Jim Cramer says Nvidia and Broadcom are the two clearest stock-market winners from OpenAI’s GPT-6 Astra launch, but the more interesting story is why both can benefit at the same time.
Nvidia supplied much of the compute behind Astra’s development, while Broadcom is helping OpenAI build custom silicon for the next stage: running models at enormous scale.
That creates a split AI trade: Nvidia for frontier-model training, Broadcom for increasingly customized inference infrastructure.
Cramer made the call Tuesday, saying Nvidia was the stock he would buy while also arguing Broadcom could gain as OpenAI expands its hardware footprint. Nvidia fell about 2% that session while Broadcom rose nearly 3%, showing investors were already treating the two names differently.
Astra Shows How Much Compute OpenAI Still Needs
OpenAI launched GPT-6 Astra on Sept. 3, calling it its most capable broadly deployed model, with major gains in coding, computer use, cybersecurity and professional work.
Nvidia CEO Jensen Huang said Astra trained on roughly 100,000 Grace Blackwell systems, with another 400,000 chips coming online for OpenAI.
That supports Cramer’s Nvidia thesis: frontier AI models are still consuming huge volumes of GPU capacity.
Our AI chip stocks coverage shows why Nvidia remains the dominant training supplier even as competitors gain ground in custom accelerators.
Broadcom’s Angle Is Different and Potentially Bigger
Broadcom is not trying to replace Nvidia everywhere.
Its opportunity is OpenAI’s push toward custom inference silicon.
In June, OpenAI and Broadcom unveiled Jalapeno, OpenAI’s first internally designed intelligence processor, built for large-language-model inference and planned for gigawatt-scale deployment.
That matters because inference becomes a larger cost as more users run AI models every day.
Broadcom already expects AI semiconductor revenue to hit about $115 billion in fiscal 2027 and double to roughly $230 billion in 2028. Its latest quarter produced $16.7 billion in AI chip revenue, up 221% year over year.
Coinpaper recently covered that AI forecast, which makes Astra more than a one-model catalyst: OpenAI is one of several hyperscale customers supporting Broadcom’s multiyear custom-chip pipeline.
The Freshest Angle Is OpenAI’s Expanding Chip Strategy
OpenAI is also widening its supplier base.
Reuters reported Wednesday that the company is deepening work with Samsung on next-generation chips, while Samsung and SK Hynix are already supporting the Stargate project with memory supply commitments.
That is the real risk to the simple “Nvidia wins everything” narrative.
OpenAI increasingly wants control over the full stack: from models and software to chips, memory and data centers. Broadcom benefits from that shift because custom silicon is exactly where hyperscalers are spending more.
Nvidia still owns the strongest position in training, but Broadcom may capture more of the inference layer as OpenAI scales Astra and future models.
