Can You Really Buy A House Without Selling Your Crypto? | Vishal Garg

Vishal Garg frames the US mortgage market as a 15 trillion dollar system where 400 basis points of spread between mortgage yields and deposit funding costs generates roughly 600 billion dollars per year in intermediation, with 300 basis points of that consumed by internal banking costs. He estimates US households lose approximately 450 billion dollars annually to the mortgage document manufacturing process, or about 4,500 dollars per homeowning household per year. The industry spends roughly 12,000 dollars to originate a single loan, with about 9,500 dollars of that being labor, and large banks like Wells Fargo, Chase, and Bank of America have pulled back partly because their compliance-driven cost per loan reached approximately 15,000 dollars.

Better has originated 110 billion dollars in mortgages over roughly 10 years and has driven its marginal origination cost below 2,000 dollars. Its direct-to-consumer channel costs about 6,000 dollars per loan all-in, with 4,000 dollars of that being customer acquisition. Better sells originated loans at roughly a 2 percent premium, generating around 2,000 dollars in profit on a 400,000 dollar mortgage. The company and its partners are running at approximately 8 billion dollars in annualized origination volume, doubling year over year, and the platform and partnership business now accounts for over 50 percent of revenue. Platform clients include Credit Karma, Coinbase, Finance of America, and Lending Club, with the platform generating between 2,000 and 4,000 dollars per loan depending on services used.

Better uses AI loan officers, processors, and underwriters to compress underwriting from 21 days or more to minutes. Garg notes that AI disruption in mortgages operates in the reasoning, communication, and orchestration layer rather than the underwriting rules layer, because those rules are preset by Fannie Mae, Freddie Mac, FHA, and VA. The core problem AI solves is that investors have up to 45 sets of guidelines running roughly 800 pages each, which no human underwriter earning an average of 210,000 dollars per year can fully memorize or apply consistently. Better has built a digital twin of a top loan officer that replicates his knowledge and communication style while knowing all 45 investor guidelines. A ChatGPT-powered product called Tin Man lets mortgage brokers get rate quotes and affordability estimates in real time at open houses, and Garg says approximately 6,000 banks want to re-enter the mortgage business using it without needing specialized loan officer training.

The headline product is a Fannie Mae eligible Bitcoin-backed mortgage developed with Coinbase. Coinbase built a triparty pledge mechanism via smart contract on Coinbase custody, allowing borrowers to pledge Bitcoin or USDC as a substitute for a down payment without selling the crypto. The product does not include margin calls as long as mortgage payments are made on time, which Garg identifies as a key differentiator from competing crypto mortgage products that require pledging 100 percent of the home value and do include margin call features. A couple in Michigan used the product to buy a house at a six and a half percent fixed rate with 100 percent financing. Garg notes that if Bitcoin appreciates more than 8 percent per year, the pledged collateral effectively pays for the house, and the rate is materially lower than other crypto loans of comparable size, which carry rates of 9 to 11 percent.

Garg argues the biggest impediment to homeownership is accumulating a down payment rather than servicing monthly payments, pointing out that US households hold roughly 5 trillion dollars in checking and savings accounts but approximately 35 trillion dollars in stocks, bonds, and digital assets. The longer-term ambition is to accept stocks, bonds, currencies, and commodities as collateral, so that an employee paid in restricted stock units could buy a house by pledging that stock rather than liquidating it. Fannie Mae and Freddie Mac leadership under the new administration are described as encouraging about accepting token-backed mortgages and tokenizing those mortgages themselves, which Garg treats as meaningful institutional validation.

Better partnered with Sky and Framework to access a 500 million dollar credit line backed by tokenized mortgages, though the facility is not yet live. A 100 basis point reduction in cost of capital from tokenization could lower mortgage rates by 50 to 100 basis points and improve affordability by approximately 1,000 dollars per month. Garg contrasts this with the publicly traded MBB ETF, which holds about 40 billion dollars in agency mortgages at a net yield of 4.1 percent while the underlying assets yield around 6 percent, with the roughly 2.5 percent gap lost to layers of intermediation.

Garg says Better could reach profitability without blockchain but not without AI, and expects to break even on an adjusted basis by September. The company has approximately 150 core corporate employees and over 1,000 total including brokers, processors, and data labeling staff. He describes the data labeling operation as a core competitive moat because it is bespoke to the mortgage process and expands with every loan originated, with each mortgage creating both a financial asset and a context graph of cleansed borrower and property data. He believes the stock, trading at roughly two to two and a half times run rate sales, is significantly undervalued relative to AI-focused peers and has been personally buying shares.