Big tech firms are investing in custom chips to accelerate Artificial Intelligence workloads, reduce vendor dependence and lower the cost of deploying advanced models. Meta has introduced a new generation of in-house processors to support ranking, recommendations and future products such as smart glasses, while Intel launched the gaudi 3 as part of a broader push by chipmakers to supply hardware tailored to training and inference. the rush mirrors moves by other players, including google and apple, which is reportedly preparing m4 processors with integrated Artificial Intelligence capabilities for its mac line.
Intel said the gaudi 3 delivers more than double the power efficiency and processes models roughly one and a half times faster than nvidia’s h100 gpu. the chip is offered in multiple configurations, including eight chips on a single motherboard and standalone cards that can be integrated into existing systems. Intel reported testing on models such as meta’s llama and highlighted support for workloads including stable diffusion for image generation and openai’s whisper for speech recognition. those performance and integration claims are central to vendors’ pitches that custom silicon can better match specific algorithms and deployment needs.
Industry experts quoted in the article said custom processors can lower per-customer training costs, enable more private and proprietary data control and make specialized deployments viable without relying solely on external model providers. amrit jassal of egnyte said in-house chips lower the bar for training per-customer, per-task models and support high-security use cases that move beyond API consumption. rodolfo rosini of vaire emphasized financial drivers and allocation constraints around popular chips, and michal oglodek of ivy.ai highlighted privacy and strategic control as reasons companies will race to own hardware as well as software.
