Nvidia's Laptop Gambit Meets the Memory Wall

Nvidia's Blackwell GPU now fits inside a laptop. Jensen Huang took the stage at Computex in Taipei on Monday to reveal the RTX Spark, a high-end mobile processor that combines an Arm-based CPU co-designed with MediaTek with a Blackwell based GPU , marking the company's first serious push into personal computing. At full strength, this chip offers up to 20 Arm CPU cores, a Blackwell GPU with 6,144 CUDA cores, 128GB of LPDDR5X RAM, and up to 300 GB/s of memory bandwidth , according to Tom's Hardware. The chip will debut in the fall on a fresh line of Windows PCs from Microsoft, Dell, HP, ASUS, Lenovo and MSI , CNBC reported.

The timing is deliberate. South Korea's exports grew at the strongest annual rate in more than four decades in May, beating market expectations, as chip sales hit a record on a global boom in AI investment , Reuters reported Monday. Exports of semiconductors jumped 169.4% to a record monthly high of $37.16 billion —nearly all of it memory chips bound for data centers. Samsung and SK Hynix, the two companies that control the high-bandwidth memory market, are printing money. But the same shortage fueling their profits is now threatening to choke the AI buildout itself.

Can You Build an AI Laptop When There's No Memory Left?

The RTX Spark announcement is impressive on paper. Nvidia says users can "render ultra-large 90GB 3D scenes with OptiX and DLSS, edit 12K 4:2:2 video with the NVIDIA Blackwell decoder, run 120-billion-parameter large language models with 1 million tokens context, and play AAA games at 1440p resolution and over 100 frames per second with ray tracing" , according to MacRumors. It's the kind of spec sheet that makes Apple's M5 look pedestrian.

But there's a problem: memory. Data centers now consume an estimated 70% of all memory chips produced worldwide , according to industry analysts. The voracious demand for HBM by hyperscalers, such as Microsoft, Google, Meta and Amazon, has forced the three biggest memory manufacturers to pivot their limited cleanroom space and capital expenditure towards higher margin enterprise-grade components. This is a zero-sum game: every wafer allocated to an HBM stack for an Nvidia GPU is a wafer denied to the LPDDR5X module of a mid-range smartphone or the SSD of a consumer laptop , IDC noted.

The economics are brutal. A single silicon wafer provides 3x as much commodity DRAM as HBM. Fab processing time for HBM is significantly longer too, making the supply problem worse. As a result, producing more HBM equates to fewer total memory chips produced , Fortune reported. Nvidia's B300 GPU requires eight HBM chips, each containing 12 individual DRAM dies. That means a single B300 GPU consumes 96 DRAM dies—and a fully configured DGX B300 system with eight GPUs requires 768 DRAM dies just for the HBM modules alone .

TrendForce this week said it expects average DRAM memory prices to rise between 50% and 55% this quarter versus the fourth quarter of 2025. TrendForce analyst Tom Hsu told CNBC that type of increase for memory prices was "unprecedented" . Lenovo, Dell, HP, Acer and ASUS have warned clients of tougher conditions ahead, confirming 15-20% hikes and contract resets as an industry-wide response , according to IDC.

How Long Until the Shortage Breaks?

Not soon. Sassine Ghazi, CEO of Synopsys, a key semiconductor design tool company, told CNBC in an interview last week that the chip "crunch" will continue through 2026 and 2027 . Google DeepMind's Demis Hassabis called it a "choke point" for the industry , Bloomberg reported. On Tesla's earnings call in late January, Chief Executive Officer Elon Musk even raised the idea of producing his own memory chips .

The supply response is underway but slow. Micron is currently building two big factories in Boise, Idaho, that will start producing memory in 2027 and 2028, and is also going to break ground on a fab in the town of Clay, New York, that is expected to come online in 2030 , the company told CNBC. Even if the major memory manufacturers were to begin breaking ground on new commodity-DRAM-focused fabs today, the 18-to-24-month construction timeline plus the additional ramp-up period for yield optimization places volume production firmly in 2028. And because the 3-to-1 dynamic means each AI chip produced today continues to suppress commodity output, the supply gap is actively widening through 2026 and 2027 rather than stabilizing , according to industry analysis.

Meanwhile, the power crunch is getting worse. Driven by data centre investments, the capital expenditure of five large technology companies surged to more than $400 billion in 2025 and is set to increase by a further 75% in 2026. Electricity demand from data centres soared by 17% in 2025, and that of AI-focused data centres climbed even faster , the IEA reported in April. Tesla Inc. and SpaceX CEO Elon Musk said earlier this year, "Very soon, maybe even later this year, we'll be producing more chips than we can turn on" , Bloomberg noted.

The constraint isn't just memory—it's the entire stack. Following the 2026 strikes on Qatari production—which accounts for one-third of global supply—spot prices for helium have doubled. Fabs in Taiwan and South Korea are now rationing helium. This net supply shortage could lead to a potential reduction in chip production , Manufacturing Dive reported. Bromine, essential for etching circuits and flame retardancy, has surged to $12,000 per metric ton. With ICL Group in Israel controlling nearly 40% of the global supply, and providing 97% of South Korea's imports, geopolitical instability has turned this once-common mineral into a strategic liability .

What Changed This Week

Nvidia crossed the Rubicon into consumer PCs with a chip powerful enough to run frontier AI models locally. South Korea's semiconductor exports hit a four-decade high, confirming the AI infrastructure boom shows no signs of slowing. But the memory shortage that was a footnote six months ago is now the defining constraint on the entire industry, with relief pushed firmly into 2028. The gap between what AI companies want to build and what the supply chain can deliver is widening, not closing.

What to Watch

RTX Spark laptops are scheduled to ship in fall 2026—watch for pricing announcements and whether Nvidia can secure enough memory allocation to meet demand. Nvidia Rubin is in full production, and Rubin-based products will be available from partners the second half of 2026 , the company announced. South Korea will report June trade data around July 1, offering the next read on whether semiconductor export growth is accelerating or plateauing. And keep an eye on DRAM spot prices: if they spike above 60% quarter-over-quarter, expect more consumer electronics makers to delay launches or cut specs.