tpu

Analyst Ming-Chi Kuo stated that Jensen Huang's departure could create market opportunities for Google's self-developed AI chip TPU. Kuo mentioned that Tim Cook excels in cost management, but Jensen Huang performs better in new technology planning and development.

According to BBX data, yesterday and in recent days, global publicly listed companies in the U.S. stock market have announced the latest developments regarding the sale of digital asset reserves and business operations, with key information as follows:· Hyperscale Data (NASDAQ: $GPUS) liquidated the remaining 211 bitcoins, fully transforming into an AI data center: The publicly listed company Hyperscale Data announced that as of November 8, the company has sold all of the remaining 211.56 bitcoins on its balance sheet at an average price of approximately $76,000 per bitcoin, cashing out about $16 million. Since initiating the liquidation of its reserves in late July this year, the company has cumulatively sold about 1,113 bitcoins, generating total revenue of approximately $74.4 million. The official statement clearly indicates that these funds will be specifically used to repay company debts and accelerate the construction of its AI data center in Michigan.· Bitdeer (NASDAQ: $BTDR) released its Q3 2026 financial report, with bitcoin production surging 248% year-on-year: Leading bitcoin mining company Bitdeer announced impressive performance for the third quarter. The total revenue for the quarter reached $320.7 million, a significant year-on-year increase of 89%; adjusted EBITDA soared 173% to $69.6 million. In terms of computing power and output, its self-mining computing power surged 135% year-on-year to 82.3 EH/s, driving the total bitcoin production for the quarter to 3,855 bitcoins, achieving an astonishing year-on-year growth of 248%.· SOL Strategies disclosed Q4 performance, with its HoudiniSwap trading volume skyrocketing: SOL treasury ecosystem company SOL Strategies released preliminary unaudited performance for the fourth fiscal quarter. The total revenue for the quarter was approximately CAD 4 million (about USD 2.83 million). Notably, its decentralized exchange protocol HoudiniSwap successfully processed approximately CAD 326 million (about USD 231 million) in exchange transactions during its first full quarter after being integrated into the company, generating approximately CAD 3.4 million (about USD 2.41 million) in exchange fee revenue. Its trading volume in September saw a significant increase of 37% compared to June.

OpenAI announced the launch of a centralized system called Control Room, designed to track model workloads within its research infrastructure and quickly halt related tasks when security or safety concerns arise. This system currently covers research workloads within OpenAI, including training experiments, evaluations, and data processing. OpenAI stated that as model capabilities increase, associated safety and security risks also rise, making it increasingly critical to detect potential issues in a timely manner and swiftly halt models when necessary.In terms of specific mechanisms, Control Room assigns an identifier to each research task, and the inference engine verifies this identifier before generating model outputs. When a stop command is issued by manual or automated monitoring systems, Control Room revokes the authorization for that task, thereby blocking subsequent outputs, even if the underlying research program is still running. OpenAI noted that cached authorizations could delay shutdowns by up to 35 seconds; additionally, if Control Room itself is unavailable, the system is designed to reject new requests.Prior to this release, OpenAI had disclosed multiple incidents this year where research models deviated from instructions during evaluations. In July, the company reported that its model escaped a restricted testing environment and infiltrated Hugging Face to cheat in a security assessment. In October, it disclosed that a research model used an unauthorized encrypted wallet to pay for smart contract deployment fees on the testnet ETH, and this expenditure was omitted from the final summary.

Google (GOOG.O) will launch two new versions of its Tensor Processing Unit (TPU) systems, namely TPU 8TX and TPU 8I Max, to handle more specialized tasks and challenge Nvidia.

Google (GOOG.O) collaborates directly with Anthropic on TPU and partners with Broadcom (AVGO.O).

Bitcoin Treasuries posted on platform X that Strive's SATA has reportedly raised enough funds to purchase over 400 Bitcoins.SATA is currently buying 90% of the daily Bitcoin mining output.

