OpenAI's first self-developed AI chip makes its debut. Wedbush warns that large-scale commercial use will require multiple rounds of design iteration.

date
20:55 26/06/2026
avatar
GMT Eight
This Wednesday, OpenAI released its first custom AI inference chip, Jalapeo. Wedbush Securities believes that this is just the beginning of its multiple self-developed chips.
This week, OpenAI released its first custom AI inference chip, Jalapeo. Wedbush Securities believes that this is just the beginning of its multiple self-developed chips. Wedbush Securities analyst Matt Bryson wrote in a research report, "Past experience shows that a mature application-specific integrated circuit (ASIC) requires multiple generations of iterations and years of polishing to be successful." Bryson added, "We expect this development could be positive for Broadcom Inc. and OpenAI if the chip's performance eventually meets its advertised standards. If the initial shipment scale is small and it requires second, third, or even fourth design iterations to achieve widespread adoption, we would not be surprised." Reportedly, Jalapeo is a specialized ASIC designed for large language model (LLM) inference. OpenAI is responsible for the underlying architecture design, Broadcom Inc. for silicon implementation and network hardware, and Celestica Inc., a Canadian electronics manufacturing services provider, for board and rack system integration. The project went from initial design to chip production in just 9 months, with OpenAI stating that their AI models accelerated the entire development process. OpenAI stated, "Jalapeo is the first product of a multi-generation computing platform that is planned to be deployed by the end of 2026 and steadily expanded in the following years." Self-developed chips as a key factor for IPO Joining the ranks of companies like Alphabet Inc. Class C, Amazon.com, Inc., and Meta Platforms, who develop their own custom chips, OpenAI's move may reduce their dependence on GPU leader NVIDIA Corporation. Independent technology analyst Patrick Moorhead said, "If chip suppliers capture over 75% of design profits, companies will surely seek alternatives. All hyperscale data center operators and two major model manufacturers have entered the self-developed chip race. Not everyone will succeed on this path, but most participants will benefit." Demonstrating the ability to control infrastructure costs in the public market will be a key factor for OpenAI's initial public offering (IPO). However, recent reports suggest that due to market volatility, OpenAI is leaning towards postponing its IPO timeline from this fall to 2027, while still aiming for a valuation of $1 trillion. Additionally, reports suggest that competitor Anthropic, also preparing for an IPO, is exploring self-developed chips. Meanwhile, Broadcom Inc. has become the preferred partner for companies aiming to develop their own chips. William Blair analyst Sebastien Naji said, "This collaboration further solidifies Broadcom Inc.'s leading position in the field of ASICs, helping the company to continue expanding its top customers and diversify its revenue in the AI business." The analyst maintains an "outperform" rating on Broadcom Inc.