律动BlockBeats

This is an in-depth research report produced by OKX Ventures. Due to its length, it is published in two parts: the first focuses on the tokenization of computing power assets, risk exposure, and derivative pricing logic; the second will analyze the computing-power financial infrastructure, opportunities for tokenizing inference assets, and industry trends. This is the second part.
4. Financial infrastructure for the computing power market: pricing, risk, and settlement
Several layers of infrastructure for the computing power financial market are beginning to take shape. Indices consolidate dispersed GPU rental rates into a unified benchmark; exchanges organize standardized risk around the benchmark; dealers take over specific data centers, tenors, and the basis brought about by SLAs; and capacity platforms then connect financial positions back to physical computing power. The layer that first accumulates a sufficient number of real orders and contract references will be closest to having pricing power in the early market.

By Jia Liu, Cheers BeatZ
“Gold has basically already hit its bottom”—this is the prevailing consensus among financial institutions recently.
After a strong rise in the first quarter and a pullback in the second, COMEX gold’s main contract rebounded from $4,022.9 per ounce on June 30 to $4,380.4 per ounce on August 14: up 8.9% this quarter and 8.2% since August. Strategy reports from Wall Street, research meetings, and gold bulls on social media are all treating “the pullback is over” as the new consensus.
The problem is that the gold shops in the marketplace don’t seem to share this consensus.

This article is a deep research report produced by OKX Ventures. Due to its length, it is published in two parts: Part 1 focuses on the assetization of computing power, risk exposure, and the logic of derivatives pricing; Part 2 will重点 analyze the financial infrastructure for computing power, opportunities for assetization in inference, and industry trends. This piece is Part 1.
Introduction: The assetization of computing power and risk exposure
AI infrastructure is entering a capital-intensive expansion phase, and the economic attributes of computing power are changing accordingly. In the past, companies mostly viewed computing power as an IT cost purchased on demand; now, GPUs, data center capacity, and multi-year procurement contracts increasingly appear on the balance sheet. Computing power is starting to take on the characteristics of capital assets: large upfront investment, capital recovery that takes several years, and future revenue and equipment value that will continue to be influenced by supply-demand dynamics and technological iterations.

Original title: Everybody Wants to Rule the (AI) Stack.
Original author: David Levy
Editor’s note: As AI computing demand continues to grow, competition for new cloud providers is shifting from “who has more GPUs” to “who can control the complete technology stack from hardware to software.” As pure GPU power rental becomes increasingly difficult to escape price competition, a more critical question is emerging: what ultimately determines the value of AI infrastructure—are the GPUs themselves, or the software layer that schedules, orchestrates, and optimizes how GPUs operate efficiently?
The author, David Levy, traces his story through Nscale’s roughly $165 million acquisition of Anyscale, mapping deals such as Nebius’s acquisition of Eigen AI, CoreWeave’s acquisition of Weights & Biases, and IREN’s acquisition of Mirantis. These five acquisitions point to the same trend: new cloud companies that have GPUs, power, and data centers are collectively extending into the MLOps and AI orchestration layer.

Original title: Neocloud Economics
Original author: APP ECONOMY INSIGHTS
Editor’s note: Generative AI continues to drive sustained growth in compute demand. Industry discussions are shifting from “Is there enough GPU?” to “Who can convert electricity, chips, and data centers into usable compute at lower capital cost?” As growth in orders and compute shortages gradually become a consensus, a more critical question begins to emerge: before revenue is realized, new cloud providers are investing billions of dollars—are they building the next generation of AI infrastructure, or are they using debt and capital expenditures to pre-finance future demand?

BlockBeats message, August 17: after Samsung, Western Digital, and Toshiba, SK hynix has also been reported to have an SSD after-sales policy of “refund only and no replacement.” A Reddit user recently filed a warranty claim for an SK hynix SSD. However, because the official side has no replacement inventory available, the final offer was not to replace the product with the same model. Instead, the user received a refund based on the original purchase price.
This is especially awkward given that SSD prices are continuing to rise. Taking SK hynix Platinum P41 and P51 as examples: for some capacities, prices have already increased noticeably compared with their earlier low prices, and some have even reached about twice the price from the same year. If an SSD that was bought at a low price in the past fails now, and the user can only get back the original amount paid, they may no longer be able to afford products with the same specifications.

BlockBeats message: On August 17, this week’s market will see the Federal Reserve’s July meeting minutes, the August PMI, and earnings reports from retail giants such as Walmart and Target. Recent U.S. CPI, PPI, and retail sales data have been weak, and market pricing for a September rate hike has fallen to about 27%. Goldman Sachs also believes a September rate hike is “very unlikely,” which has eased near-term interest-rate pressure somewhat. However, the FOMC minutes still need to be assessed for officials’ attitudes toward inflation and energy prices. If the minutes deliver a more hawkish signal, interest-rate expectations may be repriced again.
Meanwhile, whether the Strait of Hormuz can truly restore commercial shipping will directly affect oil prices and inflation expectations. According to reports, Iran and Oman have made progress on shipping routes, but actual shipping volumes remain far below normal levels. Therefore, this is still diplomatic progress rather than a resolved supply risk. If shipping continues to recover, the crude oil risk premium could decline; conversely, if talks break down or maritime tensions escalate, oil prices moving higher again would increase global inflation and interest-rate pressure.

Original title: (AI giant internship pay revealed: Anthropic over 5,000 RMB, Kimi can only rank in the fourth tier)
Original author: Wenser, Odaily Planet Daily
In 2026, the “fundamentals” of internship pay for domestic college students are: the monthly median is 5,000 RMB. If it’s in a first-tier city and the role is a technical position, you can reach 10,000 RMB.
I was pretty okay with accepting the era’s dividend setup and admitting I made a professional wrong turn. I wasn’t that bitter—until I stumbled upon—
An AI company intern makes over 5,000 RMB a day……a day? 5,000?
It’s really hard to take. I went ahead and looked up internship salaries for 12 of the world’s AI giants—from OpenAI and Anthropic to ByteDance and DeepSeek, from official public hiring pages to sites like levels.fyi and Glassdoor.