Поиск по этому блогу

Search1

123

пятница, 26 декабря 2025 г.

If You Can’t Beat Them, Buy Into Them

  -  
aim_new_logo-1-12
aim_new_logo-1-12

Sector6 
Dec 26, 2025

The bi-weekly newsletter by AIM that brings the biggest shifts shaping IT, AI, and GCCs.

If You Can't Beat Them, Buy Into Them

NVIDIA pivots from GPUs to custom chips with a $20 billion Groq deal, hiring the TPU's architect to win the high-stakes AI inference war.

By Supreeth Koundinya

Prabhu_Jensen-Huang-and-groq-story
Prabhu_Jensen-Huang-and-groq-story

While OpenAI led the narrative with product shipments in December 2024, NVIDIA chose Christmas Eve 2025 to drop a hardware bombshell. 

The chip giant has entered a non-exclusive licensing agreement with Groq, a firm that built its identity as the primary GPU alternative for AI inference. While Groq will continue to operate independently, CEO Jonathan Ross and President Sunny Madra will join NVIDIA. 

This move, along with access to Groq's technology, is expected to help NVIDIA scale its inference capabilities.

Solving the Inference Bottleneck

Media reports peg the deal at approximately $20 billion, nearly three times Groq's last reported valuation of $6.9 billion from its $750 million funding round in September. If accurate, this would rank as NVIDIA's largest deal to date.

Also Read: Top 13 Companies NVIDIA Invested in 2025

Jensen Huang, NVIDIA's CEO, has repeatedly argued that Inference will become the dominant AI workload, while admitting that it's challenging. 

GPUs remain unmatched for training. They have struggled to keep pace with custom accelerators in low-latency, high-throughput inference. 

Groq's solution is its Language Processing Unit (LPU). Ross has often contrasted it with GPUs by pointing to memory movement as the real bottleneck. Instead of relying on off-chip HBM or DRAM, Groq uses large on-chip SRAM to store model parameters close to the compute. 

This enables deterministic, statically scheduled execution, far lower latency, and very high bandwidth. Precisely the profile inference workloads demand.

Its LPU technology is accessible via GroqCloud, which hosts a variety of open-source models

From Training to Deployment 

By 2030, as much as 75% of AI workloads are expected to be inference, as AI shifts from experimentation to deployment. While GPUs will continue to be necessary for training, Ross has acknowledged that technologies like Groq's act as a "nitro boost" for inference while NVIDIA continues to sell every GPU it can produce.

Screenshot-2025-12-25-at-8
Screenshot-2025-12-25-at-8

Countering TPU Threat 

The deal appears to be a strategic counter to Google. Anthropic has extended its agreement to utilise up to a million TPUs, and Broadcom, in its Q4 25 earnings report, revealed a $21 billion order to supply Google's newest chips for the California-based AI company. Google's Gemini models are also trained and deployed on TPUs, reinforcing the case that custom silicon can compete at the frontier.

These developments also led to a dip in NVIDIA's stock last month, prompting the company to issue a statement reiterating its commitment to lead the industry. 

"NVIDIA is a generation ahead of the industry–it's the only platform that runs every AI model and does it everywhere computing is done," the company said in a social post. 

SemiAnalysis reports that Google's internal deployment costs for its newest TPUs are approximately 44% lower than those of comparable NVIDIA systems.

Even for external users, TPUs could deliver 30–40% lower total cost of ownership (TCO) than NVIDIA's GB200 and GB300-class GPUs. 

By hiring Ross, an ex-Google hardware veteran and a key architect behind the TPU, NVIDIA is looking to counter Google.  While Huang has generally been bearish on custom AI chips, he has consistently singled out Google's TPUs as a rare exception.

And by bringing Ross on board, Huang is now working with the person often referred to as the "father" of the TPU project, a technology he has always admired.

 

AIM Exclusive >>

By 2026, India aims to turn its deep-tech ambitions into reality through ₹1 lakh crore in sovereign RDI funding for AI, semiconductors, quantum computing, robotics, and advanced manufacturing. Additionally, over $67B in AI infrastructure investments from Amazon, Microsoft, and Google signal India's rise as a global build site, not just a talent pool.

video_preview_a9763a30e7aabbaef8d5ef8e6cdd20cb.jpg
video_preview_a9763a30e7aabbaef8d5ef8e6cdd20cb.jpg

Dell-NVIDIA Developer Meetup

Dell-x-NVIDIA-Developer-2
Dell-x-NVIDIA-Developer-2

As AI moves from experimentation to real-world deployment, developers are increasingly grappling with practical questions around infrastructure, performance and workflows. An invite-only Dell x NVIDIA Developer Meetup, in association with AIM, on January 17, 2026 in Bengaluru, will bring together AI engineers, data scientists, enterprise teams, and leaders from Dell and NVIDIA to share applied perspectives on building beyond proofs of concept. Click here to register. 


AWS AI Conclave 2026 in Bengaluru

Mohit_2026-The-Year-Software-Engineering-Will-Become-AI-Native
Mohit_2026-The-Year-Software-Engineering-Will-Become-AI-Native

Amazon Web Services is gearing up to host the AWS AI Conclave 2026 on January 22 at the Sheraton Grand, Whitefield, Bengaluru, bringing together the brightest minds shaping the next era of AI. 

This edition will spotlight breakthroughs in agentic AI, autonomous systems, data strategies and enterprise-scale AI adoption, offering a front-row seat to the technologies redefining global innovation. Click Here to Register Now.

Simulated Reality >>

Join us for an exclusive interview with Andy Logani, Executive Vice President and Chief Digital and AI Officer at EXL, a global data and AI company. Andy details EXL's incredible 25-year journey from a business process operations company to a data and AI powerhouse, with 55% of its revenue now stemming from Data and AI.

video_preview_759628259189fe29cb081f7c390f9a85.jpg
video_preview_759628259189fe29cb081f7c390f9a85.jpg

Best Firm for Data Scientists – Showcase your leadership in data science and build a thriving workplace. Apply Now

Join AIM Leaders Council – Connect with top data leaders and shape the future of AI. Check Eligibility

Our Events – Be part of an exclusive event uniting North America's top data leaders. More Details

AIM PeMa Quadrant – Discover the best in AI and data services with AIM's PeMa Quadrant. Explore Now

GCC Explorer Unlock Exclusive Insights into 1400+ Global Capability Centers (GCCs) in India with AIM Research's Comprehensive Database.  Access GCC Explorer

 

Stay ahead with these insights, and join us in driving the future of data science and technology. Your active engagement and participation are what make our community thrive.

 

For Brand Collaborations, Contact us at info@aim.media

To unsubscribe from future emails, simply click the "unsubscribe" link AIM. Unsubscribe

© Copyright, 2025, AIM • 1st Floor, Sakti Statesman, Marathahalli – Sarjapur Outer Ring Rd, Green Glen Layout, Bellandur, Bengaluru – 560103

AIM-logo-black
AIM-logo-black
  -  

Комментариев нет:

Отправить комментарий

Примечание. Отправлять комментарии могут только участники этого блога.