Daily: Nvidia beats expectations; China unhappy at TSMC/Intel; SK hynix yield rate
7.5 min read.
Housekeeping: Apologies for missing yesterday; I got caught up with other work. The newsletter is a bit long, catching up on things.
Highlights
Nvidia earnings report. Nvidia beat expectations yet again, with revenue rising 78% year-on-year and net profit rising 80% to US$22 billion. This is slightly lower than previous quarters when revenue and profits rose by 100+%, but it’s natural that pace of growth slows down as time goes on. Nvidia stock is roughly flat and down 5% for 2025 so far.
Elsewhere, Nvidia is expected to release their next chip, Blackwell Ultra in the second half of 2026, with new memory, new networking, etc.
China unhappy at TSMC/Intel. In a press conference in Beijing, a spokesperson for China's Taiwan Affairs Office said that Taiwan was giving away Taiwan’s semiconductor industry to the US as a “souvenir.” Last week, US media reported that TSMC might acquire Intel Foundry. This hasn’t been confirmed by TSMC, Intel, or the Taiwanese government. Taiwan’s Economy Minister said today that TSMC needs government permission for overseas ventures. TSMC is also prohibited from producing the latest-edge chips overseas by Taiwanese law, so not sure how Intel’s 18A processes would fit into the picture.
SK Hynix HBM yield rate. SK hynix reaches 70% yield rate for their 12-layer HBM4 testing ahead of mass production, which is up from 60% at the end of 2024. SK hynix is racing ahead.
Thanks for reading.
1. Policy and Geopolitics
1.1
WSJ (02/25): The AI Data-Center Boom Is a Job-Creation Bust
In Abilene, Texas, some 1,500 people are building the first data center for the Stargate artificial-intelligence venture led by OpenAI.
Once it is completed, a lot fewer people will work there. The facility will have about 100 full-time employees, according to the city’s economic development agency. That total is a fraction of the number of people who might work on the same one million square feet if it were an office park, factory or warehouse. A 286,500-square-foot cheese-packaging plant that broke ground in Abilene in 2021 was projected to employ 500 people.
Politicians and business leaders have touted data centers as a boon for employment.
The reality is data centers can employ more than 1,000 people in the several months or years it takes to build them, but rarely need more than one or two hundred once they open, according to Synergy chief analyst John Dinsdale. Stargate would have to be much larger than currently planned to create hundreds of thousands of construction jobs, let alone permanent ones.
1.2
Reuters (02/27): China says Taiwan seeks to give away chip industry to US
China said on Wednesday that Taiwan was seeking to give away the island's semiconductor industry to the United States as a "souvenir" and leverage it to seek political support from Washington.
U.S. media have said TSMC has been in talks for a stake in Intel.
Neither TSMC nor Intel has confirmed the reports and Taiwan's government says it has not received information about any overseas investment application from TSMC.
1.3
Reuters (02/27): TSMC needs government permission for overseas joint ventures, Taiwan minister says
Taiwan Economy Minister Kuo Jyh-huei said on Thursday that chipmaker TSMC would need government permission for any overseas joint ventures, but the government will not interfere in its decisions.
1.4
TrendForce (02/27): China’s Low-Cost SiC and Mature Chips Ignite Global Semiconductor Price War
China is rapidly expanding its production capacity in the mature semiconductor and niche substrate sectors, pushing prices to unprecedented lows and exerting significant pressure on the global chip industry.
Two years ago, Wolfspeed’s mainstream 6-inch silicon carbide (SiC) wafers were priced at USD 1,500 per piece. However, Chinese suppliers are now offering them for as low as USD 500 per wafer, or even less.
Notably, China has been expanding its SiC wafer production, largely due to the availability of production equipment that mostly falls outside U.S. export controls, along with state-backed subsidies, the report indicates.
TrendForce highlighted that as new capacity from China comes online, Chinese foundries’ share of mature process capacity among the top 10 is expected to exceed 25% by the end of 2025. The highest increases will be seen in 28/22 nm production. Chinese foundries are also advancing their specialty process technologies, with the HV platform in particular expected to lead, and 28 nm already in mass production as of 2024.
2. Economy, Finance, and Business
2.1
NYT (02/26): Nvidia’s Profit Jumps 80% as Company Rides Tech’s A.I. Boom
When Nvidia lost $600 billion in market value in a single day last month, it was because some investors feared for the future of the artificial intelligence chipmaker. DeepSeek, a Chinese start-up, said it had made its A.I. systems with a small fraction of the A.I. chips used by other companies, and at a small fraction of the cost.
On Wednesday, Nvidia showed those fears were overblown, even as the breakneck pace of its growth slows. The company, a bellwether for A.I., said purchases of its A.I. chips lifted its total revenue by 78 percent from a year earlier to $39.33 billion during the three months that ended in January. Profit rose 80 percent to $22.09 billion.
In previous quarters, Nvidia reported that sales and profits had more than doubled. But continuing to deliver those kinds of gains has become more difficult as its sales and profits rise.
Shares in Nvidia were almost flat in after-hours trading, after a 3.7 percent gain on Wednesday.
