In NVIDIA's latest announcement, the semiconductor giant revealed that OpenAI used its Hopper and GB200 NVL72 supercomputing systems to train and run the GPT-5.2 large language model, which was just launched in December 2025. This collaboration further solidifies NVIDIA's dominant position in providing hardware infrastructure for next-generation artificial intelligence technologies.
Background
According to NVIDIA, as artificial intelligence models grow increasingly complex and demand colossal computing power, leading developers like OpenAI have no choice but to rely heavily on specialized hardware architectures. NVIDIA's latest GB200 NVL72 system, which combines Blackwell GPUs with Grace CPUs, is currently the optimal solution for large-scale natural language processing tasks. Looking beyond GPT-5.2, OpenAI is also developing GPT-5.3 Codex—the first agentic programming model capable of building itself, slated for release in February 2026.
Developments
The deep integration of OpenAI's software with NVIDIA's hardware underscores an irreversible trend in the AI industry: powerful hardware defines the limits of the model. NVIDIA emphasized that optimizing infrastructure from the silicon level to ultra-high-speed networking systems is key to helping models like GPT-5.2 achieve breakthrough processing performance, meeting the demanding specialized workflows of professional users.
Why It Matters
For the tech community and developers in Vietnam, this event highlights a strong shift toward the era of autonomous AI agents like GPT-5.3 Codex. The technology race is no longer just a competition over algorithms; it has become a battle for access to high-performance supercomputing systems like NVIDIA's Hopper or Blackwell—a massive financial barrier for domestic tech startups.