Foxconn sees a significant shortage of AI chips impacting server business

The Chairman of Hon Hai — known as Foxconn in the West — is concerned about a lack of supply of AI-ready chips for the server market.

Speaking at the company’s annual Lunar New Year dinner, Chairman Young Liu said he expects the AI chip shortage to persist in 2024, with recovery hinging on new fabs opening up and a stable geopolitical environment.

Liu also said he predicts a slight increase in Hon Hai’s performance year-over-year, which is being held back by a lack of chip supply. The company previously gave conservative and neutral expectations for 2024.

While Hon Hai is best known as Apple’s contract manufacturer for the iPhone, the firm manufactures PCs, components, and servers for a variety of companies.

Overall, it holds approximately 43% of the global server market share, according to Digitimes Research. Last year, Liu said that sales of AI servers are expected to hit $150 billion by 2027, which is larger than the predicted size of the entire server market for 2025.

In earnings calls, executives from Hon Hai said they expect the Cloud and Networking Products division to be one of its main growth drivers for 2024 because of the strong demand for AI servers. 

The company reported on Monday in Taiwan that it saw a significant rise in cloud and networking product revenue, but a decline in computing product revenue due to slower PC demand.

Boosting fab capacity

Taiwan Semiconductor (TSMC) is working on dramatically increasing its fab capacity, with a second fab in Japan in the pipeline, scheduled to open in late February, along with a fab to serve Europe’s automotive sector to be built in Germany.

Recently, TSMC became the world’s largest semiconductor maker by revenue, beating Intel and Samsung.     

Meanwhile, companies including OpenAI and Microsoft are developing their own custom chips for AI workloads to improve cost efficiency and processing capabilities for generative AI, addressing the limitations of general-purpose GPUs by optimizing for specific AI tasks.

Open AI’s Sam Altman has been reported to be negotiating with investors in the United Arab Emirates as well as TSMC to build fabs to specifically supply OpenAI’s semiconductor needs thus reducing dependency on Nvidia.

In its earnings in November, Nvidia and industry analysts flagged fab capacity as an ongoing concern for the company.

The increased demand for AI workloads in servers means that in 2024 servers will have a 17% increase in DRAM per unit, and a 13% increase in SSDs, Taiwan’s market research firm TrendForce recently wrote.

The demand for servers capable of sustaining AI workloads is only set to increase as Apple prepares a significant rollout of new generative AI features for later in 2024. 

In the US, the Biden administration is poised to allocate billions in CHIPS Act subsidies to Intel and TSMC in order to expedite the construction of new fabs. Intel recently said it was delaying a $20 billion fab in Ohio because of the slow distribution of funds and uncertainties on where the semiconductor market was headed.