 (3).png)
1 month ago
6
ARTICLE AD BOX
Nvidia, the world’s most valuable company, launched two new chips on Monday. At its launch, chief Jensen Huang told a packed stadium that the chipmaker was changing strategy: after over three years of doubling down on training artificial intelligence, its new chips will now be optimized to run AI. Does this make any difference? Mint explains.
How are Nvidia’s new chips any different?
Nvidia launched a new generation of its Vera Rubin chips, and an all-new Nvidia Groq 3 chip—designed after it signed a $20-billion partnership deal with the founder of Google’s custom chip.
These focus on power efficiency, thereby significantly reducing the cost of AI. Huang referred to ‘inferencing’ or the cost of running AI, which he said will fall drastically when these new chips hit data centres in three years.
The new chips, for the first time, took the spotlight away from building AI and onto running it, suggesting that Nvidia sees AI at a stage of mass adoption.
Why is Nvidia doubling down on these new chips now?
Training AI is becoming templatized, while running AI will bring the bulk of the demand for specialized chips. As adoption of AI grows, super-efficient chips that can process trillions of data points will be the primary demand.
The current chips consume a lot of energy, and are thus unsustainable in the long run when AI becomes as ubiquitous as smartphones today. It is this that pushed Nvidia to focus on efficiency rather than performance this year, so it is ready for AI’s next wave. Being ready is important, as Nvidia supplies nearly 90% of the world’s AI chips.
Are any other chipmakers challenging Nvidia?
Yes. On Wednesday, AMD signed a partnership with Samsung to jointly develop AI-ready chips similar to Nvidia’s Groq 3. AMD also partnered with OpenAI in October last year to supply chips for 6GW of computing power, while on 24 February, Meta also signed a long-term partnership with AMD to use the latter’s chips. Intel, on 25 February, signed a usage partnership with niche chipmaker SambaNova to use the latter’s specialized chips.
Can this strategy shift impact India?
India is an important market for AI chips, with data centres set to grow to 16GW in six years. With most data centres using Nvidia’s older chips so far, this means that, sooner or later, frugal adoption of AI in India, using India’s data centres, would require new AI chips—rather than the older ones currently in use.
But the new chips are likely to be more expensive, so adoption could stall until companies see clear returns. Overall, though, India’s AI expenditure is likely to increase, in turn raising its contribution to global chipmaker revenues.
Is Nvidia’s trillion-dollar revenue target realistic?
Nvidia, as of 25 February, has marked 11 consecutive quarters where its operating revenue has grown 55% or more, year-on-year. For fiscal 2025, which ended on 31 January for Nvidia, the chipmaker earned $216 billion for the full year.
At the company’s annual GTC, Huang said Nvidia expects to generate at least $1 trillion from the sale of just its established AI chips within two fiscal years. The company has also projected $500 billion in revenue from chip sales by end-2026. Given its growth pace, market share and surging AI demand, analysts believe the target is likely.
Will AI chips become even rarer now?
There’s a good chance, but Nvidia has played its part to try and diversify its supply chain. The Vera Rubin chips, which also include general-purpose CPUs alongside AI-specialized GPUs, are made by Taiwan’s TSMC, the world’s largest chip fabricator.
The Groq 3 chips, however, will be made by Samsung Semiconductor, which helps prevent Nvidia from putting all its eggs in one basket. Most chipmakers are increasing their fabrication capacities, but making new fabs take at least three years.
At the same time, Nvidia expects demand for AI chips to accelerate further as companies adopt AI, which can crunch supplies despite diversification.
English (US) ·