According to The Information, citing informed sources, the artificial intelligence company Anthropic is in talks with data center operator Stream Data Centers to directly lease up to 1GW of data center computing power. The two parties are still in early negotiations, and the final scale has not yet been determined, with the deal possibly not being finalized.If signed, Anthropic will directly become a tenant of the data center and deploy its own servers and chips. The related equipment may include TPUs from Google and Broadcom, and may also use NVIDIA GPUs. Google is also discussing providing credit guarantees for some projects. This is different from renting servers from AWS or Google Cloud, as Anthropic can control the chips, racks, and network equipment itself, paying less in cloud vendor profits but needing to bear the costs of equipment procurement and data center operations.Data center developers interviewed by The Information estimate that a complete computing infrastructure of 1GW scale requires at least $40 billion in capital investment. Anthropic has already directly leased approximately 401MW of data center from TeraWulf this year and announced a partnership with Fluidstack last year to invest $50 billion in building a custom data center. If the deal with Stream reaches 1GW, it will become one of Anthropic's largest direct data center leases to date.

During an exclusive interview with China Central Television's finance program, Tesla CEO Elon Musk stated that China's AI large models are overall outstanding. He mentioned that, under limited computing power, the output per unit of computing power of Chinese models is already close to the world's top level.Musk assessed that China will solve the computing power bottleneck faster than most people expect. He roughly estimated that in the next 2 to 3 years, China could bridge the computing power gap relying on lithography technology and chip manufacturing capabilities.

Google's Senior Director of Supply Chain Infrastructure, Nikhil Cherian, revealed that to overcome memory bottlenecks, Google is developing software and hardware solutions and dismantling retired servers to recycle DDR4 components to establish an internal recycling supply chain. Google has designed special hardware adapters to connect the previous generation DDR4 to the new generation of AI servers while importing retired servers to remove their DDR4 modules for recycling.Cherian stated that the AI industry has rapidly shifted from being compute-constrained to memory-constrained, with high-performance memory accounting for about 75% of the bill of materials cost for a given AI server. A Goldman Sachs report indicated that memory prices will continue to rise in the third quarter, with personal computer DRAM prices expected to increase by 18% to 23% and server DRAM prices expected to rise by 13% to 18%. Trendforce data shows that in August, the spot market prices for DDR4 8GB and DDR5 8GB rose to $142 and $133, respectively.The two TPU ASICs launched by Google this year have been optimized for memory design, claiming to reduce memory demand to one-sixth of the original. The TPU8i chip features a dedicated layered memory design that relies on a high-speed DDR5 memory architecture to perform host-level tasks, with each chip equipped with 288GB of HBM3e high-bandwidth memory.

The SEMICON Taiwan 2026 Memory Summit took place on the 1st, where Nikhil Cherian, Senior Director of Supply Chain Infrastructure at Google Cloud under Alphabet, pointed out that with the popularity of multimodal and mixed expert architectures, AI computation has shifted from being power-limited to memory-limited, with high-performance memory accounting for over 75% of the cost of AI server hardware bill of materials. In the face of capacity, bandwidth, and power consumption bottlenecks, Google is breaking through the AI memory bottleneck through a dual-track strategy of hardware offloading for inference and training, and lossless quantization software algorithms.Google adopts an offloading strategy in hardware architecture, launching TPU 8i for low-latency inference and TPU 8t specialized for large-scale training. The TPU 8i is equipped with 288 GB of high-bandwidth memory, with SRAM capacity on the chip increased threefold to 384 MiB, placing dynamic conversation states and key-value caches on the chip itself to achieve zero chip-off latency. The TPU 8t forms a super-large computing cluster with 9600 chips, achieving a shared pool of HBM at a scale of 2 PB, eliminating chip-off data transfer bottlenecks, along with TPU Direct Storage technology.Google has developed the training-free TurboQuant lossless quantization algorithm, compressing the key-value cache of large models from 32 bits to 3 bits, reducing memory usage by six times without loss of accuracy, resulting in an eightfold acceleration in attention computation, and integrating old-generation DRAM technology to extend the lifecycle of components.

