hardware

Leifeng Network · Whale Rhino has learned that OpenAI is secretly developing a screenless AI hardware product in Shenzhen. Industry insiders say that the project's confidentiality level is extremely high.Engineers came to Shenzhen for a business trip without even bringing their phones. The product is planned to be launched in 2027.

According to CoinDesk, the U.S. Department of Commerce has allocated up to $100 million each to three quantum computing companies: Rigetti, D-Wave, and Quantinuum, totaling up to $300 million, to expand hardware, manufacturing, and error correction systems, while simultaneously acquiring minority stakes in the three companies.Meanwhile, Bitcoin and Ethereum developers are accelerating the advancement of post-quantum cryptography upgrade plans. The Ethereum Foundation has set December 2029 as the deadline for the base layer quantum resistance upgrade, covering the execution layer, consensus layer, and data layer; although there is no unified timeline for Bitcoin, proposals BIP-360 (post-quantum output types) and BIP-361 (phasing out ECDSA and Schnorr signatures) have been accelerated, with researchers also viewing 2029 as a critical window for completing a trustworthy migration path.

User chris__lu posted that they compiled all 236 companies and 470 founders from the YC Summer 2026 batch, categorizing each company into an AI technology stack layer and comparing it to the Spring batch using the same criteria. This batch still has 91% related to AI. The model companies increased from 8% to 20%, the application layer decreased from 55% to 39%, horizontal applications dropped from 58 to 32, and vertical applications remained at 25%.In the Spring, 45% of companies delivered autonomous agents, while in the Summer, it was 33%, with "agent" in a one-sentence introduction dropping from 27% to 19%. 21 companies are engaged in computational infrastructure, 11 focus on inference costs, and there are also companies for training data and reinforcement learning environments. Scale AI is listed as an alternative target by 8 companies. The industrial category increased from 12% to 24%, with 45 companies delivering physical products, 24 being robots or physical AI, and 21 companies operating their own businesses rather than selling software.This batch is the youngest, with 37% of founders being students or graduates in the last two years, 59 teams are entirely student teams, the dropout rate increased from 3% to 9%, and repeat founders decreased from 32% to 23%, with 84% having a technical background. 39 from Berkeley, 32 from MIT, and 25 from Stanford. Amazon is the largest source of talent. Sales and marketing AI decreased from 18 to 6. Only 19 founders come from AI labs, accounting for 4%. 8 founding teams come from the same previous employer.

The Ethereum Foundation stated on the X platform that the Trezor hardware wallet now supports the "Clear Signing" feature through the ERC-7730 standard, making transactions more readable and understandable on the hardware device.This progress is the result of the Clear Signing working group promoted under the foundation's "Trillion Dollar Security Initiative," aimed at reducing the risk of users suffering asset losses due to "blind signing" of unclear transactions and encouraging more wallets and decentralized applications to participate.

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 TechCrunch, the venture capital firm Andreessen Horowitz (a16z) has expanded the size of its fifth growth fund from $6.75 billion established in January this year to $8.5 billion, adding $1.75 billion in new funds. The fund primarily invests in growth-stage startups that are expanding products, exploring new markets, and scaling businesses.This fundraising comes just days after a16z announced the completion of its $1.1 billion Machine Era Fund. This fund mainly focuses on the AI hardware sector, with investment areas including infrastructure startups in chips, storage, networking, and data storage. David George, head of a16z's growth investment team, stated that the relevant growth fund has invested in over 100 companies in the past seven years. As AI companies enter the growth stage more rapidly, their financing needs and valuations are also rising. a16z is currently focusing on areas such as enterprise and consumer AI, defense technology, robotics, software and hardware infrastructure, and medical technology.

According to "DuJia," DingTalk is advancing a brand new AI office application QwenNote (Qianwen Listening Note). This application is positioned as an AI personal assistant, integrating real-time voice transcription, summarization, and translation through a combination of software and hardware, and deeply integrating with AI Agent, embedding Agent capabilities into voice input, promoting a shift from simple recording to automated execution. The application supports real-time transcription and bilingual recognition in Chinese and English, as well as language switching, and can generate structured meeting minutes, outlines, and to-do lists, with built-in AI Q&A and quick commands based on listening materials.QwenNote offers a voice memo function, requiring a QR code scan to connect to the recording device. By long-pressing the button on the back of the device, users can record inspirations, and once the recording is complete, it will automatically archive, generate a title and brief summary, and timestamp it. The product also features a stealth protection mode, which, when activated, will physically delete the original audio and only retain the transcribed text to accommodate confidential scenarios. The associated hardware QwenNote A2 has already been showcased, which is part of the Qianwen Listening Note hardware ecosystem, which also includes DingTalk A1, DingTalk A1 Pro, Cleer H1, etc., and users can complete the binding by scanning a QR code.Reports indicate that DingTalk hopes to complement the offline voice collection entry through the integration of software and hardware, forming a closed loop of on-site audio collection, real-time bilingual transcription, AI meeting minutes Q&A, and DingTalk organizational collaboration. Listening materials can be synchronized to DingTalk AI Listening Note and support personal private isolation. On the software side, it continues to embed large models into documents, meetings, and IM scenarios, while on the hardware side, it expands the Qianwen Listening Note product line. Relevant hardware has not yet been widely publicly searched for more formal release information.

