The Open Cloud Compute (OCC) network - a decentralised network of 10,000 micro data centres designed to provide affordable computing infrastructure for companies to meet growing AI demands - is yet to go live, eleven months after its launch.
According to multiple insiders, the network is yet to define its core use cases and determine whether it has the computing power to support AI workloads.
"We need to do more research and figure out what AI use cases can work with such a network," one of them said on condition of anonymity. "This is what we have been focusing on right now. We are not onboarding any new suppliers of compute."
Launched by AI non-profit based in Bengaluru people+ai, OCC was envisioned as an interconnected cloud computing infrastructure, where independent compute providers with different GPUs could collaborate on a single platform, enhancing their discoverability and utilisation by customers.
At its launch, 24 companies, including US semiconductor firm Advanced Micro Devices ( AMD), Oracle, contract manufacturer Dixon Technologies, Tata Communications, E2E Networks, IBM, and Dell Technologies had signed letters of intent to join OCC.
Some experts, however, said the launch of the IndiaAI Compute Portal with over 14,000 GPUs last month and competition from hyperscalers may have made OCC unviable.
A key challenge for the network of distributed micro data centres is ensuring necessary computing power to support AI workloads effectively.
For individual micro data centres to buy GPUs requires huge investment that they may not have, said Narendra Sen, founding chief executive of NeevCloud, one of the consortium partners.
Also, the central government's IndiaAI Mission has addressed the issue of compute supply and demand by launching the IndiaAI Compute Portal on March 6, he said. The platform will offer over 14,000 GPUs with subsidised hourly pricing. "I think OCC would be replicating the same. That is why they may not want to continue," Sen told ET.
Vishnu Subramanian, founding chief executive of GPU cloud platform Jarvis Labs, which undertook an OCC pilot in PES University in October last year, said he has not heard from the network for some time.
"We are part of the OCC consortium. Had the network gone live we would have been one of their partners offering GPUs to end users," Subramanian told ET. "We haven't heard from them for quite a long time, so it's quite possible they are not going live." people+ai, AMD, Oracle, and Dell Technologies declined to comment.
Dixon Technologies, Tata Communications, E2E Networks, IBM, and other consortium partners like Vigyanlabs did not respond to ET's request for a comment as of press time Friday.
“Distributed computing has always been a great idea until it is implemented and that is why it has not been successful anywhere,” Neil Shah, partner at Counterpoint Research, and a technology industry analyst, told ET. “Hyperscalers have always won in this battle,” he added.
According to Shah, there are two levers: cost and scale.
The bigger players have scale; they can reduce the cost and price compute cost more aggressively, he said. “Whereas microdata centres and distributed workloads make sense when it is around a particular geographical area and latency is important.”
From a scale perspective, the model will not work in a country like India because hyperscalers will always outprice them, Shah said.
people+ai is an offshoot of Infosys cofounder Nandan Nilekani’s EkStep Foundation.
According to multiple insiders, the network is yet to define its core use cases and determine whether it has the computing power to support AI workloads.
"We need to do more research and figure out what AI use cases can work with such a network," one of them said on condition of anonymity. "This is what we have been focusing on right now. We are not onboarding any new suppliers of compute."
Launched by AI non-profit based in Bengaluru people+ai, OCC was envisioned as an interconnected cloud computing infrastructure, where independent compute providers with different GPUs could collaborate on a single platform, enhancing their discoverability and utilisation by customers.
At its launch, 24 companies, including US semiconductor firm Advanced Micro Devices ( AMD), Oracle, contract manufacturer Dixon Technologies, Tata Communications, E2E Networks, IBM, and Dell Technologies had signed letters of intent to join OCC.
Some experts, however, said the launch of the IndiaAI Compute Portal with over 14,000 GPUs last month and competition from hyperscalers may have made OCC unviable.
A key challenge for the network of distributed micro data centres is ensuring necessary computing power to support AI workloads effectively.
For individual micro data centres to buy GPUs requires huge investment that they may not have, said Narendra Sen, founding chief executive of NeevCloud, one of the consortium partners.
Also, the central government's IndiaAI Mission has addressed the issue of compute supply and demand by launching the IndiaAI Compute Portal on March 6, he said. The platform will offer over 14,000 GPUs with subsidised hourly pricing. "I think OCC would be replicating the same. That is why they may not want to continue," Sen told ET.
Vishnu Subramanian, founding chief executive of GPU cloud platform Jarvis Labs, which undertook an OCC pilot in PES University in October last year, said he has not heard from the network for some time.
"We are part of the OCC consortium. Had the network gone live we would have been one of their partners offering GPUs to end users," Subramanian told ET. "We haven't heard from them for quite a long time, so it's quite possible they are not going live." people+ai, AMD, Oracle, and Dell Technologies declined to comment.
Dixon Technologies, Tata Communications, E2E Networks, IBM, and other consortium partners like Vigyanlabs did not respond to ET's request for a comment as of press time Friday.
“Distributed computing has always been a great idea until it is implemented and that is why it has not been successful anywhere,” Neil Shah, partner at Counterpoint Research, and a technology industry analyst, told ET. “Hyperscalers have always won in this battle,” he added.
According to Shah, there are two levers: cost and scale.
The bigger players have scale; they can reduce the cost and price compute cost more aggressively, he said. “Whereas microdata centres and distributed workloads make sense when it is around a particular geographical area and latency is important.”
From a scale perspective, the model will not work in a country like India because hyperscalers will always outprice them, Shah said.
people+ai is an offshoot of Infosys cofounder Nandan Nilekani’s EkStep Foundation.
You may also like
Truth behind viral Mo Salah photo as Liverpool fans get their hopes up over new contract
RSS mouthpiece article on church properties shows their 'true mindset': Kerala CM
Elton John is unable to see TV, read or watch his boys play sports
Mumbai News: BMC Conducts 'Fire Safety Awareness Rally' Covering Vikhroli & Mulund Fire Stations
AAP MLA Amanatullah Khan Moves Supreme Court Against Waqf Bill, Cites Violation Of Minority Rights
