Google unveils ARM-based Axion chips for AI, claiming they have up to 30% better performance over the top ARM rivals, coming to Google Cloud customers in 2024

Wall Street Journal 2024-04-10

Context & Ripple Effects

Axion puts Google’s ARM CPU effort into its cloud offering, extending a broader strategy of designing more of the infrastructure layer rather than relying on a single supplier path. Later coverage describes Google applying an Nvidia-like approach to turning its AI chips into a business, including TPU capacity offered beyond its own operations.

The move also sits in an evolving ARM ecosystem: Arm itself later moved beyond its traditional licensing role with an AI chip of its own. That makes CPU design, cloud distribution and accelerator access increasingly intertwined competitive levers.

First-order effects

Google Cloud customers are slated to gain an ARM-based compute option in 2024, with Google claiming up to 30% better performance than leading ARM alternatives.
Google gains a proprietary server CPU it can pair with its existing AI infrastructure, while rival ARM-based cloud offerings face a new performance claim from a major platform operator.

Second-order effects

Cloud buyers running compatible workloads gain another basis for choosing infrastructure beyond accelerator availability: CPU performance and the economics of the full compute stack.
Other cloud providers and ARM ecosystem partners are pressured to differentiate through their own silicon, instance performance, software support or pricing rather than treating ARM architecture alone as the differentiator.

Third-order effects

If hyperscalers continue to deploy in-house CPUs alongside specialized accelerators, cloud competition shifts toward vertically integrated, heterogeneous fleets rather than standardized server configurations.
The strategic value of chip design may increasingly lie in controlling deployment and customer access; Arm’s later move into its own AI silicon suggests that supplier and cloud-platform roles could overlap more often.

The trend: AI infrastructure is moving toward vertically integrated, heterogeneous cloud stacks in which proprietary CPUs and accelerators are both product features and strategic leverage.