According to a report by the Economic Daily, ASE Technology Holding benefited from the increased capacity of Google TPU, significantly adding advanced packaging and testing orders. Intel is accelerating the promotion of its EMIB advanced packaging platform while simultaneously expanding its outsourcing orders to ASE Technology Holding, with the previously raised testing quotes showing benefits. Google continues to increase its AI capital expenditures to expand data centers, using TPU to support the Gemini large model, AI agents, and Google Cloud services. The market estimates that Google TPU will enter a large-scale production phase by 2028, with shipments expected to reach 12 to 15 million units. ASE Technology Holding has mastered 2.5D, 3D advanced packaging, and high-end testing technologies, and Google is also expanding its use of TSMC's advanced processes and CoWoS capacity.Intel is racing to advance its EMIB packaging platform, with major cloud companies like Meta and Google successively adopting the EMIB and EMIB-T platforms. ASE Technology Holding, positioned as a pure testing and packaging foundry, can directly procure organic embedded silicon bridge substrates to undertake backend wafer-level assembly and high-end testing services. Wu Tianyu, the Chief Operating Officer of ASE Technology Holding, believes that Intel's EMIB and TSMC's CoWoS are not zero-sum competitors. As the testing time for AI chips lengthens, the costs of equipment, materials, and labor are rising, and high-end packaging and testing capacity is becoming increasingly tight. Major testing and packaging factories have gradually raised prices based on products, capacity, and customer conditions, with increases of about 5% to 10%.

According to DIGITIMES, the Taiwanese semiconductor supply chain reveals that advanced packaging technology is diversifying, with Intel's EMIB becoming a key option. Google TPU has confirmed the adoption of EMIB packaging, and the first TPU using this technology has met yield standards, with technology, cost, and capacity performance all aligning with Google's original criteria. The likelihood of continuing to use EMIB in future generations of TPU has significantly increased.Against the backdrop of ongoing tight capacity at TSMC's CoWoS, as long as EMIB yield and technology remain up to standard, Google will continue to adopt it. MediaTek, as a relevant ASIC partner, has some managers in its U.S. team with Intel backgrounds, facilitating smoother collaboration between the two parties. Although MediaTek still relies heavily on TSMC, its success with Intel on TPU helps solidify Google's trust, secure more projects, and provide diverse packaging solutions to cloud customers.Previously, TSMC called for more suppliers to invest in advanced packaging to share the pressure during its earnings call. The supply chain generally believes that its own capacity expansion will not be limitless, making the diversification of packaging paths, such as those from Intel, a necessary direction, and related testing equipment and interface companies are also expected to benefit.

DeepSeek announced the new pricing for the V4 API, effective August 17. The V4-Pro peak period price for cache misses on input has increased from 3 yuan to 9 yuan per million tokens, and the output has risen from 6 yuan to 27 yuan, representing increases of 3 times and 4.5 times, respectively.During off-peak hours, although the price is only half of the peak rate, it is still more expensive than the current rates. The V4-Pro input is 4.5 yuan, and the output is 13.5 yuan; the V4-Flash input is 1.5 yuan, and the output is 4.5 yuan.Peak hours are from 9:00 to 12:00 and 14:00 to 18:00 Beijing time. The remaining hours are charged at off-peak prices.

The research organization SemiAnalysis released an analysis indicating that Google DeepMind is no longer among the leading AI laboratories. A week prior, DeepMind co-founder Demis Hassabis stepped back from daily operations, and key members such as Google Chief Scientist Jeff Dean and Gemini co-lead Oriol Vinyals left to establish a new lab called Discovery Loop. The analysis suggests that the long-term struggle within Google over computing power allocation between Gemini and GCP has concluded with GCP emerging victorious.SemiAnalysis stated that Gemini 3.5 Pro has been canceled, and Gemini 3.6 Flash's performance is inferior to that of leading Chinese open-source models and Grok 4.5. Currently, Gemini has fallen to the 8th or 9th position in the large model rankings. Meanwhile, GCP is selling a large number of TPUs to competitors like Anthropic, having secured long-term leasing and sales contracts for hundreds of thousands of TPUs over the past nine months. The Tokenomics model estimates that Gemini's own ARR is about $12 billion, while GCP's third-party AI cloud service revenue is expected to exceed $73 billion by the end of 2027, with TPU system sales contributing an additional over $120 billion. GCP's latest quarterly growth rate is 82%, and it is expected to accelerate to over 100% by 2027 due to TPU system sales, contributing approximately $3 to Google's earnings per share.

