Deeper-than-expected AI integration ‘speaks to higher incremental cost risk’ for Google – Morgan Stanley

This post was originally published on this site

https://i-invdn-com.investing.com/news/LYNXMPEB0Q00C_M.jpg

On Tuesday, Alphabet (NASDAQ:GOOGL) owned Google unveiled its new AI search assistant, Bard, a rival to OpenAI’s ChatGPT.

The company said it will begin rolling out Bard in the coming weeks, opening it up to “trusted testers” before making it more widely available to the public.

It is powered by Google’s LaMDA, or Language Model for Dialogue Applications.

Reacting to Google’s new AI chatbot, Morgan Stanley analysts told investors that the firm believes the tech giant has the AI tech and scale to maintain/grow its leading user base.

However, a deeper-than-expected AI integration “speaks to higher incremental cost risk, as we see every 10% of searches moving to language models adding ~ $1.2bn of opex.”

“The AI race is on,” added the analysts. “Our work on natural language queries suggest they could be 5x more expensive (on average).”

The analysts explained that the compute intensity of natural language models that store, recall, analyze and compile large amounts of text into answers in a natural language format is significant.

The firm’s updated analysis of ChatGPT’s model size, time to compute, average words generated per query, Azure Nvidia (NASDAQ:NVDA) A100 GPU pricing tiers, and an estimated 50% Azure gross margin (for this analysis) leads them to estimate that GOOGL’s incremental natural language average cost per query is likely to range from $0.0022-$0.0220.

“Early demos show how the NL compute costs are potentially more incremental… But, in our view, the greater risk from NL compute comes from the higher potential incrementality of the costs given the deeper than expected integration of NL into basic search results,” said the analysts. “Every 10% of Google Search queries that add AI/Natural Language could add $1.2bn to ’24 Opex.”

He concluded that if 50% of queries have natural language integration by 2024, it would add $6 billion of incremental costs.