NVIDIA brings Co-Packaged Optics to Feynman GPUs in 2028
Updated
Updated · Wccftech · May 3
NVIDIA brings Co-Packaged Optics to Feynman GPUs in 2028
8 articles · Updated · Wccftech · May 3
The move pulls commercialisation forward from 2033 by five years as AI systems stretch beyond 10km and require data links of several hundred gigabits per second.
NVIDIA has also said Feynman will use 3D die stacking, a new Rosa data-centre CPU and custom HBM, with Intel expected to provide advanced packaging including EMIB.
The shift aligns with the OCI-MSA optical interconnect push involving Broadcom, AMD, Meta, OpenAI and Microsoft, while AMD is reportedly targeting similar co-packaged optics deployment around 2028.
As NVIDIA bets on Intel's foundry for its 2028 chips, can rival AMD exploit this supply chain gamble to win the AI hardware race?
While new optical tech cuts data transfer power, will the explosive growth of 'AI factories' still trigger an unprecedented global energy crisis?
With NVIDIA solving the data bottleneck, will a crippling memory shortage become the new 'wall' that halts its 2028 AI ambitions?