Nvidia raises the stakes with $1T revenue forecast and new Groq chip
What's the deal? Nvidia CEO Jensen HuangDealroom has a profile for this one. Try Dealroom → used the company's annual GTC developer conference in San Jose to forecast at least $1 trillion in AI chip revenue through 2027 — double its previous $500 billion projection through 2026.
He also unveiled a new chip system built on technology licensed from startup Groq, designed to make Nvidia's servers faster and cheaper for AI inference tasks.
Why now? The inference market — running AI models rather than training them — is rapidly growing in importance as tools like AnthropicDealroom has a profile for this one. Try Dealroom →'s Claude Code and AI agents gain widespread adoption. Rivals including Google and Cerebras have been winning inference business from Nvidia's longtime customers.
The Groq deal, struck in December 2025 for $20 billion, is Nvidia's direct response: Groq's chips, which embed memory directly on-chip, are optimised for exactly the fast, cheap inference Nvidia's GPUs have struggled to match.
The Groq 3 LPX system will ship in Q3 2026 as an add-on to Nvidia's Vera Rubin servers. It marks the first time Nvidia has integrated a third-party AI processor into its rack systems — a tacit admission that its GPUs alone aren't always the best fit for every workload. Notably, the chip will be manufactured by SamsungDealroom has a profile for this one. Try Dealroom →, not Nvidia's usual foundry partner TSMC, easing supply-chain pressure.
What could go wrong? Wall Street wasn't convinced. Nvidia's stock briefly rose 4.8% before closing up just 1.6% — the trillion-dollar forecast extending projections by a year rather than accelerating near-term growth. Analyst consensus for Nvidia's fiscal 2027 and 2028 revenues sits around $835 billion, well below the new target. Investors remain cautious about the sustainability of AI infrastructure spending, Middle East supply-chain risks, and tight high-bandwidth memory supply.
Competition is intensifying too. AMDDealroom has a profile for this one. Try Dealroom → is pushing harder into AI chips, while major customers including Meta and Amazon are developing in-house silicon. Nvidia's projected share of the inference chip market may be only around a third, compared to its 90%-plus dominance in AI training.
The signal: Nvidia's trillion-dollar bet reflects a broader shift in how AI value is being created. The centre of gravity is moving from building AI models to running them at scale — billions of daily inference requests demanding chips optimised for speed and cost, not just raw training power. By absorbing Groq and pushing into CPUs (Intel's home turf), Nvidia is expanding well beyond its GPU origins to defend its position across the entire AI compute stack. The company that built the AI era now has to evolve fast enough to stay on top of it.
Sources:
TechCrunch
Bloomberg
The Financial Times
The New York Times
The Information
Business Insider
CNBC
Image credit:
Nvidia
J.V.
