?? Compute as a Bond ??

?? Compute as a Bond ??

India’s UPI moment in generative AI has finally arrived, but deploying it at scale requires the right infrastructure. Just as Microsoft revamped Azure for OpenAI, India’s DPI needs a transformation too. However, it’s easier said than done, as the infrastructure must support the needs of billions of citizens.?

The Indian government recently struck a deal with NVIDIA to procure 10,000 GPUs and offer them to local startups, researchers, academic institutions, and other users at a subsidised rate under its INR 10,000 crore India AI Mission.?

Former chief architect of Aadhaar Pramod Varma described it as quite ‘tricky’ to evaluate which startups will receive how many GPUs.?

"Besides procuring for their own purposes, the Indian government can potentially create a financial product, like a compute bond, where startups and VCs can buy vouchers by investing into it,” added Varma in an exclusive interview with AIM.?

He further said that this bond would allow the government to pool the demand for computational resources from multiple startups and VCs, giving the government an upper hand.

“The government can say if so many investments have come through this bond, I have a collective negotiation power, instead of each startup/VC trying to negotiate [independently] for their startup,” said Varma.?

Coincidentally, Varma’s idea echoes OpenAI chief Sam Altman and NVIDIA CEO Jensen Huang's perspectives, where they envision compute as the new currency unlocking a $100 trillion industry that can be transformed with AI.?

Altman recently said in an interview with Lex Fridman, “Compute is going to be the currency of the future. It may become the most valuable commodity in the world. We should invest significantly in expanding compute resources.”

In a similar vein, Altman proposed a concept where everyone would have access to a portion of GPT-7’s computing resources. “I wonder if the future looks something more like ‘universal basic compute’ than universal basic income, where everyone receives a slice of GPT-7 compute,” Altman speculated.

Towards Open Cloud Compute

Recently, People+AI launched Open Cloud Compute (OCC), a project that seeks to create an open network of compute resources, making it easier for businesses, especially startups, to access the compute power they need without being locked into specific cloud providers.

“Our idea is to have many, many micro data centres collectively behaving like a mega data centre. That is our network model,” said Varma, drawing an analogy that one bee doesn't matter, but a swarm of bees is suddenly very powerful.

Several compute providers are emerging in India, including Yotta, E2E Networks, Johnaic, Ola Krutrim, and Jarvislabs. Although these providers are not as large as hyperscalers like Microsoft Azure, AWS, and Google Cloud, they have the potential to challenge them collectively.

Yotta, backed by the Hiranandani group, is set to establish a GPU infrastructure with 32,768 GPUs by the end of 2025. Meanwhile, NeevCloud plans to acquire 40,000 GPUs by 2026. Jarvislabs has also secured access to thousands of NVIDIA H100s.

“One of the challenges these smaller players are facing is discoverability. Besides helping them with marketing, OCC also aims to support them in figuring out the right tax subsidies, infrastructural policies and ease of doing business,” Tanvi Lall, the director of strategy at People+AI, told AIM.

At the same time, India's data centres' rapid growth is currently being fueled by coal-generated power, threatening sustainability goals and exacerbating the nation's water crisis. Read more here.?

Democratising compute?

Varma said that OCC allows lower capital businesses and SMEs to come in and play the larger game. “Today, you can perfectly imagine a 105,000 square feet data centre suddenly becoming 500,000 square feet of compute.”

To date, OCC has already teamed up with 24 technology partners, including Oracle Cloud, Vigyan Labs, Protean Cloud, Dell, NeevCloud, and Tata Communications, among others.

He said the concept is similar to how Microsoft Azure or AWS works. They could have ten or more data centres in the country, but users are only concerned about the APIs they provide.?

Any user can simply log into the Open Cloud Compute Network, and upon typing their requirements in the search box, they will be shown multiple providers. Each provider card displays the location, usage cost, and a green rating of the provider. It also includes the SLAs promised.

“The idea is to create an open network of providers coming together powered by protocols rather than a platform aggregating everything. They can remain decentralised, but they can be discovered, transacted, or contracted through a protocol, and that's a very, very interesting way, as big as the internet,” added Varma.

OCC aligns with national initiatives like the India AI Mission, which aims to build a scalable AI computing infrastructure by deploying thousands of GPUs through public-private collaborations.

According to Lall, hyperscalers like AWS, Azure and Google Cloud could even be part of the network. “ Our entire approach is plus one. The idea is not to leave certain folks out. There is room for everybody and that is the reason we approached Oracle Cloud.”?

“If the hyperscalers feel they have something to offer in the Indian context, they are welcome. It would also be foolish for us to imagine that everyone will abandon the big CSPs as and when the network is up and running,” said Lall.?

Check out the full story here.?


Vector Databases are Ridiculously Good

Building large language models requires complicated data structures and computations, which conventional databases are not designed to handle. Consequently, the importance of vector databases has surged since the onset of the generative AI race.?

This sentiment was reflected in a recent discussion when software and machine learning engineer Santiago Valdarrama said, “You can’t work in AI today without bumping with a vector database. They are everywhere!”

He further added that vector databases, with their ability to store floating-point arrays and be searched using a similarity function, offer a practical and efficient solution for AI applications.

Enjoy the full story here.?


INDIA?

要查看或添加评论,请登录

社区洞察

其他会员也浏览了