Investing.com -- Piper Sandler analysts identified Nvidia (NASDAQ: NVDA ) and Taiwan Semiconductor Manufacturing Company as the "clear winners" from the growing demand for inference compute power, driven by the increasing use of AI applications.

Following discussions with a data center engineering expert, the firm’s analysts reiterated their belief that AI infrastructure spending will continue rising.

"Compute and further capex requirements are only increasing as inference applications for LLMs continue to proliferate," the expert noted.

While DeepSeek’s advancements have made headlines, Piper said the expert believes leading AI labs like OpenAI and Anthropic remain six months ahead.

One of the key takeaways from the discussion was that inference, rather than training, is becoming the dominant cost driver for AI labs.

"Inferencing, which is the monetization engine for most of these labs, is what drives the capex up and there are no shortcuts associated with inference compute deployments," the expert told Piper.

Currently, OpenAI and Anthropic allocate about 75% of their compute resources to inference, but this is expected to increase to 90% as training requirements stabilize.

The analyst note emphasizes that Nvidia is best positioned to capitalize on this shift. "According to our expert, NVDA followed by TSMC are the clear winners for capex dedicated to inferencing."

The report also highlights that networking, particularly optical solutions, will play a crucial role in supporting this trend.

Despite the excitement surrounding DeepSeek, analysts found its cost efficiency claims less impressive when compared to OpenAI.

"A similar run on OpenAI would have cost somewhere in the $10 million range, thereby making the DeepSeek training run cost not so impressive relative to the U.S. comps," the expert noted.

With inference applications rapidly expanding, analysts remain bullish on Nvidia and TSMC as key beneficiaries of AI's continued evolution.