Daily: Meta-Google's multi-billion dollar deal
7 min read.
Highlights
US allows H200 shipment. The US government has granted a licence to Nvidia to sell a small number of H200 chips to China. The licence is still conditioned on a security inspection on US soil and a 25% tax. There is also no news on whether Beijing will allow this shipment. I maintain my impulse that the Chinese government wants to limit to as few American chips as possible, especially if the chips are subject to opaque inspections that could feasibly add location-trackers and other features that Beijing doesn’t want.
China’s semiconductor self-sufficiency drive seems to be gaining steam, with tech giant Baidu touting the quality of its in-house Kunlunxin chips during an earnings call. Chinese AI chip designers Hygon and Sugon both posted a big jump in earnings.
Meta-Google deal. Meta has agreed to rent billions of dollars of Google’s TPU AI chips to train Meta’s own models. The deal is still developing with little reporting on the details yet. This is the latest in Meta’s compute buying spree. Just this past week Meta entered multibillion dollar deals with AMD and Nvidia.
GPU-backed debt. Interesting perspective in an FT piece on GPU-backed debt that has become a trend in recent years and has only increased in number. Using chips as collateral to buy more chips has a sheen of circularity that’s come to characterise many of the deals in the AI boom. One big question is how long investors think the lifespan of a GPU is. It’s probably shorter than what people want it to be.
Thanks for reading.
1. Policy and Geopolitics
1.1
Bloomberg (02/26): Nvidia Gets US License for Small Amount of H200 Exports to China
Nvidia Corp. said it secured a license to ship a small number of its less advanced H200 chips to customers in China, inching forward in its bid to return to the world’s largest semiconductor market.
The leading maker of artificial intelligence accelerators said that the US government granted it a license for H200 shipments, subject to inspection in the US and a 25% duty. But it’s still unclear whether Beijing will allow even that limited return, and Nvidia said Wednesday that it wasn’t including any China data-center revenue in its first-quarter sales outlook.
1.2
Nikkei (02/27): Japan to hold 10% voting rights in chipmaker Rapidus but with veto power
The Japanese government will become the largest shareholder in domestic chipmaker Rapidus under plans revealed Thursday, holding voting rights of only around 10% but reserving the ability to raise that above 50% if the company runs into trouble.
Japan also will possess a golden share giving it veto power over important management decisions to bolster the company against economic security risks.
The Ministry of Economy, Trade and Industry soon will announce a 100 billion yen ($641 million) investment in Rapidus, which aims to mass-produce cutting-edge chips, through the Information-technology Promotion Agency, Japan. An additional 150 billion yen is earmarked in the fiscal 2026 budget proposal, bringing Japan’s investment total for fiscal 2026 to 250 billion yen.
2. Economy, Finance, and Business
2.1
Reuters (02/27): Meta signs multi-billion-dollar deal to rent Google AI chips, The Information reports
Meta Platforms has signed a multi-billion dollar deal to rent artificial intelligence chips from Google to develop new AI models, The Information reported on Thursday, citing a person involved in the talks.
The report of the multi-year deal comes as companies pour billions into chips and AI infrastructure to meet demand for artificial intelligence.
2.2
FT (02/26): Tech stocks dip as Nvidia results fail to quell AI spending fears
US tech stocks fell on Thursday on renewed concerns of a possible bubble in AI tech spending, though the decline stalled in afternoon trading after progress on US-Iran nuclear talks tamped down oil prices.
The Nasdaq had fallen by as much as 2.1 per cent earlier in the day as Nvidia earnings failed to calm investors’ jitters about the sector. The company’s fourth-quarter earnings, released after the market closed on Wednesday, showed stronger than expected revenues and surging profits, prompting shares to initially rise in after-hours trading.
2.3
FT (02/26): Tech groups turn to more chip-backed loans to fund AI arms race
Tech companies are increasingly turning to loans backed by the chips on which their large language models are trained as they hunt for ways to fund their massive AI investments.
Such loans, which are secured against graphics processing units and backed by leases to the tech groups, are popular with a sector burning hundreds of billions of dollars a year in an AI arms race on chips that can quickly become obsolete.
Pioneered by cloud computing provider CoreWeave in late 2023, GPU-backed debt has grown in popularity as demand for advanced chips skyrockets and prices soar. Citigroup estimates that GPUs and associated servers can account for 30 to 40 per cent of total project costs for data centres.
