Yesterday Amazon announced that Twitter will begin using AWS for its core Timeline product (with a good write-up from Eric Jhonsa at The Street here). This announcement is important in its own right, one of the largest Internet sites is moving away from owning their own data centers, another big milestone in the move to the Cloud. However, for us the biggest element of this news was a prominent mention in Amazon’s press release that Twitter will be running the service on instances of Amazon’s Gravitron Arm-based server CPU. It is hard to overstate how significant a milestone this is.
For 20+ years, data centers were built on Intel CPUs. Intel makes something like half of its profits from selling these high-performance chips. And for 10+ years, as Intel struggled and then failed to enter mobile, and its core PC market saturated, they held onto the commanding heights of the data center business. Recently, AMD has made significant inroads with its x86 CPUs as Intel has struggled with its transition to advanced manufacturing nodes. For years, companies, both large and small, have tried to break into that market using Arm-based chips, constantly nipping at Intel’s heels. These companies have made inroads, but so far none has won a significant share at any of the “Super 7” – the top cloud scale data center building Internet companies who consumer ~50% of data center silicon. (The Super 7 are AWS, Google Cloud, Microsoft Azure, Alibaba Aliyun, Facebook, Baidu and Tencent. If it were Super 10, Twitter would probably make the cut.)
But now we have a Super 7 member winning business with their own Arm-based solution. The press release even includes a quote from Twitter’s CTO extolling the virtues on Arm based products.
When comparing Intel’s x86-based CPUs with Arm-based CPUs, the commonly held wisdom has been that x86 offers better raw performance while Arm offers better power consumption. This has become a major selling point for Arm because data centers are largely power constrained. Saving 5%-10% on the power budget is an interesting proposition. But we believe the case for Arm goes further, because we have likely now reached the point where Arm CPUs offer better raw performance than Intel products. This likely would have happened in a few years anyway, but Intel being stuck at 10nm has advanced that tipping point considerably.
This does not mean the end is nigh for Intel’s server products. There are still many workloads where x86 products will have a performance advantage, and Intel still makes ~$5 billion a quarter in profits which it can invest into fixing itself. And they have immense talent.
That being said, this is a major blow to Intel. Graviton is designed by Amazon, so it comes with some built-in advantages when earning design wins at AWS. But this is still a clear signal. Suddenly, other Arm CPU offerings have a rallying point for their marketing – Arm works for AWS and Twitter. Companies like privately-held Ampere and Nuvia can point to this as validation which will likely force many other customers to take a much closer look at Arm servers. We suspect that those companies already have significant design wins among data center builders, and yesterday’s news will accelerate those and future conversations.
So while it is not lights out for Intel, the doomsday clock has definitely ticked ahead. Intel now desperately needs to solve its manufacturing process. And in the years it will take to do that, they will have to contend with material market share losses. In the past, they have responded to Arm inroads through aggressive pressure on customers and partners, but with the biggest customer out there now flagging its own Arm chips, those tactics are more likely to fail. Another response is to further their bundling of other products – networking, memory, FPGAs and AI. They have a wide offering of products, but none of them are truly category leaders, at least not the way their CPUs once were. They still have many levers to pull, but they will need to move faster and smarter than they have been.