Circle Internet Group has applied for a national trust charter, aiming to establish a national trust bank called First National Digital Currency Bank, N.A. If the application is approved by the Office of the Comptroller of the Currency (OCC), the bank would oversee the management of the reserve of the USDC stablecoin on behalf of Circle’s U.S. issuer. First National Digital Currency Bank, N.A. would also offer digital asset custody services to institutional customers. For Circle, the charter would also help it meet the expected requirements of the proposed stablecoin legislation, the GENIUS Act. “By applying for a national trust charter, Circle is taking proactive steps to further strengthen our USDC infrastructure,” Circle Co-founder, Chairman and CEO Jeremy Allaire said. “Further, we will align with emerging U.S. regulation for the issuance and operation of dollar-denominated payment stablecoins, which we believe can enhance the reach and resilience of the U.S. dollar, and support the development of crucial, market neutral infrastructure for the world’s leading institutions to build on.” Becoming a publicly traded company requires Circle to comply with a higher standard of regulatory oversight, including audits, disclosures and governance practices.
Kraken and Bybit listing their tokenized U.S. stocks just two hours apart indicates growing momentum behind tokenized finance and the broader ambition to decentralize access to traditional markets
In a sign of growing momentum behind tokenized finance, two major crypto exchanges, Kraken and Bybit, unveiled their listings of tokenized U.S. stocks just two hours apart. Kraken is launching 60 tokenized equities under the xStocks brand, powered by Swiss issuer Backed. The offering includes prominent names like Apple, Tesla, and ETFs such as SPY. Two hours later, Bybit, currently the second-largest exchange by crypto trading volume, announced the same product integration on its Spot platform. Kraken’s launch signals a broader ambition to decentralize access to traditional markets. Its xStocks are built on the Solana blockchain and allow users not only to trade them on the exchange but also to withdraw them to self-custody wallets. From there, users can deploy them as collateral across decentralized finance protocols, something conventional stocks can’t match. The exchange plans to expand access to xStocks across more than 185 countries in the coming weeks, with support for additional blockchains to follow. Bybit’s listing supports Ethereum (ERC-20) and Solana (SPL) versions of xStocks, and includes the same basket of high-demand equities. Emily Bao, Bybit’s Head of Spot, said the exchange aims to provide users with more control and choice while remaining within the crypto ecosystem. xStocks offer features such as traditional equities can’t, fractional ownership, on-chain mobility, and round-the-clock trading. By listing them nearly simultaneously, Kraken and Bybit are positioning themselves at the frontier of financial infrastructure. Meanwhile, Robinhood also announced the launch of tokenized versions of U.S.-listed stocks and ETFs, besides a blockchain network.
OpenLedger enables deploying thousands of fine-tuned models using a single GPU without preloading them, by dynamically merging and infering on demand using quantization, flash attention, and tensor parallelism to offer 90% savings in deployment costs
OpenLedger has launched OpenLoRA, a new open protocol that enables developers to deploy thousands of LoRA fine-tuned models using a single GPU, saving up to 90% of deployment costs. Built on cutting-edge research and an open-source foundation, OpenLoRA allows developers to serve thousands of LoRA models on one GPU without preloading them, dynamically merging and infering on demand using quantization, flash attention, and tensor parallelism. This means builders can now scale AI deployment without bloating compute bills. Deployed as a SaaS platform, OpenLoRA makes it radically easier for startups and enterprises alike to launch AI products across verticals, from marketing, legal, education, crypto, customer service, and beyond, without having to replicate the entire model architecture for each use case. It’s a paradigm shift in how fine-tuned intelligence can be deployed at scale. Ram, Core Contributor at OpenLedger said, “With OpenLoRA, we’re redefining the economics of AI deployment, offering the first protocol where developers can serve massive fleets of fine-tuned models with minimal cost and maximum performance.”
