Nvidia Strikes $20 B Groq Inference Tech Licensing & Talent Deal

Big chip houses have scooped up specialised rivals before—Intel’s 2015 $16.7 b purchase of Altera or AMD’s 2020 $35 b Xilinx deal—but almost always through outright M&A that later drew regulatory heat. After regulators scuttled Nvidia’s $40 b bid for Arm in 2022, the company is now emulating Facebook’s 2012 ‘acquihire’ of Instagram and Google’s steady AI-startup licensing sprees: get the IP and brains, leave the shell behind, and dodge antitrust thresholds. The move underscores two structural shifts: (1) the profit centre of generative AI is tilting from energy-hungry model training to latency-sensitive inference, and (2) cash-rich platform firms can use licensing plus talent migration to entrench dominance while regulators focus on formal mergers. If successful, Nvidia extends its 2020s GPU hegemony into the inference epoch much as IBM’s 1964 System/360 locked in mainframe standards for a generation. Over a 100-year arc, the pattern rhymes with Standard Oil’s early-1900s vertical lock-ins: concentration, followed by regulatory break-up or technological leapfrog. Whether this deal cements a monopoly or triggers the next round of antitrust confrontations will shape how open—or closed—the AI hardware stack becomes for decades.