OpenAI Unveils In-House Inference Chip Jalapeño, Claims Up to 1.9x Energy Efficiency Over Nvidia

Habib Ur Rehman Thu, August 27, 2026 at 10:58 AM PDT | 3 min read

• AVGO
• CLS
• TSM
• NVDA

OpenAI has announced that its first homegrown inference chip, codenamed Jalapeño, can deliver up to 1.9 times more AI work per watt than comparable systems. The chip went from initial design to tape-out in just nine months. However, the more investable surprise lies in the partners that helped turn the design into silicon.

Broadcom Inc. (NASDAQ: AVGO) handled Jalapeño's physical implementation and supplied the Tomahawk networking. Celestica is responsible for board, rack, and system integration, while TSMC manufactured the chip. OpenAI designed the architecture and stated that its own models helped accelerate the timeline from initial design to tape-out, significantly faster than industry norms.

OpenAI's Nvidia Win Is Narrower Than It Sounds

In OpenAI's published InferenceX results, Jalapeño delivered:

• 1.5 to 1.9 times more work per watt
• 1.7 to 3.6 times lower end-to-end latency

These results were measured across three large open models, with comparisons including Nvidia GB200 and GB300 systems.

However, these results come with important caveats:

• Jalapeño is an inference chip, not a training accelerator.
• It was not compared with Nvidia's Vera Rubin.
• Some tests emphasized single-token prediction, while Nvidia systems can leverage multi-token techniques.
• Initial deployment is expected by year-end, meaning production economics remain unproven.

OpenAI rated Jalapeño at 700 watts but noted that measured sustained power stayed at or below 550 watts.

The Bull and Bear Case for Broadcom and Nvidia

Broadcom Inc. (NASDAQ: AVGO)

• Bull case: Jalapeño connects to a much larger prize — the companies plan to deploy 10 gigawatts of OpenAI-designed accelerators through 2029. Broadcom's AI semiconductor revenue already jumped 143% to $10.8 billion last quarter.
• Bear case: Risks include customer concentration, costly project financing, and margin pressure from expensive HBM memory.

NVIDIA Corporation (NASDAQ: NVDA)

• Bull case: OpenAI still states it needs plenty of Nvidia hardware, and Jalapeño does not replace the CUDA training ecosystem.
• Bear case: Every efficient custom inference chip shifts another recurring workload away from premium GPUs, posing a strategic threat.

Hedge Funds Remain Heavily Invested in Both

• Nvidia: Hedge-fund holder count rose from 277 to 286 in Q2, a 3.2% increase.
• Broadcom: Insider Monkey's latest count was 171 funds in Q2, down from 174 in Q1.

As of August 14, short interest represented only 1.20% of Broadcom Inc. (NASDAQ: AVGO)'s float and 1.23% of NVIDIA Corporation (NASDAQ: NVDA)'s. Short sellers are barely leaning against either stock.

Jalapeño is not yet an Nvidia killer. It is proof that Broadcom can help Nvidia's biggest customers become chip companies — and get paid even when Nvidia loses the socket.

Story Continues

While we acknowledge the potential of AVGO and NVDA as investments, we believe certain other AI stocks offer greater upside potential and carry less downside risk. If you're looking for an extremely undervalued AI stock that also stands to benefit significantly from Trump-era tariffs and the onshoring trend, see our free report on the best short-term AI stock.

READ NEXT: NVIDIA (NVDA): What Foxconn and Super Micro Are Telling Us about the AI Boom and Pony AI Is Scaling Robotaxis Fast—Can the Stock Reach BofA's $17 Target?

Disclosure: None. Follow Insider Monkey on Google News.