When I saw the official announcement from @ falconflance, the first thing that came to my mind was the RWA attempted by Centrifuge during the Defi Summer period At that time, the income from music copyright was put on the chain as collateral for loans, but later it was unexpectedly abandoned. The RWAs that have proved effective in the past few years are basically "standardized" or "homogenized" assets - the earliest stable currency, later treasury bond, gold, and now token American stocks GPU， In the current era of AI, it may be an asset form that is large enough in size and standardized/homogeneous, which is still considered "good" First, let me briefly explain what Falcon officially announced - the establishment of an RWA issuance pipeline under the regulatory framework of El Salvador (why El Salvador - because Tether Gold - XAU ₮ is also issued under this framework, veteran driver). Currently, the first choice is tokenized GPU Forward - to finance the construction of computing infrastructure through tokenization. NEAR AI, as the Anchor Buyer for future computing power, provides demand anchors and plans to enter Dex transactions such as Uniswap in the future. Even more interestingly, once the trading volume is deep enough, it can still be accepted as collateral for the Falcon platform to mint USDf. Simply put, it is to turn non-standard assets such as "computing power financing" that only circulate in the hands of private banking groups into a chain asset that connects issuance, trading, and collateral. Why do I think this matter is more Make Sense? Because you can tell by combining the other two news articles in the past few days Firstly, let's talk about Lao Huang. On August 10th, NVIDIA announced that it will work with six global leading institutions, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, to build an AI Compute Financing Platform. The goal is to leverage over $500 billion in third-party capital to invest in AI infrastructure 2. Next is OpenRouter - On August 16th, Stripe officially announced the acquisition of OpenRouter for $7-8 billion, which is more than five times the valuation of OpenRouter's financing three months ago (btw, Alex is really amazing, he first made OpenSea and then OpenRouter, curious about what he will do next, Founder must follow his direction to start a business) When you look at the three pieces of news together, they are actually three aspects of the same thing: financial infrastructure is moving towards every link of AI economic activity. 1. Lao Huang's side is equivalent to setting a benchmark - computing power is no longer just a technical infrastructure, but an asset class that institutional capital is willing to allocate on a large scale, which is like issuing a pass to the entire track 2. Falcon's approach is to deposit this pass onto specific assets, using tokenization and collateral to transform GPU finance, which used to only circulate in private banking groups, into a tradable and withdrawable on chain asset 3. The OpenRouter card is on the AI consumer side, who is responsible for routing and billing model calls. Now that Stripe has accepted it, it means that even the payment process has been included Lao Huang sets the tone, Falcon conducts upstream financing and exits, OpenRouter collects downstream traffic billing and payment. The direction is different, but the logic is the same - capital is now willing to pay a very high premium for the "pipeline that can catch the cash flow of AI economic activities", regardless of whether this pipeline is long on the computing power end, financing end, or consumption end. Of course, this structure is not without risks. Here are a few points I can think of: The first is that Oracle Compute Asset has not yet had the mature independent secondary market price like treasury bond. Who will price the GPU forward and how to prevent price manipulation The second is depreciation - GPU hardware itself has a Moore's Law depreciation curve, how can the forward term structure hedge this risk. In recent years, inference has exploded, and old cards such as H100 and A100 are also running at full capacity, with depreciation much better than before. However, this thing cannot continue to be so good in the future, it has to return to the mean Third, liquidity - historically, except for treasury bond, most RWA transactions are dead after token. Hanging a pool on Uniswap does not necessarily mean depth and true pricing. What I will focus on myself is not whether GPU forward can be released, but whether Falcon has the ability to make markets and divert funds, injecting the collateral demand of USDf into the secondary market in reverse, forming a positive cycle. This step cannot be taken, and the subsequent incident of 'low liquidity RWA being pledged into USDf' itself is adding a layer of liquidity mismatch exposure to the USDf system. Our circle should not suffer too much losses due to collateral issues The earliest RWA experiments conducted by Centrifuge have already proven that not all real-world cash flows are suitable for tokenization, only assets that naturally have homogeneous units of measurement and public pricing benchmarks are acceptable. Gold and treasury bond have proved this. In the era of AI, computing power should most likely be the first standardized attribute of a new asset class, so this matter is worth taking a look at