Tech Newsround – AI Semiconductors

Samsung – Crunch time.

• Nvidia is expected to qualify its memory suppliers for the successor to Blackwell (Rubin) in Q1 26, which is a must-win for Samsung if it wants to maintain and extend the recovery it has seen in its share price.
• Samsung’s failure in the current generation (HBM3) cost Samsung dearly with a loss of 40% of its market capitalisation and red faces at the Samsung stand at every trade show where I have asked the HBM4 question.
• Several quarters ago, Samsung decided to abandon the current generation and put all of its efforts into getting HBM4 right, regaining Nvidia as a customer and retaking its crown as the king of memory.
• This obviously takes some time, and we are currently in the trough where there is no action on HBM3 and no news on HBM4, which will cause some to doubt Samsung’s ability to pull this off.
• I have postulated that Samsung has a very good chance of succeeding as its history points to a depth of corporate character and resilience that has allowed it to emerge from worse crises than this (Note 7).
• Nvidia has already signalled that Rubin will be launched in 2026 (presumably at GTC in Q1 26) by which time it will have had to figure out who will be supplying the silicon and equipment that is needed to build a system around Nvidia GPUs.
• This means that Samsung needs to qualify its HBM4 with Nvidia before GTC 2026 to be a supplier from launch.
• Speculation is already running high, and a three-way meeting between Jensen Huang, Samsung chairman Jae-yong Lee and SK group chairman Tae-won Chey has increased the expectation that Samsung will get it right this time.
• Samsung has already rallied 30% to KRW70,000 since I bought it, and I think that qualifying HBM4 with Nvidia should push it back to KRW90,000 or higher.
• I still think that there is a very good chance of success, and so I am sitting on my position in Samsung’s London-listed GDRs.

Cambricon – A rapidly growing speck.

• Cambricon is a home-grown Chinese AI chip maker that is seeing massive growth in the home market, but it is a rounding error in global terms, and I suspect it has no chance of winning any business outside of China.
• H1 2025 revenues / operating profit were RMB2.9bn ($406m) / RMB1bn ($140m) with the company reporting an expansion of revenues of 44x YoY.
• Despite the growth, Cambricon is a speck in the market with 3% of the Chinese market and around 0.3% globally.
• However, these numbers are an indication of how effective the Chinese state can be when it comes to encouraging local companies to help it fulfil its policy objectives.
• One of these objectives is to become fully independent from foreign technology so that Chinese technology development can never again be hindered, as it has been in silicon chip manufacturing.
• I think that there is every chance that this strategy will be successful, but it will come at the cost of the competitiveness of Chinese technology in the global market.
• DeepSeek put to bed once and for all the notion that the Chinese can’t develop cutting-edge AI, but what no one is paying attention to are the economics of developing and running Chinese AI on homegrown silicon.
• This is because in 2022, Cambricon had to end its relationship with TSMC and switch to SMIC, meaning that the most advanced node that it now has access to is the inefficient and expensive 7nm multipatterning technique that is also used by Huawei.
• This technique is now over 7 years old and several generations behind the latest nodes that are available at TSMC and Samsung, which is why I think that its chips are not globally competitive when it comes to economics.
• This makes the 34.4% operating margin reported in H1 2025 interesting reading as it implies that someone (either the state through some form of subsidisation or SMIC through lower prices) is allowing Cambricon to earn unrealistic margins on a globally uncompetitive product.
• In the long run, I think that neither Huawei nor Cambricon nor anyone else making AI chips in China are going to trouble Nvidia and its peers very much.
• This is because foreign companies are almost certain to lose the entirety of the Chinese market in time, and the Chinese are not going to win market share overseas.
• Hence, what they lose from China, they will make up elsewhere, which is why when Jensen bemoans the potential loss of China, I don’t worry too much.