According to Gate market data, U.S. stocks opened with the Dow Jones up 0.26%, the S&P 500 index up 0.29%, and the Nasdaq up 0.12%. Computer hardware concept stocks strengthened, with HP up 6.4% and Dell up 3.4%. Cisco fell 7.9%, as the company's performance exceeded expectations but did not meet Wall Street expectations; the Nasdaq Golden Dragon China Index fell 0.65%, and JD.com fell 2.3%.

According to Gate market data, the Hang Seng Index opened lower and then rose sharply this morning. After opening more than 150 points down in the early session, it surged back, recovering losses and turning positive. As of the midday close, the Hang Seng Index rose by 0.07%, reporting at 25,457.33 points, while the Tech Index rose by 0.3%, reporting at 4,790.56 points. The total market turnover reached HKD 132.028 billion, with strong performance in the computing hardware sector. In terms of individual stocks, Cambridge Technology rose by 15.37%, and Shengbang Technology rose by 7.69%.

According to TechCrunch, OpenAI is fighting back against Apple's lawsuit regarding the leak of trade secrets, stating that Apple itself has vulnerabilities in its information security management and employee departure processes, and cannot prove that the relevant information is protected trade secrets.Apple previously accused OpenAI of obtaining hardware-related secrets through former employees who left Apple and requested the court to expedite the evidence investigation. In the latest legal documents submitted, OpenAI stated that Apple allows employees to use personal iCloud accounts to handle work materials and did not promptly revoke access after employees left. OpenAI claimed that Apple did not specify the exact trade secrets that were allegedly stolen, but only referred generally to information such as product development processes, supply chains, and testing.OpenAI believes that Apple's lawsuit is more about restricting its development in the AI hardware field and denies that the company needs or uses Apple's trade secrets. The focus of the dispute between the two parties centers on employee mobility, intellectual property protection, and competition in the AI hardware sector.

According to Decrypt, Ledger's Chief Technology Officer Charles Guillemet stated that the Coldcard vulnerability incident exposed weaknesses in hardware wallet random number generation and emphasized that AI is reshaping the cybersecurity offensive and defensive landscape. Guillemet pointed out that the security model of hardware wallets "lives or dies by random numbers," and this incident proved that in the most expensive way.Ledger stated that its hardware wallets were unaffected because their mnemonic phrases are generated by a hardware random number generator within a certified secure element, with no software fallback path, generating a complete 256-bit random number each time. Guillemet believes that open-source code, if not thoroughly reviewed, does not equate to security, and AI is enabling attackers to scan code and identify vulnerabilities at machine speed, requiring defenses to advance at the same speed, which relies on secure design, hardware, and mathematics. Guillemet suggested that users should understand how the random number generation of a hardware wallet works and whether it has been independently certified when choosing one.

The Coldcard hardware wallet attack has expanded to approximately $116 million, with attackers transferring 1,816 BTC from over 5,200 addresses. Galaxy Research confirmed that a fourth wave of transfers has occurred, with this wave alone transferring about 449 BTC after adjustments.Alex Thorn, head of Galaxy Research, stated that the latest activity may be "the fourth round of organized theft," with a transfer rate reaching 13.8 transactions per block, which is about 45 times the baseline level before the incident. Their analysis suggests that the related patterns indicate multiple groups may be scanning the affected key space in parallel. The root of the incident can be traced back to a 2021 firmware update for some Coldcard devices, which changed the wallet seed generation process from a hardware random source to a predictable software model. The Canadian manufacturer Coinkite recommends that users who generated wallet seeds on Coldcard devices transfer their funds to a new secure wallet as soon as possible.

Zhao Changpeng posted on platform X, stating that regarding the suspected vulnerability of the Coldcard wallet that led to over $38 million in Bitcoin being stolen, even hardware wallets may have vulnerabilities.Even long-established wallets may have vulnerabilities. To reduce risk, one might consider diversifying funds across several wallets, but this brings another set of risks. Nothing is 100% secure.

