Axion by Google DeepMind: status and timeline

Axion

Cut the cost and energy of ordinary cloud computing (web servers, databases, analytics) by designing its own CPU, so more of Google's power budget can go to AI.

ScalingStage 5 of 5

Two generations on sale: C4A (Neoverse V2) since Oct 2024 and N4A (Neoverse N3), generally available Jan 2026.

Updated 27 Jan 2026Checked 11 Oct0 updates this week

Milestones

No announced next step

Axion unveiled, built on Arm Neoverse V29 Apr 2024Complete.
C4A virtual machines generally availableOct 2024Complete.
N4A (Neoverse N3) generally available27 Jan 2026Complete.

Most important updates

Google Cloud makes Axion-based N4A virtual machines generally available27 Jan 2026

Current obstacles

Software must be rebuilt for ArmPrograms compiled for x86 must be recompiled and retested; Google has ported many internal services, but some customer software still lags.
Rival Arm chipsAWS Graviton, Microsoft Cobalt and Nvidia's Arm CPUs compete on the same pitch, so Axion's edge must come from price and Google's offload chips.

Physics limits

Power density of siliconFaster clocks raise power steeply, so modern CPUs gain mainly by adding cores; but much software cannot split its work across many cores.
Memory wallCores can compute far faster than memory can feed them data, so for databases and analytics, memory bandwidth, not core count, often sets the speed.

How it works

3 parts

Cores

Licensed Arm cores

Google licenses core designs from Arm (Neoverse V2 for speed, N3 for efficiency) and builds its own chip around them, skipping x86's older instruction baggage.

Offload

Titanium does the chores

Separate Google chips called Titanium handle networking, storage and security, so all CPU cores are left for the customer's software.

Two lines

Fast and frugal versions

C4A targets steady high performance such as databases and caches; N4A trades peak speed for lower price per job on web and microservice work.

Spec sheet

Spec	Axion
C4A max vCPUs (bare metal)	96^R (reported)
C4A max memory (bare metal)	768 GB DDR5^R (reported)
C4A networking	Up to 100 Gbps^R (reported)
N4A max vCPUs	64^R (reported)
N4A price-performance vs x86	Up to 2x^R (reported)
N4A performance per watt vs x86	+80%^R (reported)
Energy efficiency vs x86 (2024 claim)	Up to 60% better^R (reported)

^R reported by the company

Update log

1 update

Tue 27 Jan

Minor: Google Cloud makes Axion-based N4A virtual machines generally availableBlogCommercial

About Google DeepMind

The team behind Axion

Google DeepMind

Gemini, AlphaFold, Genie, Veo, TPU chips

Google DeepMind is Alphabet's AI lab, formed in 2023 from DeepMind and Google Brain. It builds the Gemini models and leads in AI for science. AlphaFold won a share of the 2024 Nobel Prize in Chemistry.

Gemini 3 (Nov 2025) led many benchmarks. Gemini 3.5 Pro, announced in May 2026, is delayed.
AlphaFold predicted structures for over 200 million proteins and won a share of the 2024 Chemistry Nobel.
Google's own Ironwood TPU chip is used in-house and sold to outside labs, including Anthropic.

deepmind.google (opens deepmind.google)Company profile →

Founded: 201016 yrs
Headquarters: United KingdomUnited States
Status: Subsidiaryof Alphabet
Valuation: Alphabet-owned
Works in: AIFrontier modelsComputingBiotech
Coverage: 16 programs · 50 updateslatest 3d agochecked 10 Oct
Customers: Anthropic
Partners: Isomorphic LabsBroadcom
Acquired by: Alphabet
People: Demis HassabisCEO and co-founderShane LeggCo-founder and Chief AGI ScientistJohn JumperAlphaFold lead; 2024 Nobel laureate

Spec

Axion

C4A max vCPUs (bare metal)

96^R (reported)

C4A max memory (bare metal)

768 GB DDR5^R (reported)

C4A networking

Up to 100 Gbps^R (reported)

N4A max vCPUs

64^R (reported)

N4A price-performance vs x86

Up to 2x^R (reported)

N4A performance per watt vs x86

+80%^R (reported)

Energy efficiency vs x86 (2024 claim)

Up to 60% better^R (reported)