SDE -> Video creator pipeline
instantly recognisable that they got into tech in the hiring bull run and are burnt out in life with nothing but twitter likes helping them
8
217
the more india will grift, the less india will need to grift
straight up discouraging in the name of being pragmatic, bullish on indian talent killing it in deeptech, we shall realize our potential soon regardless of the obvious hurdles
last few days in mumbai dm me to catch up if you :
work in a data center/like football/like coffee/like walking/like art
5
20
1,081
happy to connect stranded gpu capacity with buyers
dm me with whatever type of node you want to offtake
1
3
309
Siddharth retweeted
We just executed the first OTC option on GPU rental prices with @Architect_Fi and @luxor 🤝
A put hedging ~$5M of our B200 rental revenue against AI compute price volatility.
Thank you @BrettHarrison, Matt Williams, and both teams!
covered this 2 months ago with the research piece on token derivatives
theres a lot of room for compute tail risk and lifecycle to be hedged and the risk premium that comes with being a counterparty is being priced slowly
great to see it being pioneered by @hyperbolic_labs / @Architect_Fi / @luxor / @silicondata
Announcement: Architect, @Luxor, and @hyperbolic_labs have executed the first OTC option on compute. Through our CFTC Introducing Broker we structured a bespoke $5M put option on B200 GPUs settling against @silicondata’s index. Luxor priced and was counterparty to Hyperbolic.
1
14
717
it begins
Announcement: Architect, @Luxor, and @hyperbolic_labs have executed the first OTC option on compute. Through our CFTC Introducing Broker we structured a bespoke $5M put option on B200 GPUs settling against @silicondata’s index. Luxor priced and was counterparty to Hyperbolic.
10
707
full forward and backward passing with an adapter
neat tech
Today we’re launching cross-model KV cache transfer from @OneTriangle, a step towards efficient model communication.
Up to 10.8x faster TTFT at 1M context. Prefill small model, transfer KV cache, decode on large. SDK live for Qwen3.5-4B → Qwen3.8-27B, beta open for more model pairs.
178
what sarvam’s announcement taught us
1. Dudes have no clue about capex for a frontier/specialized lab
2. Dudes have no clue about inference-first as a business model for a lab
oh wow the underlying infra was just more useful to rent out than attempt making something useful
known pivot pipeline now for "specialized labs" but I believe inference has a lot of potential at the optimisation layer
Today, we are launching Sarvam Token Factory for everyone.
We host our own models, and the full range of best-in-class open-weight models, in India. We match leading inference providers on performance and reliability.
For enterprises, this means more models to choose from, and an AI stack they can own without lock-ins.
2
12
1,223
super cool
Late-night talks about inference, GPUs and everything SGLang 🍸
Thank you to everyone who came to the SGLang Summit Afterparty, co-hosted with @hyperbolic_labs and SourceM3d! It was great catching up with old friends, meeting new ones, and hearing what everyone is building, from inference platforms to GPU clouds and compute marketplaces.
Big congrats again to @radixark on the $100M seed. The SGLang community keeps growing.
Where should we host the next one?
5
393
never feel bad about yourself
there are people out there with ‘getting drunk in a crowded place’ as a personality
gpu lifecycles is the biggest open ended question going forward
whenever someone mentions trading compute - “who is long compute” is the most obvious thought
I think longer lifecycles can introduce long counterparties that can at the least get a risk premium ?
fun times
V100 came out in 2017.
It’s still renting at ~ $0.50/hour in 2026.
Thank you @MichaelDell for citing our data.
1
8
552


















