Arm FQ2 26 – The Power Game

Power constraints play to Arm’s advantage.

• Arm reported good results, and with power rapidly becoming a bottleneck in the data centre, its competitive edge just got a bit keener, meaning a better medium-term outlook.
• FQ2 26 revenues / Adj-EPS were $1.14bn / $0.39 ahead of estimates of $1.06bn / $0.33 as interest in using Arm in a wide variety of products continues to expand.
• This can be seen in the License revenue, which grew by 56% YoY to $515,m leaving Royalties way behind but still decent on 21% YoY.
• Licenses are up-front payments that customers make when the customer decides to use Arm in a product, and then they go on to pay royalties when the product ships.
• Hence, this is a good sign that the growth story that Arm has been outlining for some quarters, still has legs and will continue even once the migration from v8 to v9 is essentially complete.
• There are two main components to this story:
  • First, Power: which is quickly becoming an issue for anyone who needs to roll out gigawatts of compute power to provide inference for their AI service.
  • OpenAI intends to roll out just over 30GW on its own, which amounts to just over 2% of total electricity generation capacity in the USA.
  • When one adds everyone else’s ambition to the list, it is easy to see how power is becoming a scarce commodity, leading to price rises, making data centre construction less popular in local communities.
  • This plays directly to Arm’s strengths and makes Nvidia’s deal with Intel for x86-based data centre CPUs look like window dressing.
  • Second, Compute Sub Systems (CSS): which are basically IP building blocks which a customer can use with other pieces of IP to design and build a full chip.
  • This is similar to buying a processor design from Arm instead of just licensing the IP, but it is one step up in terms of complexity, meaning a higher price.
  • Arm has signed 19 deals with 11 companies, of which 3 came in the last 3 months which is helping companies like Samsung, whose DNA is not in CPU design, keep their products competitive with Qualcomm, MediaTek and Apple.
  • This should also be a good revenue driver as it means higher royalties per device shipped, which goes straight to revenue, profit and cash flow.
• This is what gives Arm the confidence to discuss the long-term so optimistically, and also provides support for short-term guidance.
• Here, the numbers are good with FQ3 26 revenues / adj-EPS expected at $1.18bn – $1.28bn ($1.23bn) / $0.37 – $0.45 ($0.41), which is nicely ahead of estimates if $1.15bn / $0.37, which is why the shares are up 5.6% in pre-trading.
• While the outlook remains pretty good and the long-term story that the company has been forecasting is coming in to view, the valuation of the shares for a value-oriented contrarian is challenging.
• Here, FY 2026 and FY 2027 PER are at 100.0x and 75.1x, respectively, which are easily the most expensive in the sector, with the exception of perhaps Palantir.
• This is why I continue to prefer the other parts of the AI-related trade, such as inference at the edge or nuclear power, which is attracting even more attention now that power is becoming a bottleneck for AI expansion.