Discussion

Phoronix Phoronix on x
Google Announces Axion ARM-Based CPUs For The Cloud
@danielnewmanuv Daniel Newman on x
As always top insights on CNBC from @PatrickMoorhead with @MorganLBrennan breaking down the nuance of the AI chip space. I agree on Foundry and it's long tail for @Intel and also agree that not just GPU but OneAPI is the big Inflection for Intel to compete more broadly with...
@benbajarin Ben Bajarin on x
We all knew it was coming, because what else was @jonmasters working on :), Google now officially with an @arm CPU. All three hyperscalers now with @arm cpus for their software stack. This gets super interesting as the ratio of CPUs to GPUs gets closer in future datacenter....
@danielnewmanuv Daniel Newman on x
Yep, @DivesTech is directionally correct on the what and why of @Google launching its home grown CPU Axion. Vertical integration and workload optimization plus oft better economics is a good combination. Plus, saying this all while wearing that jacket should yield at least...
@stevendickens3 Steven Dickens on x
Google Cloud Axion - A quick take. Google's venture into the realm of custom silicon chips is not a new endeavor; the company has consistently demonstrated its prowess in designing and deploying these chips on a large scale with its TPU portfolio being on its 5th generation. The.…
@stevendickens3 Steven Dickens on x
.@googlecloud is getting further into custom silicon with the launch of Axion. Listen to @ThomasOrTK give the 30 second overview below 👇 My quick take: This move makes perfect sense and the ARM collaboration is a wise move. Custom silicon will be the battleground for AI... [video…
@aschilling Andreas Schilling on x
This is the @googlecloud #Axion that runs general-purpose workloads on C4A like web and app servers, microservices, open-source databases, in-memory caches, data analytics engines, media processing, AI training and inferencing. Its based on Arm Neoverse V2. [image]
@googlecloud @googlecloud on x
Google has a long history of designing silicon chips and deploying them at scale. So, what's next? Introducing Google Cloud Axion—our new custom ArmⓇ-based CPU designed for the data center #GoogleCloudNext. [image]
@jimcramer Jim Cramer on x
so far i think Arm and Palo Alto are big winners of this Google lovefest
@iancutress @iancutress on x
Here's the official image. Monolithic, DDR5, BGA. No word on core counts, frequencies, memory bandwidth, PCIe, CXL, TDP, multi-socket compatibility. Yet. That's why I'm here. [image]
@patrickmoorhead Patrick Moorhead on x
BOOM! The worst kept semi secret is out. @Google announces Axion, an @Arm-based general purpose processor. Already deployed internally on services like @YouTube, in preview for enterprise customers soon. More details later from @MattKimball_MIS. [image]
@iancutress @iancutress on x
On perf, Google are comparing it to Graviton v3 - 30% better perf. To x86, they're saying 50% better perf and 60% better perf/watt. They've not said if that's a per-core or per-socket statement.
@quinnypig Corey Quinn on x
“We're announcing the Google Axion processor; it'll be available later this year.” Proving the Google Axiom that “only OpenAI can announce things that are shipping that day.”
@cnbcovertime @cnbcovertime on x
“This is a flex the muscle...we're talking about a trillion dollar market opportunity over the next decade. When you look at what Kurian and Google are doing, this is exactly what they need to go after,” says @divestech on $GOOGL's announcement of a new custom arm-based chip. [vi…
@dhinchcliffe Dion Hinchcliffe on x
The #AI Chip Era: @Google unveils new custom @Arm-based chips named #Axion, following similar efforts at rivals Amazon and Microsoft https://www.cnbc.com/... 60% more energy efficient than same workload on x86, which is a large cost differential. #cloud #AI #CPUs #compute [image]
@danielnewmanuv Daniel Newman on x
Noteworthy as @Google ramps up it's silicon ambitions that it also called out that it trains Gemini with its in-house TPU. Not suggesting it doesn't use @nvidia but worth mentioning since many like to claim nvidia is the only training chip on the planet. $NVDA $GOOGL
@mattkimball_mis Matthew Kimball on x
Just as expected, @Google announces its @Arm based CPU - Axion. It will be used for internal workloads (e.g. Spanner, Google Earth) and for @googlecloud enterprise workload support. https://cloud.google.com/...
@pramod_bhatotia Pramod Bhatotia on x
The x86 dominance is diminishing, and the cloud is becoming even more heterogeneous! https://cloud.google.com/...
@jonmasters @jonmasters on x
Proud to be a member of the team that brought you @Google Axion, Google's first custom @Arm CPU #ArmServers Read more in Amin's blog, below: https://cloud.google.com/...
@jonmasters @jonmasters on x
I remember telling my former CEO Jim at @RedHat back in the mid-2010s that all the hyperscalers were going to build their own CPUs. Now with Graviton, Cobalt, and Axion, the set is complete #ArmServers
@felixclc_ @felixclc_ on x
Congrats to @jonmasters and so many other Googlers on the launch of Axion! With its launch today, it marks all 5 highest valued tech companies all building their own in house @Arm devices. Also interesting is how all but one are using the arm Neoverse vendor core IP model! #HPC
@thomasortk Thomas Kurian on x
We're introducing custom silicon advancements with Google Axion, our first custom ArmⓇ-based CPU designed for the datacenter.
@thomasortk Thomas Kurian on x
We're announcing key advancements to our AI-optimized infrastructure, including the GA of TPU v5p, A3 mega coming soon, and that NVIDIA HGX B200 and GB200 NVL72 will be coming in early 2025. We're also enabling AI anywhere with new capabilities in Google Distributed Cloud.
@jkowall @jkowall on x
Today's most interesting announcement from @googlecloud is their new ARM-based CPUs called Axion, which will likely start to compete with @awscloud Gravitons. It's about time, and it's about to get a lot more exciting in compute. https://cloud.google.com/...
@madscapital @madscapital on x
“Axion processors combine Google's silicon expertise with Arm's highest performing CPU cores to deliver instances with up to 30% better performance than the fastest general-purpose Arm-based instances available in the cloud today” $GOOG https://cloud.google.com/...
@firstadopter Tae Kim on x
The big Google hyperscaler chip design win for ARM “Axion processors improve performance by up to 30% compared with the fastest similar Arm-based chips available in the cloud” “Axion chips will become accessible to external customers later this year” https://www.wsj.com/... [imag…

Chronicles