The world’s leading AI chip manufacturer delivered first-quarter earnings that surpassed Wall Street forecasts, yet questions emerge about obstacles that could slow the sector’s explosive expansion. Nvidia reported strong growth, with quarterly revenue up 85% year over year to $81.6 billion. Net profit more than tripled to $58.3 billion. The company also expects sales of about $91 billion in the current quarter. Big tech spending backs up the strong demand picture. The four major cloud computing companies plan to invest roughly $700 billion combined in 2026 on infrastructure, a jump exceeding 60% from last year’s already record-breaking amounts. Chief executive Jensen Huang pegs the total market opportunity for the company’s Blackwell and Vera Rubin chip lines at $1 trillion running through 2027. But three major roadblocks stand between current momentum and continued growth: trade restrictions, infrastructure limits, and borrowing costs. China market vanishes China used to make up about 20% of the chipmaker’s data center revenue, but that has now fallen to zero. The company reported no sales from China last quarter and expects none this quarter either. The drop is due to changing trade rules. H20 chip sales to China were first banned in April 2025, then later allowed again in July. In December, l imited H200 exports were approved , but with a 25% revenue share required for the U.S. government. However, Chinese customs stopped the shipments soon after. Even though ten major Chinese tech firms, including Alibaba, Tencent, and ByteDance, were approved to buy large quantities of H200 chips, none of the deliveries actually happened. In the end, China still has not approved the imports because it wants to focus on its own chip companies instead. At the current company size, losing one-fifth of business equals roughly $38 billion annually. Huang himself estimates China’s total AI chip market at $50 billion. Meanwhile, Chinese buyers are building their operations around Huawei’s competing Ascend chips. “By effectively excluding China and conceding that market to Huawei, Nvidia is demonstrating that global AI demand outside China is more than enough to sustain its growth,” said Alvin Nguyen, senior analyst at Forrester. Power grid becomes the limiting factor Physical limits are now as big a problem for data center growth as chip shortages. The main issue is power infrastructure, especially long delays in connecting large facilities to the grid. While data centers take 12 to 24 months to build, getting high-capacity grid connections can take 3 to 7 years. The U.S. grid queue is now over 2,600 gigawatts. Of the 12 gigawatts of U.S. AI data center capacity planned for 2026, only about 5 gigawatts are under construction, with the rest delayed due to power shortages and transformer delays that can take up to four years. To bypass this, companies like xAI, Meta, OpenAI, and Oracle are building their own power systems, now totaling over 130 gigawatts in the U.S., even though it is more expensive than grid power. Jensen Huang also said that supply limits like ASML machines and TSMC wafer production could be resolved in the next two to three years. Gavin Baker, founder of the hedge fund Atreides Management, gave a bold view on the situation. He said that if TSMC followed what Jensen Huang wanted, then Nvidia could potentially reach $2 trillion in GPU sales in 2026 or 2027. Borrowing binge raises concerns The last constraint comes from debt markets. Big tech companies raised $121 billion in U.S. corporate bonds in 2025, over four times their usual average. Analysts at Bank of America expect this could rise to $175 billion in 2026, especially after Amazon issued a record $54 billion global bond sale in March. This heavy borrowing competes with government and other corporate debt, increasing overall supply and making capital more expensive for tech firms. Looking ahead, competition is expected to intensify. John Blank, chief equity strategist at Zacks, noted buyers may soon pursue “anti-Nvidia” strategies to capture those profit margins themselves. Purpose-built chips from Broadcom and Marvell are gaining ground in inference workloads, where power efficiency matters more than raw performance. The smartest crypto minds already read our newsletter. Want in? Join them .