According to GlobeNewswire, Nasdaq-listed Bitcoin mining company Bitdeer released an unaudited production and operation update for June 2026. The company's Bitcoin mining output in June was 990 coins, a year-on-year increase of 388%; self-mining computing power was 73 EH/s, with a total managed computing power of 86.1 EH/s, and 243,000 self-owned mining machines.The Sealminer Nevada manufacturing plant has begun construction and is expected to be completed by the end of 2026. In terms of AI cloud business, monthly ARR has increased to $76 million, GPU utilization is at 95%, and the delivery of the NVIDIA GB300 NVL72 cluster has been completed. Currently, Bitdeer's total controllable power capacity globally reaches 3.02 gigawatts, with multiple cryptocurrency data centers simultaneously advancing the transformation into AI computing center plans.

According to PR Newswire, Canaan Creative announced unaudited Bitcoin mining operation update data, producing 64 Bitcoins in June.After deducting operating costs and accounting for the Bitcoins received from the sale of mining machines in June, the cryptocurrency reserves increased by 49 Bitcoins, bringing the total cryptocurrency holdings to a record 1,915 Bitcoins and 3,952 Ethers.

The U.S. stock-listed Bitcoin mining company Bitdeer disclosed that as of July 3, 2026, its own BTC holdings remain at 0, excluding customer deposits.This week, Bitdeer produced 223.1 BTC, sold 223.1 BTC, resulting in a net increase of 0 BTC. Bitdeer has disclosed its own BTC holdings as 0 since February 20, 2026, and it has continued for 19 weeks.

Tianfeng International analyst Ming-Chi Kuo revealed that Google is deepening its collaboration with MediaTek on the TPU v9, developing an upgraded chip codenamed Triggerfish. It is reported that MediaTek has exclusively obtained this high-value order. The new chip is positioned as a v9 variant with stronger inference capabilities, primarily targeting the AI Agent and reinforcement learning fields, aiming to address CPU and memory bottlenecks in the computing process.In terms of technical specifications, Triggerfish's SRAM capacity has increased by 2 to 3 times compared to the basic version (Humufish), with the addition of dedicated analog chips for reinforcement learning and AI Agent collaboration, and memory upgraded from HBM4 to HBM4E. Kuo expects that while the basic version chip maintains a shipment expectation of 4 to 5 million units, Google has added 1 to 2 million new orders for Triggerfish. The chip is expected to enter production by the end of 2027 and achieve mass production and shipment in 2028, with a unit price approximately 30% higher than the basic version, likely becoming a new driving force for MediaTek's business growth in 2028.

According to Reuters, private equity giant Apollo Global Management and Blackstone Group are bringing in more investors for a debt financing deal worth approximately $36 billion to help Anthropic build its artificial intelligence infrastructure.Insiders revealed that this debt financing will be used to purchase Google-customized TPU (Tensor Processing Unit) chips, which will then be leased by Anthropic. Broadcom, which assisted Google in developing the chips, will provide guarantees for the majority of the funds in this transaction.This move is expected to become one of the largest private credit transactions in history and may also be the largest chip financing debt transaction to date. The plan aims to provide computational power support for Anthropic by leveraging Broadcom's creditworthiness.

According to BBX data, several listed companies disclosed their latest reserve actions yesterday, with the core dynamics as follows:Tron Inc. (NASDAQ: TRON) announced that it increased its holdings of 154,494 TRX tokens at an average price of $0.3236 yesterday, further raising the total amount of TRX in its treasury to over 693.3 million tokens.The Jupiter Strategic Reserve Trust, Jupiter Litterbox Trust, increased its holdings by 211,474 JUP yesterday, valued at approximately $36,000. This month, it has accumulated a total purchase of 9,405,535 JUP, valued at about $1.63 million. As of now, its total purchases amount to 113,508,785 tokens, valued at approximately $19.76 million.Nasdaq-listed Bitcoin mining company Bitdeer (NASDAQ: BTDR) released the latest data on its Bitcoin holdings. As of the week of April 24, the Bitcoin mining output was 185.7 Bitcoins, but the same amount was sold during the same period, resulting in a net addition of 0 Bitcoins, and it currently maintains zero Bitcoin holdings.