Under the artificial intelligence craze, a large number of new power stations, chips and data centers have been launched one after another. At the same time, a more implicit supporting financial system has also emerged, establishing a feasible channel for the mortgage loan business of high-priced hardware assets. Miami financial technology startup Barkr is one of the companies building this financial service system. It is mainly engaged in AI chip mortgage asset evaluation business and has attracted the attention of Nvidia.
The author previously met the company’s CEO and co-founder Thomas Galbraith when he hosted a forum in San Francisco on how lenders conduct credit reviews for artificial intelligence infrastructure projects.
Galbraith's track belongs to a very segmented sector in the complex asset-backed financing field.
The company mainly provides insurance-backed valuation services for physical hard assets. The valuation categories include automobiles, artworks, and commercial equipment. Now it has also officially included AI chips. Barkr itself does not directly issue loans, but instead provides lending institutions with authoritative and credible asset valuation results that can be directly used in credit business.
Carrying out this business requires extremely strong confidence: once there is a deviation in Barkr's valuation and the borrower defaults, the company will be responsible for the relevant spread losses.
Barkr was founded in 2023. Its initial business was to price collectible luxury cars, private jets and other assets to facilitate lenders to reasonably set loan limits. About six months ago, a lender presented Galbraith with a new problem.
He recalled the other party's original words: "We are preparing to launch a graphics card mortgage loan business, but currently we can't find any institution that can give a fair price, and no one is willing to provide guarantee for the asset price."
In the past few years, NVIDIA graphics cards have become a new investment asset category, but the asset-backed financing business for this type of chips has not yet become popular. The core crux is that borrowers and lenders often disagree on the residual value of graphics cards, and there is no neutral third party that can mediate disputes. This also makes the terms of graphics card mortgage financing generally less relaxed than those of traditional physical asset financing.
There have always been two core risks in asset-backed loans: one is the borrower's repayment ability, and the other is whether the value of the mortgaged assets can be stably maintained. After decades of hard work, lending institutions have already matured and mastered the borrower credit review system; however, it is still extremely difficult to judge the value of the assets themselves, especially emerging assets such as graphics cards. Galbraith said that when five institutions were asked to evaluate the same batch of graphics card clusters, they would often come up with five completely different valuation results.
Barkr has independently developed a number of artificial intelligence valuation models to specifically calculate the value of various mortgage assets. It also cooperates with insurance institutions such as Munich Re to issue legally binding guarantee agreements for the valuation results. Galbraith said this model completely changed the business considerations of lenders. Fair and credible asset pricing can not only reduce lender risks, but also help financing companies obtain lower lending rates.
In the past six months, Barkr has assisted in completing graphics card-related financing transactions with a total value of approximately US$200 million, and is expected to undertake similar business with a scale of US$300 million before the end of this year. Compared with Nvidia's overall sales volume, the scale is not large, and the vast majority of current graphics card purchase orders are still completed by the world's leading companies relying on their own funds.
By comparison, three years ago, the investment institution Upper90 issued a US$200 million mortgage loan to the cloud service provider Crusoe, with the collateral being 20,000 graphics cards. Although similar transactions have been implemented since then, they have never been popularized on a large scale.
Make graphics cards a standardized and robust asset
This company, which has only 14 employees and has completed a total of US$3.5 million in venture capital financing, uses the same logic for evaluating graphics card clusters as professional institutions evaluate commercial equipment: that is, it measures the actual transaction price of this batch of hardware in the market after the borrower defaults.
Galbraith said: "We are building graphics cards into standardized, low-volatility, stable assets. When the insurance industry evaluates assets and risks, the more unified, more conventional, and more stable the value trend, the easier it is to be accepted and recognized by the industry."
He pointed out that the cost of renting graphics cards for AI companies has fluctuated wildly and is still rising, but the price trend of second-hand resale of graphics cards is much more stable.
Although there is asset depreciation in graphics cards, the depreciation pattern is clear and predictable. This is what insurance institutions and lending institutions value most. Take the H100 graphics card launched by Nvidia in 2023 as an example. Its initial price is about US$30,000. Today, the unit price in the second-hand market remains in the range of US$25,000 to US$30,000, which is far lower than the previous price peak of over US$40,000.
As early as the beginning of the 21st century, Galbraith was deeply involved in the field of asset valuation. At that time, he worked for a global insurance company, specializing in appraising the value of art and various luxury goods for high-net-worth clients. The company name Barkr is derived from this obsession and pays tribute to the late host Bob Barker of the classic price guessing variety show "Price Guess".
Nvidia secretly expresses support
After the forum, Galbraith went to Santa Clara to visit NVIDIA's headquarters and held talks with NVIDIA Vice President of Artificial Intelligence Infrastructure Nico Capres and members of its start-up investment team.
Nvidia has clearly seen that opening up chip financing channels contains huge market opportunities. According to Galbraith, Nvidia insiders have proactively connected him with a number of lending institutions interested in developing graphics card financing business.
NVIDIA is actively promoting this matter mainly for two major considerations: first, with the help of a neutral third-party agency, it has been confirmed that the value-preserving period of graphics card assets is much longer than some skeptical opinions have determined; second, to open up financing channels in the capital market, so that non-high-rated large-scale technology companies can also purchase graphics card computing resources in large quantities with the same preferential borrowing conditions.
Galbraith said bluntly: "We are solving an industry pain point that NVIDIA also attaches great importance to."