Lookonchain APP

App Store

Arthur Hayes: FLOP miners' revenue depends on inference demand, and their hardware configurations need to be optimized in line with market demand.

14 minutes ago

Arthur Hayes stated in a post that recently, numerous individual and enterprise FLOP miners have reached out to him for advice on hardware selection and performance optimization, though he cannot provide specific configuration recommendations. FLOP miners’ block rewards are tied to the volume of inference tasks they process, measured in floating-point operations (FLOPs). For instance, if a miner handles 100 GFLOPs of inference tasks in one block, while the total network processing volume hits 200 GFLOPs, the miner will earn 50% of that block’s reward. Hayes cautioned that hardware performance is not the sole determinant of miner returns—actual market demand for AI models is equally critical. If a miner’s deployed hardware supports an AI model that lacks market demand, the device will not secure inference tasks and thus generate no revenue. He advised miners to reference popular model data from platforms such as OpenRouter to forecast model demand for agentic tasks, and refine their hardware configurations through hands-on testing and revenue performance tracking.

Relevant content

Binance hacker transfers 78,415.24 ETH to 7 addresses

According to Beosin’s monitoring, the attacker behind the Binance hot wallet breach has just transferred 78,415.24 ETH, valued at approximately $142 million, to six newly created wallet addresses and one previously used wallet address.

24 minutes ago

Cloudflare open-sources Kitesurf: No Chromium dependency, memory usage as low as one-seventh of comparable solutions.

Beating AI Express News: Cloudflare has officially open-sourced Kitesurf, a browser engine built specifically for AI agents. Kitesurf first debuted in August this year. Cloudflare points out that AI agents mainly read and interact with web pages, so many features of standard browsers are unnecessary. Instead of adopting Chromium, Kitesurf was rebuilt as a lightweight engine using existing open-source technologies, without needing to launch a full browser. Official tests show its CPU usage is roughly one-third that of Chromium, and memory usage as low as approximately one-seventh. Developers can integrate Kitesurf into their AI agents to directly control web pages, execute scripts, and take screenshots, or continue using existing tools like Puppeteer. Currently, Kitesurf remains an experimental project with incomplete support for many web standards, so it cannot fully replace Chromium.

24 minutes ago

Arthur Hayes-linked addresses received 35,738 ETH

According to on-chain analyst Ai Yi (@ai_9684xtpa), an address linked to Arthur Hayes has received 35,738 ETH, valued at over $64.49 million. Tracing the fund’s source backwards, this address is closely associated with asset management firm Galaxy Digital; its purpose and ownership remain unclear. The address may belong to Galaxy or its related entities, having previously interacted with Arthur Hayes.

24 minutes ago

Masayoshi Son: Failing to invest in Anthropic was a 'mistake', SoftBank deeply regrets it.

According to the Financial Times, SoftBank founder Masayoshi Son said Wednesday at the Nikkei Future of AI event held in Tokyo that he regrets failing to invest in Anthropic ahead of its large-scale initial public offering (IPO), calling it a mistake. He noted that he had already started investing in Anthropic’s rival OpenAI at the time, leading him to miss the opportunity to back the AI company. Son’s remarks come as OpenAI has delayed its IPO to at least next year and cut its ongoing financing target by $10 billion, aiming to raise $20 billion at a $1.25 trillion valuation. In stark contrast, Anthropic plans to launch its IPO this week, targeting a maximum valuation of up to $1.4 trillion.

24 minutes ago

Bland launches Tofu-1: The AI solution can detect interruptions from TVs, and its AI customer service can process business operations while speaking.

Beating AI News – Voice AI company Bland has unveiled a new model, Tofu-1, specifically designed to handle complex calls in industries including banking, healthcare, and insurance. The model processes audio directly, enabling it to speak and listen simultaneously, allowing customers to interrupt or correct at any time, and distinguish between the customer, bystanders, and sounds from televisions. For example, if an AI customer service agent asks a customer, “Confirm cancellation of the insurance policy?” and before the customer can respond, a “Yes” is suddenly heard from a nearby TV, traditional voice assistants may mistake this for the customer’s consent. Tofu-1, however, identifies the sound source to avoid such errors. Tofu-1 also supports processing requests mid-conversation. For instance, when checking a claim status, the AI can continue interacting with the customer, and promptly follow up if the customer modifies information midway. Bland said the model will only report results after a task is fully completed. Currently, Tofu-1 is only open for enterprise research preview applications, with no released measured data on latency, recognition accuracy, or business task success rates.

24 minutes ago

Axe Compute’s Q3 revenue rose 125% quarter-over-quarter, with customer advances pushing its cash balance to $148 million.

According to official announcements, Axe Compute (NASDAQ: AGPU) disclosed on November 17 that its third-quarter 2026 revenue for the period ending September 30 reached $7.24 million, a 125% increase from $3.22 million in the second quarter. The company posted a quarterly operating profit of $6.83 million, net profit of $11.5 million, and diluted earnings per share of $0.52. On the cash front, its ending cash and cash equivalents jumped to $147.7 million from $21.9 million at the end of Q2; contract liabilities from customer advances surged from $60.8 million to $569.3 million. Axe Compute said these advances are primarily used for GPU infrastructure procurement: the firm invested $349.6 million in the quarter to purchase cluster equipment for its Build project. For business expansion, the company noted it has signed orders with a total contract value (TCV) of over $3 billion so far this year, with an expected annualized revenue run rate of more than $7 billion once all deployments are completed. On September 30, Axe acquired a Georgia-based cluster equipped with 2,304 NVIDIA B300 GPUs for $124.4 million, its first self-owned cluster. The project is pending customer acceptance, targeting launch in the fourth quarter, and the customer’s contract has been extended to 2031. The company also completed the sale of Helomics, officially transitioning to a single business focused on GPU-as-a-Service (GaaS). Management did not issue formal financial guidance, only stating that the Georgia cluster and the Build contract signed in July will be deployed gradually starting in Q4 2026 and continuing into 2027. Notably, revenue from the Build contract has not been recognized; all third-quarter revenue came from GPU computing services under Axe Compute’s Access model.

24 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano