When it comes to companies rolling their own custom chips, our main thesis is that doing this to save a few dollars on chips is profitable at best. Instead, companies want to build their own chips when they convey some form of strategic advantage.
The classic example is Apple, which ties its chips to its own software to significantly differentiate their phones and computers. Or Google, which customizes chips for their most intense workloads like search algorithms and video encoding. A few hundred million dollars in chip design costs are more than paid for in billions in additional sales for Apple or billions in capital expenditure and operating expense savings for Google. It’s important to point out that in both of these cases, the company has full control over the software running on its local chips.
Guest Author Jonathan Goldberg is the founder of D2D Advisory, a cross-functional consulting firm. Jonathan has developed growth strategies and alliances for companies in the mobile, networking, gaming and software industries.
So what’s in it for Amazon?
For Amazon, and more specifically for AWS, software control goes beyond them. AWS runs everyone’s software and therefore, by definition, AWS cannot control it. They have to run almost literally every form of software in the world. Nevertheless, AWS seems to be working very hard to push its customers to run workloads on their Graviton processors. AWS offers many ways to lock in customers, but silicon is not one of them. At least not yet.
AWS probably isn’t doing this to save money on the AMD and Intel x86 processors they buy. The fact that they only have two providers means they have plenty of leeway when it comes to pricing. To some extent, Graviton can be a safeguard against the day Intel stops being competitive in x86. (A point we may have already reached.)
That being said, we think there’s a bigger reason: power. Today, the main constraint in the construction of data centers is electricity. Data centers consume a lot of energy, and when designing new centers, companies need to stick to an energy budget. Now imagine they could reduce power consumption by 20%, which means they could fit more equipment into the same electrical footprint, which means more revenue. A reduction in the energy consumption of one part of the system means a much higher return on the overall investment. Then multiply that gain by 38 as the savings spread across all AWS global data centers.
Now, of course, the math is a bit more complicated than that. CPUs are only part of a system, so even though Graviton is 20% more power efficient for the same performance compared to an x86 chip, that doesn’t really translate to 20% more profit from the center of data, but the scale is about right. Moving to an in-house designed Arm processor can generate a sufficient increase in data center capacity to more than offset the chip design cost.
Going deeper, a major hurdle preventing more companies from moving to Arm workloads is the cost of optimizing their software for a new instruction set. We touched on this previous topic, porting software can be a lot of work. AWS has a strong incentive to encourage its customers to switch providers and seems to be doing what it can to facilitate this process. However, we have to wonder if it is not a one-way street.
Once customers switch to Graviton, it just shifts the friction. As we said above, today AWS can’t use x86 silicon to lock its customers into its service, but once customers switch to Graviton, all that optimization friction shifts in favor of it. ‘AWS, creating a new form of lockdown. Certainly, the barrier today exists between Arm and x86, not among the different versions of Arm servers. But one of the beauties of working with Arm is the ability to semi-customize a chip, and so it’s entirely possible that AWS could introduce proprietary features in future versions of Graviton.
We think Amazon has plenty of other good reasons to encourage a switch to its Arm-based Graviton processor, but we have to wonder if that lockout lingers somewhere deep in their brains. If true, it just gives other hyperscalers more reason to switch to Arm servers as well.