Google escalates Nvidia rivalry with specialised versions of AI chips
AI generated
Alphabet’s Google is rolling out a broader array of artificial intelligence chips, aiming to capitalise on its rapid ascent as a challenger to Nvidia in a market worth hundreds of billions of dollars.
asserted
Google → escalate → dollars
The company is introducing two new variations of its tensor processing units, or TPUs, custom-made semiconductors that help develop and run AI software.
asserted
that → introduce → software
The latest chips will address the need for confidential computing and help models access a wider range of information when handling queries.
asserted
models → address → queries
More broadly, they are meant to serve an industry that is shifting towards the use of AI agents and confronting security challenges.
asserted
that → mean → challenges
The roll-out marks an acceleration in product development for Google’s chip unit.
asserted
out → mark → unit
The business once released a new AI processor every two years.
asserted
business → release → processor
In 2026, it put out two at the same time.
asserted
it → put → time
And the latest variations, due to be announced on Nov 9, will add additional flavours to that pair of chips.
asserted
variations → announce → chips
The four TPU offerings will optimise different needs, Google chip chief Amin Vahdat said in an interview.
asserted
Vahdat → optimise → interview
“This really follows the thesis that we have here at Google, which is that the future is going to be about specialisation, and the upsides for specialisation continue to be significant.
asserted
upsides → follow → specialisation
Google is staking its claim at a tumultuous time for the AI chip industry.
uncertain
Google → stake → industry
Concerns about a potential slowdown in infrastructure spending have roiled tech stocks, and US midterm elections that swept Democrats into power have raised fresh doubts about data centre projects.
asserted
that → roil → projects
But the demand for AI processors from Nvidia remains larger than it can satisfy, and that has left room for Google and other contenders to build a sizeable business.
asserted
Google → remain → business
The new TPU 8tx and 8i Max variations are not completely new silicon designs – rather, they are reconfigured to take on tasks that customers have been requesting, Vahdat said.
asserted
Vahdat → reconfigure → that
The 8tx is aimed at handling secure, confidential work for customers like government agencies and companies in regulated industries.
asserted
8tx → aim → industries
The system works with chips from Advanced Micro Devices, part of a collaboration between the two companies.
asserted
system → work → companies
The 8i Max, meanwhile, is meant to run AI applications that require long-context windows – the amount of data an AI system can consider in a single request.
asserted
system → mean → request
That is necessary for things like drug discovery and mathematics, but is also important for coding applications, Vahdat said.
asserted
Vahdat → cod → applications
TPUs compete in a market dominated by Nvidia’s graphics processing units, or GPUs.
asserted
TPUs → compete → units
Google set out to develop the technology more than a decade ago, alongside Broadcom, which helped with design and coordinating production.
asserted
which → set → production
The initial idea was to more efficiently handle the search giant’s burgeoning computing needs.
asserted
idea → handle → needs
It then realised the potential of TPUs for doing AI work and has now begun selling them outside its cloud-computing operation.
asserted
It → realise → operation
Google’s new life as a chip seller is proving lucrative.
asserted
life → prove → seller
By 2028, TPU sales will top US$135 billion (S$173 billion), including related software and hardware, according to Bloomberg Intelligence.
uncertain
sales → top → Intelligence
That is up from an estimated US$12 billion in 2026, and it does not include sales from clients using the chips through Google Cloud.
asserted
it → estimate → Cloud
TPUs have netted some high-profile clients, including Apple, Meta Platforms and Oracle.
asserted
TPUs → net → Apple
And offering specialised versions of the chips should help Google make further progress, Vahdat said.
asserted
Vahdat → offer → progress
The variations will be available in 2027, he said.
asserted
he → say → 2027
“We could build a single chip that tries to address the whole market or, by building specialised versions of the chips, we actually can deliver much more value,” he said.
uncertain
he → build → value
Even Nvidia, which has historically promoted its GPUs as a product for all AI tasks, is starting to offer more specialised chip systems.
asserted
which → promote → systems
That includes offering its own central processing units, or CPUs, a category pioneered by Intel.
asserted
That → include → Intel
Those processors do not handle the heavy-duty data crunching needed to create AI models, but they are useful for managing AI agents.
asserted
they → handle → agents
Nvidia is also offering chips more focused on the inference stage of AI – where already-trained AI models carry out tasks – using technology it acquired from start-up Groq.
asserted
it → offer → Groq
Google’s 8tx pairs with a CPU from AMD, a rival to both Nvidia and Intel.
asserted
8tx → pair → Nvidia
The technology allows data to be encrypted at the chip level, making it more secure.
asserted
it → allow → level
Normally, the information has to be unlocked in memory while a computer processes it, Vahdat said.
asserted
Vahdat → have → it
The 8i Max puts two TPU chips in a single package to deliver more computing power and memory.
asserted
Max → put → power
Part of the appeal of TPUs is their price for performance, said Mandeep Singh, an analyst at Bloomberg Intelligence.
asserted
Singh → say → Intelligence
They cost about 30% less than Nvidia’s GPUs but do a competitive job in terms of throughput – the amount of data that can be processed over a given time.
asserted
that → cost → time
The approach appeals to a variety of clients, including users in finance and other fields.
asserted
approach → appeal → finance
…and 35 more, not listed.
The Aporia
