OpenAI switches to new chips, Samsung launches HBM4: tech digest

# OpenAI Transitions to New Chips, Samsung Launches HBM4: How the Artificial Intelligence Market Will Evolve

The technological landscape of artificial intelligence is undergoing a significant shift. OpenAI presented the GPT-5.3-Codex-Spark model running on Cerebras chips, Samsung began shipping HBM4 memory for accelerators, and startups like MiniMax and ByteDance are demonstrating alternatives in the generative AI segment. This is not just a routine update — it is a signal that Nvidia's monopoly on AI components is beginning to crack.

OpenAI's transition to Cerebras chips is symbolic. Years of refinement of processors from Nvidia gave the company an unprecedented advantage in the market. Its A100, H100 and newer models became the de facto standard for training large language models. But Cerebras, specializing in creating processors for AI workloads, offers its own approach — an architecture optimized specifically for transformers and deep learning. OpenAI's choice to use these chips means the company is actively diversifying its computational resources. This strategic decision not only reduces dependence on a single supplier, but also signals to investors and competitors: viable alternatives have appeared on the market.

In parallel, Samsung is expanding its presence in a critically important segment of high-speed memory. The release of HBM4 is a qualitative leap in high bandwidth memory technology. HBM is necessary to accelerate model training, as it allows rapid transfer of huge volumes of data between computational elements. Samsung forecasts a threefold increase in sales of AI components, and this is not just optimistic prediction. With the growing number of data centers training their own models, demand for quality memory will only increase. The company is betting that HBM4 standardization will become the next wave in the evolution of AI infrastructure.

Against the backdrop of hardware innovations, software competition is also occurring. MiniMax released the M2.5 model, and ByteDance presented the Seedance 2.0 video generator — solutions that call into question the monopoly of OpenAI and Google in the field of generative AI. If previously the quality of models directly depended on the volume of computational resources, now architectural and methodological breakthroughs allow small companies to compete with giants. This is AI democratization in action.

The financial sector is also reacting to these shifts. Elon Musk topped Forbes' innovators list, surpassing Jensen Huang, the head of Nvidia. This is symbolic — despite Nvidia's dominance in hardware, public attention is shifting to creators of services and platforms. SoftBank recorded record profits thanks to investments in AI startups, and Lenovo reached a historic revenue peak by integrating neural networks into its products. This shows that value is extracted not only from the hardware itself, but from its application in real solutions.

The picture that emerges is this: the artificial intelligence ecosystem is becoming more fragmented and competitive. New players are appearing in the chips, memory, and software sectors. This is good for innovation, good for prices, and good for the ecosystem as a whole. However, it also means that investors and developers need to choose partners more carefully, as the technological landscape becomes increasingly complex and rapidly changing. The future of the AI industry will not belong to a single player — it will belong to those who best integrate hardware and software into a unified system.