Mr Viral
Groq is a leader in AI inference and today announced its first AI infrastructure in Asia-Pacific, selecting Equinix’s data center in Sydney, Australia as the location of choice. The development is part of its continued global data center network expansion, following launches in the U.S. and Europe.
Groq’s investment brings fast, low-cost and scalable AI inference closer to organisations and the public sector across Australia.
Under this partnership, Groq and Equinix will establish one of the largest high-speed AI inference infrastructure sites in the country with a 4.5MW Groq facility in Sydney, offering up to 5x faster and lower cost compute power than traditional GPUs and hyperscaler clouds.
Leveraging Equinix Fabric, a software-defined interconnection service, organizations in Asia-Pacific will benefit from secure, low-latency, high-speed interconnectivity, ensuring seamless access to GroqCloud for production AI workloads, ensuring full control, compliance and data sovereignty.
If you’ve ever used an LLM and seen ‘thinking’ this can be a result of inference resources being limited. By expanding inference capacity in this region, users are likely to get responses faster, resulting in a better experience, increasing the likelihood they’ll come back.
– Advertisement –
Groq is already working with customers across Australia, including Canva, to deliver inference solutions tailored to their business needs, from enhancing customer experiences to improving employee productivity. This momentum reflects growing demand for Groq’s LPU systems across Asia-Pacific, where half of GroqCloud’s two million users are based.
“The world doesn’t have enough compute for everyone to build AI. That’s why Groq and Equinix are expanding access, starting in Australia.”
Jonathan Ross, CEO and Founder, Groq
In the world of AI, one company has dominated every conversation: NVIDIA. Their powerful GPUs (Graphics Processing Units) have been the undisputed kings for training the massive AI models we all use, like ChatGPT, Claude, and Llama.
But training a model and using a model are two very different things.
Training is a one-time, monstrously complex task that takes months and billions of dollars. Using the model, or “inference,” happens millions of times every second, every time you ask a question.
Groq decided not to compete with NVIDIA on training. Instead, they built a new kind of chip from the ground up, designed for one thing and one thing only: making AI inference mind-bogglingly fast.
They call it the LPU, or Language Processing Unit.
Think of it this way: a GPU is like a massive, general-purpose factory that can build an entire car, from engine to paint. An LPU is a hyper-specialised, lightning-fast assembly line that just puts the wheels on, but it can do it a thousand times faster than the general-purpose factory.
The result is AI that doesn’t “think.” It just responds. Groq’s demos show models spitting out hundreds of tokens (words) per second, an order of magnitude faster than what most of us are used to.
“Groq is a pioneer in AI inference, and we’re delighted they’re rapidly scaling their high-performance infrastructure globally through Equinix. Our unique ecosystems and wide global footprint continue to serve as a connectivity gateway to their customers and enable efficient enterprise AI workflows at scale.”
Cyrus Adaggra, President, Asia-Pacific, Equinix
This new “Groq Cloud” node is expected to be operational in the coming months. Given that the Asia-Pacific region includes locations like China, Singapore and others, this is a massive vote of confidence in the Australian tech market.
“We’re entering a new era where technology has the potential to massively accelerate human creativity. With Australia’s growing strength in AI and compute infrastructure, we’re looking forward to continuing to empower more than 260 million people to bring their ideas to life in entirely new ways.”
Cliff Obrecht, Co-Founder and COO of Canva
Now let’s talk latency.
Right now, when you use most advanced AI models, your request travels from your computer, under the ocean via cables, to a massive data centre in the United States, gets processed, and then travels all the way back. That round-trip time, adds significant lag to the process. When this is text, it’s not a massive issue, but as we move to more-real-time multimedia applications, latency can be significant.
By having Groq’s hardware physically located in Sydney, the network latency disappears. For developers building AI applications in Australia, this is a game-changer.
This could unlock a new wave of real-time AI applications that simply weren’t possible before. Think AI-powered customer service agents that can interrupt and respond naturally, real-time language translation without the awkward pauses, or developer tools that suggest code instantly.
The sovereign AI advantage
The second piece of the puzzle is the word “sovereign.”
Businesses that play with sensitive data like finance, healthcare, and defence sectors, these organisations are often restricted from sending data outside of Australia. This has meant they’ve been locked out of using many of the world’s best AI tools, which are hosted in the US.
By hosting this high-performance AI hardware on Australian soil, Groq is giving these critical industries a way to tap into next-generation AI without compromising on data security or sovereignty.
For more information, head to https://groq.com/
GIPHY App Key not set. Please check settings