Alphabet: Technical Paper on Energy Consumption Supports Our Cost Advantage Thesis

Key Morningstar Metrics for Alphabet

• Fair Value Estimate: $237.00
• Morningstar Rating: ★★★★
• Morningstar Economic Moat Rating: Wide
• Morningstar Uncertainty Rating: Medium

On Aug. 21, researchers at Google published a technical paper that delineates the energy cost profile of a median Gemini text prompt. The headline number, 0.24 watt-hours of energy consumption, is materially lower than many commonly cited estimates.

Why it matters: We see Alphabet GOOG having a clear cost advantage in deploying artificial intelligence, due to its vertically integrated tech stack that includes its AI accelerators. Further, we believe that over time, this cost advantage will let Alphabet deploy AI across its businesses.

• One key investor concern around AI, especially AI infused search, has been the cost of running AI search and its inevitable impact on Alphabet’s margins. We see this paper as throwing serious doubts on the “Rising costs will destroy margins” argument.
• Encouragingly, the median Gemini prompt’s energy consumption decreased 33 times over the last year. Also, according to Alphabet’s past disclosures, the energy consumption of traditional search is estimated to be around 0.30 Wh, slightly higher than the median Gemini prompt.

The bottom line: We maintain our $237 fair value estimate for wide-moat Alphabet and continue to see the firm’s economic moat enforced by its cost advantage, network effect, and switching costs. Despite shares being up more than 40% from their April lows, we view them as undervalued.

• At the firm’s I/O conference in May, Alphabet noted that its monthly token usage had grown 50 times over the last year, even as its margin profile has continued to expand. We believe the material gains in model efficiency are the core reason for margin resilience.

Between the lines: While the median Gemini prompt’s energy consumption is encouraging, as is the immense model efficiency the firm gained over the last year, we’d imagine that more complex prompts would require more energy and are likely to be costlier than traditional search.