Nvidia faces turbulence as Google Cloud unveils game-changing new AI chip

22 Apr 2025

Google Cloud diversifies away from third party hardware with a new Ironwood AI chip that could set the energy efficiency benchmark while Nvidia bleeds trillions on the frontlines of the global trade war.

Nvidia can’t catch a break right now. After ascending the AI hardware throne, increased sanction pressure forced the company to design a gimped version of its H100 GPU for Chinese customers, which DeepSeek then used to develop its AI model to show the world what can be achieved on cheaper hardware.

That caused the first catastrophic stock plunge. Then came a devastating couplet of circumstances just as the once hardware king was shaking off the tariff damage.

Google Cloud unveiled its seventh-generation Tensor Processing Unit (TPU), codenamed “Ironwood”, with claims of remarkable improvements in both performance and energy efficiency that could reshape the environmental footprint of artificial intelligence operations — a much-needed development after the recent International Energy Agency report.

Google’s announcement comes against the backdrop of Trumpian tariff flip-flopping where the US government imposed fresh restrictions requiring licences for the export of Nvidia’s H20 AI chips to China, resulting in a 6% plunge in Nvidia’s after-hours trading and erasing approximately R3-trillion in market capitalisation.

In a US Securities and Exchange Commission filing, Nvidia disclosed it expects “to record charges of up to approximately $5.5-billion” (R103-billion) in its first fiscal quarter ending 27 April 2025, associated with H20 products inventory, purchase commitments, and related reserves. The licensing requirement is stated to be “in effect for the indefinite future”.

Despite these challenges, Google Cloud is still a top five Nvidia customer and is reportedly in advanced discussions to lease Nvidia’s next-generation Blackwell (H200) chips, which just demonstrates the complex interdependencies in the AI infrastructure landscape.

Reaching into the bag of tricks

The new Ironwood chip, specifically designed for generating AI results at scale, achieves nearly 30 times greater power efficiency compared with Google’s first cloud TPU (tensor processing unit – optimised for AI tasks) from 2018. It also doubles the efficiency of its predecessor, the sixth generation “Trillium” TPU.

“Ironwood is the most powerful chip we have ever built and will enable the next frontier of AI models,” the Google Cloud team explained during a media briefing Daily Maverick attended before the public announcement. The chip delivers “significantly more capacity per watt for customer workloads”, addressing growing concerns about the energy consumption of increasingly sophisticated AI systems.

This energy efficiency gain comes at a critical time, as data centres currently account for approximately 1.5% of global electricity consumption — a figure that continues to grow rapidly as AI adoption accelerates.

Google says it’s “the only major cloud provider with a platform fully optimised for AI”. The Ironwood chip is a key part of Google’s larger “AI Hypercomputer” — a powerful, unified system that powers all AI tasks on Google Cloud.

A recent Forrester report named Google Cloud a leader in AI infrastructure, calling out its strong performance for all types of AI work, its advanced plans, and its innovation.

Ironwood TPUs have a fast, high-capacity connection system for communication between groups of chips. These liquid-cooled Ironwood chips can be scaled up to 9,200 units with fast connections and networking using almost 10MW of power.

What this means for you

Local access to cutting-edge AI tools: With Google’s new Johannesburg data centre, South African developers, businesses and researchers can now access the powerful new Ironwood TPU — a custom-built AI chip that’s faster and more energy efficient than its predecessors.

Faster, more reliable cloud performance: Data hosted in-country means quicker response times, improved performance, and better control over your data — ideal for industries like finance, healthcare and education.

More than just hardware — it’s a strategic shift: Google’s Ironwood TPU launch shows the company is reducing its reliance on Nvidia, which has been caught in the crossfire of global trade tensions. This makes Google Cloud more resilient, and its services more stable — good news for African clients looking for long-term infrastructure partners.

Designed with energy use in mind: Ironwood offers nearly 30 times the power efficiency of Google’s first AI chips, and is double that of the previous generation. That’s a big deal in South Africa, where load shedding and rising electricity costs are a constant challenge.

Benefits for researchers and students: South African universities are already tapping into Google’s TPU programme to give students access to world-class AI tools — sometimes even better than local high-performance computing resources.

A growing digital ecosystem: Google’s investment in submarine cables and local infrastructure like Cloud WAN is helping expand rural internet access and create a foundation for future AI jobs and innovation across the continent.

An African dream

Google’s recent launch of its Johannesburg data centre – the company’s first cloud region in Africa – takes on added significance in light of this hardware announcement. The facility promises to provide South African businesses and researchers with local access to energy-efficient AI infrastructure like the Ironwood TPU.

“I’m excited just for the country. I’m excited (about) the investments we’re putting in... I’m excited about what the opportunity is bringing,” said Tara Brady, president for Google Cloud in EMEA, speaking about the Johannesburg launch.

Alex Okosi, Google’s MD for Sub-Saharan Africa, shared his enthusiasm: “It’s just an amazing opportunity for us to be able to have the foundation to really enable AI and all the things that we want to be able to build in continental cloud.”

Read more: AI boom reshapes energy needs — and the world needs to be able to power it

Google’s AI infrastructure is also having an impact in South African universities, making advanced hardware accessible for research. Professor Jonathan Shock of the University of Cape Town (UCT) told Daily Maverick about the value for students: “So many of my students at the moment actually use the Google TPUs. Google actually has an amazing scheme where students can get six to nine months of free TPU access.”

He explained that this often surpassed UCT’s current high-performance computer (HPC): “Most of my grad students actually use that rather than the HPC currently just because having access to those Google TPUs is amazing.”

However, Shock highlighted scalability challenges this had created for the university HPC expansion plan. “I’ve spoken with Google, and essentially they can take on a relatively small number of these students, so having our own computer is going to be absolutely vital.”

Commitments to Africa

Alongside the Ironwood announcement, Google Cloud revealed several other infrastructure advancements, including the global availability of Cloud WAN, its worldwide private network designed to improve application performance while reducing costs. According to Google, Cloud WAN “leverages Google’s planet-scale network” and "delivers over 40% faster performance while reducing total cost of ownership by up to 40%”.

For Africa specifically, Google has invested in submarine cables like Equiano “to improve digital inclusivity at a country level, especially parts of the world that don’t have connectivity into basic internet”, according to Okosi. This infrastructure investment builds the planet-scale network out to extend connectivity into rural parts of South Africa and make it more affordable.

Read more: SA must learn from China’s digital white elephants and build a sustainable AI tech ecosystem

Looking ahead, Brady reminded the media of Google’s commitment to creating 300,000 jobs in South Africa in the coming years, across both technology and business skills. Brady sees the potential for South Africa to become a “genuine AI leader in the world” given its large population and the opportunity to upskill it.

As the trade tensions between the US and China continue to evolve, with US President Donald Trump ordering probes into potential new tariffs on critical minerals imports, companies with diversified AI hardware strategies may be better positioned to navigate an increasingly complex global landscape.

For South Africa, Google’s local data centre and continued investment in energy-efficient AI infrastructure represent a significant opportunity to develop sovereign AI capabilities while managing energy demands in a region facing ongoing electricity challenges. DM

Categories:

Tags:

DM is much better if you are signed in...

Business Maverick

Reaching into the bag of tricks

What this means for you

An African dream

Commitments to Africa