AI ALGORITHMS ARE BEING PACKAGED AS A FINANCING ASSET

The Silicon Data data show that rents of A100, H100, H200 and B200 in non-Hyperscaler markets are approximately 1.65, 2.72,.29, $5.61/GPU hours. H100 has clearly rebounded at $2 since the end of last year, with B200 and H200 synchronizing. Based on a 36-month forward rent curve, the residual value of H100 was estimated to have risen from $14,000 in October last year to about $20,000 at present (moderated, non-actually traded). The rent movement is evidence of Wong In-hoon ' s logic of calculating assets: once the GPU stabilizes and generates reasoned rents, it can be financed with the same future cash flow as aircraft and ships. On August 10th, United Apolloto, BlackRock and others promoted over $500 billion in AI infrastructure finance, some of the loans were mortgaged with equipment, and Yveida or up to 25 per cent guaranteed with residual value, a model that, if established, would reduce the cost of Neo-claud financing and feed the GPU. It should be noted, however, that Silicon Data's own second-hand market data show that the H100 mark price for about three years is only 20-30 per cent of the peak of the new product and that technical depreciation remains significant; A 100-year rent stabilization only indicates that old money is still valuable in the low-cost scenario of reasoning, fine-tuning, etc. More precisely, the economic life of the GPU may have been longer than the previous market assumption of 2-3 years, and strong reasoning demand is offsetting some of the technical depreciation, which is critical to the valuation of the $500 billion financing plan. Kim Xian