NVIDIA invests another $20 billion; is the Groq deal the dawn of decentralized AI?

NVIDIA recently agreed to invest approximately $20 billion to acquire the core assets of AI chip startup Groq. This not only becomes one of NVIDIA’s largest deals in history but also once again confirms its consistent strategy of preemptively “acquiring” potential competitors before they grow into threats. Although the transaction is described as a non-exclusive licensing agreement rather than a direct acquisition, the market generally perceives its actual effect as equivalent to a deep integration.

Just three months ago, Groq completed a $750 million funding round, reaching a valuation of $6.9 billion, with its valuation soaring in a short period. The funding lineup included BlackRock, Samsung, Cisco, and 1789 Capital, which has close ties to former President Donald Trump. In this deal, Groq’s cloud computing business will be retained, while its core technology and key talent will join the NVIDIA ecosystem. Groq’s founder and CEO Jonathan Ross (one of the key designers of Google TPU) will join NVIDIA alongside several other executives, while Groq will continue to operate independently under its current CFO.

This move continues NVIDIA’s recent “license-first” strategy. In September, NVIDIA introduced the Enfabrica team and obtained its technology licensing for over $900 million. Through licensing rather than acquisition, NVIDIA is believed to aim at reducing antitrust review risks. Previously, NVIDIA’s plan to acquire Arm for $40 billion was halted due to regulatory pressure.

From a technical perspective, Groq’s Language Processing Unit (LPU) uses on-chip SRAM instead of external DRAM, offering significant advantages in real-time inference and low-latency scenarios. The company claims its energy efficiency can be improved by about 10 times. Although this architecture has certain limitations in model scale, after being integrated into NVIDIA, this technological approach is expected to be reevaluated and utilized within a broader AI ecosystem.

The timing of this deal is also quite meaningful. Recently, Google released the seventh-generation TPU “Ironwood” and launched the Gemini 3 model trained entirely on TPU, demonstrating strong performance across multiple benchmarks. NVIDIA subsequently publicly emphasized that its platform can run all mainstream AI models, which is interpreted by the market as a sign of increasing competitive pressure.

Although this deal has no direct impact on the crypto market, it once again reinforces the long-term narrative of decentralized AI and decentralized computing power. Decentralized AI computing projects, including io.net, are attempting to replace centralized AI infrastructure by integrating dispersed GPU resources. However, as NVIDIA continues to absorb low-latency and high-performance technologies, its technological moat deepens, making the performance and scale challenges for decentralized AI projects even more severe.

Overall, NVIDIA’s acquisition of Groq’s assets is not just a business transaction but may also accelerate the centralization of global AI computing power. Meanwhile, whether the survival space for independent AI chip companies will continue to shrink remains a long-term concern for the market.