2.2
Bloomberg (02/25): Nvidia, Cisco Extend Partnership Aimed at Promoting AI
Nvidia Corp., the top provider of chips used in new artificial intelligence computers, is extending a partnership with networking-gear maker Cisco Systems Inc. in a push aimed at making it easier for corporations to deploy AI systems.
Many businesses remain in the early stages of adopting AI systems because of the complexity the shift adds to their data centers, Cisco and Nvidia said Tuesday in a joint statement. The two companies are broadening the list of products that include each others’ technology in an attempt to remove those hurdles.
Nvidia offers networking add-ons to its AI server machinery, called Spectrum-X Ethernet. That product line will start including Cisco chips that handle broader connectivity functions, the two companies said.
2.3
Reuters (02/27): Synopsys revenue forecast beats estimates on AI chip boom, shares up
Synopsys on Wednesday forecast second-quarter revenue above Wall Street estimates, citing growing demand for its software used in designing chips.
The company's shares rose 2.45% to $482 in after-hours trading after the results were released.
The company sees second-quarter revenue between $1.59 billion and $1.62 billion, the midpoint of which is slightly above analysts' estimates of $1.60 billion, according to data compiled by LSEG.
2.4
WSJ (02/25): ASM International Orders Miss Forecasts Amid Weak China Demand
ASM International posted orders below analysts’ expectations for the fourth quarter amid weak demand for chip-making equipment in China, a market increasingly affected by U.S. export restrictions as Washington seeks to curb Beijing’s technological ambitions.
The Dutch group booked 731.4 million euros ($765.7 million) in orders, up 8% on year at constant currencies. Analysts had forecast €786.61 million in orders, according to consensus estimates by Visible Alpha.
ASM International provides tools–mostly for the deposition of thin films—that chip makers need to produce increasingly powerful semiconductors as demand for smaller but more efficient chips to power artificial intelligence is on the rise.
2.5
WSJ (02/25): Onsemi to Cut 9% of Workforce Amid Restructuring
Onsemi is planning a 9% cut to its workforce as it tries to maintain its pace of innovation amid a drop in demand and falling revenue.
The company said in a filing Tuesday it plans to cut 2,400 employees. Onsemi had 26,400 regular employees as of mid-February, according to an annual filing. The cuts are expected to save $105 million to $115 million annually once completed. Charges related to the layoffs are anticipated to cost around $50 million to $60 million and be recorded in 2025.
Onsemi’s silicon carbide devices have beaten Chinese companies’ products in terms of quality, but the Scottsdale, Ariz., company’s challenge will be to keep up that level of performance as rivals develop, Benchmark analyst David Williams said.
3. Technology
3.1
The Verge (02/27): Nvidia’s next AI chip, Blackwell Ultra, will be unveiled next month
Nvidia is hosting its GTC keynote on March 18th, and its keynote speaker has just revealed his talk. “Come to GTC and I’ll talk to you about Blackwell Ultra, Vera Rubin, and then show you the one click after that,” Nvidia CEO Jensen Huang told analysts on the fiscal Q4 2025 earnings call.
He says Blackwell Ultra will come in the second half of next year, with new networking, new memory, new processors, but on the same system architecture as Blackwell.
3.2
TrendForce (02/27): SK hynix Fast-Tracks HBM4 Development, Reportedly Hits 70% Yield in 12-Layer Testing
While HBM4 has become the next battleground for memory giants, SK hynix has hit a key milestone, as it has reportedly achieved a 70% yield in 12-layer HBM4 testing ahead of mass production, according to South Korean media outlet etnews.
Citing an insider familiar with the situation, the report indicates that at the end of 2024, SK hynix reached a 60% yield in HBM4 test already, and the company is improving yields at an impressive pace.
3.3
TrendForce (02/26): Micron Races Ahead in 10nm-class DRAM with 1γ DDR5 Samples Delivered to Intel & AMD
As memory giants ramp up their efforts on 10nm-class DRAM, Micron seems to be gaining an advantage, as it announces the shipment of 1γ (1-gamma) DRAM. According to its press release, the company has delivered 1γ (gamma) DDR5 samples to customers like Intel and AMD.
Micron says its 1γ DRAM builds on its 1α and 1β leadership, and could further drive the innovation for cloud, industrial, consumer, and Edge AI devices like AI PCs, smartphones, and automobiles.
3.4
Reuters (02/27): Startup PsiQuantum says it is making millions of quantum computing chips
PsiQuantum said on Wednesday it had cracked one of the crucial challenges in quantum computing: a method to manufacture quantum chips at the volumes needed to build commercially viable machines.
PsiQuantum's founders decided almost 20 years ago to pursue an approach that could be made using photonics, the same semiconductor manufacturing technology widely used in the communications industry.
The company has partnered with GlobalFoundries (GFS.O), opens new tab to fabricate the chips at the latter's Albany, New York, factory.
PsiQuantum's method uses chip industry-standard foot-wide wafers on GlobalFoundries' 45-nanometer process and has achieved manufacturing yields that match standard semiconductors, according to PsiQuantum's Chief Scientific Officer Pete Shadboldt.