Xenveo runs trained field teams for robot data collection and annotation with full QA, so your operators can stay focused on the hardest tasks while volume grows.

[AMAN VAIBHAV @amanvaibhav]

@sankaet @Scobleizer @arian_ghashghai Complex skills needing longer, richer data is exactly where deployment stalls. Xenveo runs field teams that collect egocentric and teleop demonstration data across factory and logistics sites, annotated and QA checked, so pilots can scale past the demo stage.

[AMAN VAIBHAV @amanvaibhav]

https://t.co/zZa0NHI6SV

[AMAN VAIBHAV @amanvaibhav]

Banks just lined up $63B to buy Broadcom chips and lease them to Anthropic, with Broadcom guaranteeing the chips' value. The chipmaker is now backstopping debt for its own product. AI compute is no longer bought, it is financed in a circle.