Not only NVIDIA's GPUs, but multiple links in the AI industry chain are facing shortages.

Not only NVIDIA's GPUs, but also supply chain bottlenecks in hardware devices are hindering the AI industry's progress towards greater prosperity.

According to Brady Wang, an analyst at consulting firm Counterpoint, "Both advanced packaging and high-bandwidth memory (HBM) suffer from insufficient capacity, which limits production."

TSMC plans to double the production capacity of CoWoS advanced packaging, but warns that this bottleneck will not be resolved until at least the end of 2024. HBM has two major suppliers, SK Hynix and Samsung from South Korea.

Driven by the demand sparked by the AI boom, NVIDIA has exceeded expectations, with second-quarter revenue doubling compared to the previous year and EPS profit increasing more than fourfold, significantly surpassing analyst expectations.

Three sources close to NVIDIA revealed that the company plans to at least double the production of the H100 graphics card. By 2024, the shipment volume of H100 is expected to reach 1.5 to 2 million units, a significant increase from this year's expected 500,000 units.

The fact that AI processors for 2024 are already sold out indicates that the high demand for NVIDIA chips is impacting the broader computing device market. Major buyers are investing heavily in AI servers at the expense of general-purpose servers.

Foxconn, the world's largest electronics contract manufacturer, predicted strong demand for AI servers in the coming years, but also warned that total server revenue will decline this year.

Similarly, Lenovo, the world's largest computer manufacturer, announced an 8% decline in second-quarter revenue last week, citing strong demand from cloud service providers (CSPs) for software servers and a shortage of AI servers.

Lenovo CEO Yang Yuanqing stated, "CSPs are shifting their demand from traditional computers to AI servers. Unfortunately, the supply of AI servers is limited by GPU shortages."

TSMC predicted last month that the demand for AI server chips will grow by nearly 50% annually over the next five years. However, the company stated that this growth is not enough to offset the downward pressure caused by the global economic downturn.

Data from Counterpoint Research shows that global CSP capital expenditure is expected to grow by only 8% this year, lower than the nearly 25% growth in 2022, as interest rates rise and companies cut spending.

TrendForce, an industry research firm, predicts a 6% decline in global server shipments this year, with a slight increase of 2% to 3% expected by 2024.