Canadian hardware wallet manufacturer Coinkite has issued a security warning, advising users who generated mnemonic phrases with Coldcard Mk3 firmware versions 4.1 to 5.3 to migrate their assets as soon as possible. The company stated that preliminary analysis shows wallets using BIP-39 passphrases are at lower risk, and Mk4, Q, and Mk5 devices are not affected; the investigation is still ongoing.Meanwhile, security researchers are investigating an anomalous transfer event involving 594.48 BTC (approximately 38.3 million USD). AnchorWatch CEO Rob Hamilton stated that attackers transferred 1,324 UTXOs through 500 transactions within three blocks and speculated that the issue may stem from insufficient random number entropy during the wallet generation process. Wizardsardine CEO Kevin Loaec believes the vulnerability may be related to a specific software library, secure chip, or a particular batch of devices, as well as low-entropy random number generators in firmware versions, with attackers possibly using AI-generated scripts to brute-force the affected wallets. There is currently no conclusive evidence linking this fund transfer to a direct vulnerability in Coldcard Mk3.

According to The Block, Trezor's Chief Business Officer Danny Sanders refuted recent criticisms of cryptocurrency hardware wallets made by blockchain investigator ZachXBT.Earlier this week, ZachXBT posted on Telegram, "All hardware wallets are complete garbage and I do not recommend using them for important tasks such as signing transactions or storing funds," and suggested that users use a dedicated iPhone to store funds and sign transactions.In response, Sanders acknowledged in an interview that firmware/software updates for hardware wallets can sometimes interfere with urgent or large transactions, and that the user experience can indeed be "clunky," understanding the users' frustration. However, he opposed ZachXBT's blanket statement: while hardware wallets may not be the best single solution for advanced users managing large amounts of funds, they are not garbage.He pointed out that compared to hardware wallets, the operating system and connectivity features of an iPhone present more "attack risks." For the average cryptocurrency user, hardware wallets remain the best self-custody option available.

Ledger researchers found that all Tangem hardware wallet cards' passwords can be reset through a laser attack.This attack requires physical contact with the device and approximately $250,000 worth of laboratory equipment, and the cards that are already in circulation cannot be patched to fix this issue.

According to The Block, the Donjon security team under Ledger has bypassed the firmware verification mechanism of the TROPIC01 chip used in the Trezor Safe 7 through precise laser attacks in a laboratory environment, allowing unauthorized firmware to be loaded as long as the attacker possesses the physical device.Chip manufacturer Tropic Square further discovered that there are additional attack vectors in the MAC-and-Destroy security mechanism used for PIN verification on this chip, but details will not be disclosed until the enhanced version of the chip is launched at the end of 2026.Trezor stated that the PIN, mnemonic backup, and private keys have never been stored on a single chip, and they have notified partners, so ordinary users do not need to take action. Currently, the feasibility of attacks can be reduced by disabling the chip's MAINTENANCE mode.

According to The Block, Bitcoin hardware wallet manufacturer Foundation has completed a new round of financing of $6.4 million, led by Fulgur Ventures with participation from Arche Capital and others, aimed at expanding from Bitcoin self-custody wallets to scenarios such as identity, multi-factor authentication, and AI agent authorization. This round of financing brings the company's total funding to $16.5 million, with valuation and round structure undisclosed.Foundation also announced that its flagship device, Passport Prime, is now available for sale to all users, integrating a Bitcoin hardware wallet, FIDO security key, 2FA storage, confidential information vault, and 50 GB of encrypted storage, emphasizing "human authorization" through dedicated hardware. In addition, the company is expanding the scope of the KeyOS developer platform and plans to launch the KeyOS app store by the end of the second quarter, with initial partners including Cake Wallet.

According to Decrypt, the cryptocurrency technology company Quantography Labs has announced the launch of the early access version of the cryptocurrency wallet Lock.com. This product uses "Isolated Signing" and an Air-Gapped architecture, enabling offline private key management without the need for dedicated hardware wallet devices.The official statement indicates that Lock.com separates the signing environment from the network broadcasting environment: private keys are always stored in a completely offline device, while transactions are created and broadcast on a connected device, thus avoiding private key exposure to the internet. The platform emphasizes that users can directly use existing devices without relying on third-party hardware manufacturers or supply chains. Additionally, Lock.com integrates post-quantum cryptography architecture, including ML-DSA digital signatures and ML-KEM key encapsulation mechanisms, to enhance security in future quantum computing environments.Quantography Labs stated that the current early access phase is mainly used to collect user feedback and prepare for a subsequent full launch.