SambaNova revives older NVIDIA GPUs in fresh benchmarks — potential extension of the GPU depreciation curve
Nvidia to invest $5bn in Safe Superintelligence, Sutskever firm to substantially increase GPU compute
Nvidia to invest $5bn in Sutskever's Safe Superintelligence
Nvidia considers $250bn backstop for OpenAI's planned 10GW Ohio data center - report
Nvidia weighs $250bn backstop for OpenAI's 10GW Ohio data center
AMD and Cerebras join forces against Nvidia’s Groq LPUs
AMD and Cerebras join forces against Nvidia's Groq LPUs
AMD attacks the rack with Helios systems that rival Nvidia's
AMD's first rack-scale Helios claims to beat Nvidia's Vera Rubin on paper
Nvidia’s LPU gamble
Nvidia's LPU gamble: the GPU kingpin looks beyond the GPU
Nvidia shows off Vera Rubin platform for tokenmaxxing
Nvidia shows off its Vera Rubin platform aimed at "tokenmaxxing"
Nvidia partners with Noetra for 140MW Vera Rubin cluster in Japan
Nvidia partners with Noetra on a 140MW Vera Rubin cluster in Japan
Intel-backed AI chip startup SambaNova breathes new life into aging Nvidia GPUs in latest benchmarks
SambaNova's H200+SN50 hybrid hits 763 tok/s on MiniMax M2.7 benchmark
The leading-indicator read is that, amid tight AI-compute supply and demand, a concrete path is emerging to extend the useful life of GPUs that were on the retirement track. If the 3-to-5-year depreciation assumption breaks and the secondary GPU market expands, hyperscaler GPU-refresh cycles (and Nvidia's sales cadence) could be fundamentally rewritten.
The read forks on (a) whether SambaNova's approach gets adopted through hyperscalers and the used-GPU market becomes the primary secondary channel, or (b) whether it stays at the benchmark level while software-porting, power-efficiency, and TCO integration costs prevent it from scaling. The first suppresses demand for Nvidia's new products; the second reconfirms Nvidia-side lock-in.
Watch (1) SambaNova's disclosure of supported GPU generations and specific workloads, (2) used-GPU reuse cases at hyperscalers (particularly Oracle, CoreWeave, Lambda), (3) Nvidia's warranty and software-update policies for older GPUs, and (4) Intel's follow-on investment in SambaNova and any Foundry plans for related chips.