2.4
SCMP (02/26): Baidu touts AI chips’ potential amid sluggish fourth quarter
Baidu founder, CEO and chairman Robin Li Yanhong touted the performance and potential of artificial intelligence chips developed by its Kunlunxin semiconductor design unit during the fourth-quarter earnings call, offering a rare bright spot in an otherwise lacklustre quarter.
The Kunlunxin chips are built on a proprietary architecture and deliver “stable, high-performance AI computing at scale with broad compatibility across different [AI] models”, Li said, citing strong inference efficiency that has driven broad enterprise adoption across sectors including finance, telecommunications and energy.
2.5
SCMP (02/26): China computing stalwarts Hygon, Sugon post revenue surge on AI boom, tech self-reliance
Semiconductor designer Hygon Information Technology and supercomputer maker Sugon, two stalwarts of China’s tech self-reliance efforts, reported surging revenue for 2025 amid strong domestic demand for home-grown computing systems.
The two Shanghai-listed companies’ latest financial results reflect a broader trend across China’s public and private sectors to increase engagement with domestic technology suppliers, as artificial intelligence development projects in the country intensified.
Headquartered in Beijing, Hygon posted revenue of 14.4 billion yuan (US$2.1 billion) last year, a 56.9 per cent jump from 2024. Profit attributable to shareholders reached 2.54 billion yuan, up 32 per cent from a year earlier.
2.6
Yonhap News (02/26): Samsung Electronics reclaims No. 1 spot in global DRAM market in Q4
Samsung Electronics Co. reclaimed the top spot in the global dynamic random-access memory (DRAM) market in the fourth quarter of last year, driven by increased sales of high bandwidth memory (HBM), data showed Thursday.
SK hynix Inc. saw its fourth-quarter DRAM sales rise 25.2 percent on-quarter to $17.2 billion, dropping to the second place with a 32.1 percent market share.
It marks the first time Samsung has led the DRAM market since losing the top position in the first quarter of 2025 -- its first loss of the title in 33 years.
2.7
Bloomberg (02/27): Smartphone Market Set to Shrink 13% Due to Memory Chip Crisis, IDC Says
The global smartphone market will contract 12.9% in 2026 because of the unprecedented memory chip shortage, marking “a crisis like no other,” according to researcher IDC.
3. Technology
3.1
Reuters (02/27): ASML says next-gen EUV tools ready to mass-produce chips, marking key shift for AI chip production
ASML Holding’s next-generation chipmaking machine is ready for manufacturers to start bringing it into use for production at high volumes, a senior executive told Reuters, a big step for the chip industry.
The Dutch company produces the world’s only commercial extreme ultraviolet lithography (EUV) tools, which are a critical piece of equipment for chipmakers. This new tool will help chipmakers such as Taiwan Semiconductor Manufacturing and Intel produce more powerful and efficient chips, by eliminating several costly and complex steps from the chip-manufacturing process, data from ASML shows.
3.2
TrendForce (02/26): Apple Reportedly Eyes Chinese Memory as Leverage Against Samsung, SK hynix Amid Rising Costs
Apple is reportedly exploring Chinese memory suppliers as prices continue to climb. According to Commercial Times, citing Wccftech, sources say Apple is evaluating products from major Chinese memory makers for potential use as early as the iPhone 18 series, as well as MacBooks. The move aims to enhance supply flexibility and strengthen its negotiating position with the three dominant suppliers — Samsung Electronics, SK hynix, and Micron.
3.3
Bloomberg (02/26): Broadcom Ships New AI Chip to Fujitsu, Plans Wider Rollout
Broadcom Inc. is shipping a new custom AI chip design to Fujitsu Ltd. that stacks components to save energy — an approach it expects big data center operators to adopt later this year.
Broadcom’s 3.5D eXtreme Dimension System in Package takes two pieces of the chip called dies and stacks them top to top, rather than a previous system pioneered by Advanced Micro Devices Inc. that piles them top to bottom. The stack allows for more data to be transferred with better power efficiency, said Harish Bharadwaj, vice president of marketing in the company’s custom chip unit.
Fujitsu will use the chip for data centers and possibly supercomputers in the future, Bharadwaj said. The next phase will be having so-called hyperscalers — the largest data center operators — adopt the technology, he said.
–