Ajay (@ajaysagar99) profile picture

Proud Indian 🇮🇳

India
Joined December 2011
Photo by Ajay (@ajaysagar99)Photo by Ajay (@ajaysagar99)Photo by Ajay (@ajaysagar99)Photo by Ajay (@ajaysagar99)Photo by Ajay (@ajaysagar99)Photo by Ajay (@ajaysagar99)
Mahatma Gandhi promoted khadi, he wore it himself. When he made charkha a symbol of India’s self-reliance, he sat and spun it himself. He lived it, breathed it. An article to take us back in time -indianexpress.com/article/re…
2
886
I am back at @GoogleDeepMind as Research Engineer. Couldn't stay away for long. Feels good to be in London and diving back into research with Neon :)
46
13
718
21,368
Ajay retweeted
i was in a meeting with an investor and he said "there are so many inference providers in the market" i said, how many? so i went and scraped every catalog i could find -- models dev, OpenRouter, LiteLLM, Hugging Face, etc.. 425 inference API providers in the world right now what i found: - 58 got listed on models dev in all of 2025 - 232 so far in 2026 - 34 in august alone - 94 in november, and the month isn't over - 32 are routers or resellers by name i'm pretty sure 99% of them took vLLM, put an API key in front of it, and shipped the only thing that they can compete with is the is the price, and a price war on rented compute loses more money the more you scale this is not sustainable from all the views, a lot of them will shut down, their compute will come back to the market, and GPU prices will drop for whoever is left @rightnow_co started in march 2025 as a kernel optimization lab. we were writing GPU kernels long before we sold a single token we can break even on GPU cost at 20% utilization. even at $13.5 per B300 GPU hour we don't lose money that margin comes from the kernels. we're confident ours beat anything on the market, and we'll benchmark against anyone to prove it rightnowai.co
32
23
3
213
24,133
Ajay retweeted
Replying to @rivuchakraborty
aur space mein data centre banane waale the usme bhi kya Chinese model lagayenge.. abhi bhi chu banana baaki hai ki ya wo bhi deepseek se hi guidance lenge
1
1
18
Ajay retweeted
Replying to @rivuchakraborty
Sarvam mein kaam karte ho bhagwan nahin ho dkbose. Jawo jaake apna chacha logon ko bolna pichwade mein dam nahin that kyun market kiye sovereign model bolke aur kitne chote startupps ka gala ghot diya GPU access mein. koyee baith ke code check kar raha hai kahan kya train ho raha
1
1
22
Replying to @rishdotuk
Subsidy was paid for by govt i.e. public money. 1) what was the output of the money 2) if subsidy for GPUs was not sufficient for creating top class frontier models & progressing further does it mean another Galgotia doggy? The Galgotiya doggy aptly describes Indian AI efforts.
1
1
28
Arrogance of entitlement, when questioned on usage of public money. Its what Sarvam's PR team is doing. There r Indians using their own money & time & creating whatever little break throughs they can on AI. Even publish research papers. Sarvam's mandate was not to b a dhando.
A junior hardworking engineer contributes on open source, using their own personal claude account. Le some random non techy on twitter - Bhokhne chala aya. These guys can't do any good, but one step ahead to demotivate anyone. I don't respond to trolls in Twitter generally, but can't stand them doing this to a kid.
3
6
17
1,348
Replying to @business
bro raised money to only use deepseek ai models😂😂😂
1
2
1,197
Replying to @business
Shame shame @SarvamAI thoo 🤡... Traitors getting govt subsidy which are Indian tax payer money.. But giving access to Chinese AI models where claimed Arunachal as their land.
1
2
15
2,467
Replying to @business
Such embarassing. Instead of keeping the excess gpu to themselves and give better pricing of their models they sell chinese models. India is indeed a sh1thole.
1
2
63
6,237
Replying to @business
Its the same stupid story everywhere. Indian entrepreneurs pursuing short term profits instead of long term vision.
1
1
18
2,500
Replying to @business
It is called dalali One motabhai somewhere on top branch is architect & purveyor of this new format MakeInIndia Yojana soon to be renamed Dalali-India IP is for copy-paste Act Contract Act paper to eat fafda wt Laddu (wearing pants nowadays) Paisa pheko, tamasha dekho !
1
2
452
Again same old bullsht. And a loop can't show a good model on half the compute. If GPUs are being underutilised give it to acadmia. Atleast while we get upto date . Our scientist and researcher can use it but these people want to sell it overseas.
1
1
38
🚨 SHOCKING DISCLOSURE BY META ITSELF! 54 LAKH posts and accounts restricted in India in just 6 months on BJP orders. That's nearly 30,000 EVERY SINGLE DAY! Not our allegation. META's own figures, reported by Reuters. And BJP is upset that we protested at META offices? 😂 If not META's doorstep, WHERE should we go? Sit down clowns 🤡
𝐎𝐧𝐥𝐲 𝐨𝐧𝐞 𝐭𝐫𝐞𝐚𝐭𝐦𝐞𝐧𝐭. 💀 #IYKYK
37
494
14
1,499
39,541
We told you Aadhaar has 13ft high walls
The claim being circulated is false and misleading. The incident being referred to relates to a claim reported in 2023 and it has no connection with UIDAI or its Aadhaar database. UIDAI has not suffered any breach of its Central Identities Data Repository (CIDR). The Union Government had informed the parliament about the claim and explained its position in December 2023 itself. People are advised not to believe or share such misleading information.
12
81
1,759
Breaking🚨🚨🚨 As the next business move of Sarvam AI, they will server nigerian and somalian ai models in their model factory. They will use govt. subsidized GPUs to run somalian and nigerian llms and sell it. What an innovative company, I hope they offer afghani and monkey models soon.
7
5
1
68
2,695
Replying to @cneuralnetwork
I think what people are is saying to sarvan is to just do better
1
1
66
Ajay retweeted
Replying to @cneuralnetwork
Yc takes risks and gives money. The government takes taxpayers' money and decides to distribute it to its favorites that's the problem. Why not open the market for everyone and reduce taxes and duties to zero? Why not have a level playing field for everyone rather than handpicking someone?
1
2
115
Ajay retweeted
Replying to @cneuralnetwork
How is this a defence of Sarvam? Why waste taxpayer money on GPUs if it is either insufficient or inconsequential? Also don't see how compute cannot be enough, it may take longer to train a model but will not make it impossible right?
1
1
2
109
Replying to @ajaysagar99
I do not know of any Series A company that raises $28.5 and then allows a no-name investor named Avaanti advisor to invest $228K while raising valuation from $111M to $196M. Why was this done as soon as Sarvam got access to GPU from Indian government? How did Vrijesh Natural Fiber and Fabrics ended up getting allocation in HCL led round that brought the company to $1.5B. Who are they? What did Sarvam team allow them to invest this tiny amount? cc: @protosphinx
1
1
40
Ajay retweeted
Replying to @cvkrishnan
Are the cash flow statements public that you make quite confident claims about how they are utilising them My question remains - 4k size cluster was given, where is our 600billion + model?
1
1
57