TikTok owners reportedly working with Raspberry Pi chip maker on AI hardware
Raspberry Pi chip maker Broadcom is apparently working with TikTok’s parent company, ByteDance, on new AI chips. This comes from a Reuters report, with both sides refusing to comment.
ByteDance’s alleged partnership is to acquire a batch of 5-nanometer AI chips. These will then be implemented across the social media networks that ByteDance runs.
For instance, this could assist with the algorithm on TikTok or, as Reuters points out, run Doubao, its ChatGPT competitor.
These chips will be produced by TSMC, which the report states doesn’t plan on making until next year.
While known for its work with Raspberry Pi, US-based Broadcom develops a massive array of hardware outside the British company. However, working with a Chinese company like ByteDance can be like walking across a minefield these days.
On top of the looming TikTok ban, the two companies will have to work around US sanctions against a particular sect of hardware.
These sanctions were implemented to block China from accessing powerful parts to boost its artificial intelligence efforts. Part of this means having to work within the limitations set out by the Biden administration.
Nvidia has been affected the most, with its A and H-series AI chips blocked from being sold in China. The company’s flagship gaming card, the RTX 4090 was also affected.
Team Green then had to create the 4090D to skirt sanctions by stripping it of some of its potential.
ByteDance reportedly allotted $2 billion to acquire Nvidia hardware last year, according to the report. On top of “stockpiling” Nvidia hardware, Reuters mentions ByteDance has also purchased Huawei’s competing AI chip too.
While China is currently hampered by the US sanctions, hardware makers are repeatedly proving that they might not need America at all.
The hardware might not be neck-and-neck with Western chips, but Huawei claims its rival chip is already 80% of the way there and can even best Nvidia by 20% in certain use cases.
Source, dexerto, Read the full article here