Nvidia, AMD, and the Ghost of Groq

The big tech companies have been throwing money at AI like it’s the last gasp of a dying star. Most of it went to GPUs from Nvidia and AMD, and networking gear from Broadcom. Capital is a fickle lover, though. It’s moving downstream now, where Silicon Valley startups are making noise.

Groq just raised $750 million, valuing itself at $6.9 billion. Its investors include Samsung, Cisco, and BlackRock. That’s not just money. It’s a vote of confidence in a world where semiconductors are supposed to be the domain of old gods.

Groq doesn’t make GPUs. It makes language processing units, or LPUs. These are for inference, the part of AI where models go live. GPUs are good for training. LPUs are faster, more efficient, and less prone to lag. It’s like comparing a bulldozer to a race car. Both get the job done, but differently.

Nvidia owns 90% of the AI accelerator market. Its CUDA software is a fortress. But Groq’s rise shows the AI world isn’t as simple as it seems. So it goes.

AMD has always been the cheaper option. But Groq’s arrival complicates that narrative. If cloud companies start mixing chips, AMD might benefit. Or maybe not. The future is a mystery, even to those who think they know it.

Nvidia is still the safer bet. It’s profitable, innovative, and has next-gen chips on the way. AMD is riskier. It lacks the ecosystem lock-in that makes Nvidia hard to dislodge. But Groq’s presence could open doors. So it goes.

The AI chip race isn’t over. It’s just getting weirder. Nvidia remains the favorite, but the underdog has a way of showing up when you least expect it. So it goes. 🤷‍♂️

2025-09-26 21:25

Read More