Alphabet’s reported Gemini chip project puts AI costs back in focus

Alphabet is reportedly working on a new server chip aimed at making Google’s Gemini AI models cheaper and more efficient to run. For retail investors, the point is straightforward: AI growth is exciting, but the cost of powering it has become one of the biggest questions hanging over Big Tech earnings.

The Information reported that the chip is known internally as “Frozen v2” and is expected to arrive sometime in 2028, citing anonymous sources. The report said the chip could be six to 10 times more efficient than Google’s current AI chips, based on tokens produced per unit of power. A token is a small unit of text that an AI model reads or generates, so more tokens per watt means more output for the same electricity use.

Google did not directly confirm the report in a statement to TechCrunch. The company said its teams are “constantly researching and experimenting with new innovations” to improve performance and efficiency, and added that not every project reaches production.

Google also pointed to its “full stack approach,” meaning it designs hardware and software together rather than treating chips and models as separate pieces. In plain English, that can help a company tune the machine for the exact workloads it expects to run, such as Gemini answering prompts or generating code.

Why custom AI chips are getting attention

AI companies have been trying to build more of their own chips for two reasons highlighted by TechCrunch: they want their internal models to run more efficiently, and they want to reduce pressure from limited global AI computing capacity.

There is also a market-power angle. Nvidia has long dominated AI hardware, leaving many major AI developers reliant on its chips. Building custom silicon gives companies another path, even if Nvidia remains central to the broader AI buildout.

The cost side has become harder for investors to ignore. TechCrunch cited broader concerns about AI spending that have cooled some of the earlier enthusiasm around the sector. Alphabet, in particular, has faced investor scrutiny over the money it plans to put into AI infrastructure.

Earlier this year, Google said Alphabet planned to spend between $180 billion and $190 billion, according to TechCrunch. With numbers that large, investors are watching for evidence that the company can turn AI infrastructure spending into products, revenue and efficiency gains.

Alphabet is not alone

The reported Frozen v2 project fits a broader pattern across the AI industry. In June, OpenAI announced its first custom chip, an inference processor called Jalapeño, according to TechCrunch. Inference means running an already trained AI model to produce an answer, image, code or other output.

TechCrunch also reported earlier this month that Anthropic was in talks with Samsung about a possible custom chip partnership. Those efforts show how central chip design has become to the AI race, especially as companies try to lower the cost of serving users at scale.

The market reaction was positive after The Information’s report. CNBC reported that Alphabet shares rose about 3% on Monday morning following the news, ahead of the company’s earnings report later this week.

This story draws on original reporting from TechCrunch.