Success of Pix and UPI is paving way for a three-stage framework for state-led fast payment systems that involves weighting pre-requisites, implementation and scaling and establishing engagement mechanisms and regulatory adjustments
Pix and Unified Payments Interface (UPI), Brazil and India’s respective instant payment systems, provide two key lessons for governments interested in implementing new fast or immediate payment systems. First, the significant effect that government-led instant payment systems can have on citizens and the financial market transforms financial inclusion and market structures. Second, decisions made during the early stages of the process, such as system pricing and ownership structure, shape the power dynamics between local and international players, as well as incumbent and new entrants. These lessons are shaping an emerging framework governments can use to evaluate their need for central bank-led immediate payment systems, their potential structure, organizational features, and trade-offs involved in implementing a similar approach. The framework is composed of a three-step approach, including prerequisite weighting (i.e., “do we need this system”), the preparations needed to hit the ground running, and the process of setting up new immediate payment systems.
Researchers from MIT, McGill University, ETH Zurich, Johns Hopkins University, Yale and the Mila-Quebec Artificial Intelligence Institute have developed a new method for ensuring that AI-generated codes are more accurate and useful. In the paper, the researchers used Sequential Monte Carlo (SMC) to “tackle a number of challenging semantic parsing problems, guiding generation with incremental static and dynamic analysis.” Sequential Monte Carlo refers to a family of algorithms that help figure out solutions to filtering problems. This method spans various programming languages and instructs the LLM to adhere to the rules of each language. The group found that by adapting new sampling methods, AI models can be guided to follow programming language rules and even enhance the performance of small language models (SLMs), which are typically used for code generation, surpassing that of large language models. João Loula, co-lead writer of the paper, said that the method “could improve programming assistants, AI-powered data analysis and scientific discovery tools.” It can also cut compute costs and be more efficient than reranking methods. Key features of adapting SMC sampling to model generation include proposal distribution where the token-by-token sampling is guided by cheap constraints, important weights that correct for biases and resampling which reallocates compute effort towards partial generations.
Crowdsourced AI benchmarks should be dynamic rather than static datasets, and tailored specifically to distinct use casesAI benchmarks
Over the past few years, labs including OpenAI, Google, and Meta have turned to platforms that recruit users to help evaluate upcoming models’ capabilities. When a model scores favorably, the lab behind it will often tout that score as evidence of a meaningful improvement. It’s a flawed approach, however, according to Emily Bender, a University of Washington linguistics professor and co-author of the book “The AI Con.” Bender takes particular issue with Chatbot Arena, which tasks volunteers with prompting two anonymous models and selecting the response they prefer. To be valid, a benchmark needs to measure something specific, and it needs to have construct validity — that is, there has to be evidence that the construct of interest is well-defined and that the measurements actually relate to the construct,” Bender said. “Chatbot Arena hasn’t shown that voting for one output over another actually correlates with preferences, however they may be defined.” Asmelash Teka Hadgu, the co-founder of AI firm Lesan and a fellow at the Distributed AI Research Institute, said that he thinks benchmarks like Chatbot Arena are being “co-opted” by AI labs to “promote exaggerated claims.” Benchmarks should be dynamic rather than static datasets,” Hadgu said, “distributed across multiple independent entities, such as organizations or universities, and tailored specifically to distinct use cases, like education, healthcare, and other fields done by practicing professionals who use these [models] for work.” Wei-Lin Chiang, an AI doctoral student at UC Berkeley and one of the founders of LMArena, which maintains Chatbot Arena said that incidents such as the Maverick benchmark discrepancy aren’t the result of a flaw in Chatbot Arena’s design, but rather labs misinterpreting its policy.
Capital One 1Q25: Credit card purchase volume is up 5%, auto loan originations are up 22%
Capital One Financial Corporation announced net income for the first quarter of 2025 of $1.4 billion, or $3.45 per diluted common share, compared with net income of $1.1 billion, or $2.67 per diluted common share in the fourth quarter of 2024, and with net income of $1.3 billion, or $3.13 per diluted common share in the first quarter of 2024. Adjusted net income(1) for the first quarter of 2025 was $4.06 per diluted common share. “Last week, we received regulatory approval for our acquisition of Discover and we’re fully mobilized to complete the transaction on May 18th,” said Richard D. Fairbank, Founder, Chairman, and Chief Executive Officer. “The combination of Capital One and Discover will create a leading consumer banking and payments platform with unique capabilities, modern technology, and powerful brands. It leverages Capital One’s technology transformation and digital capabilities across a significantly larger customer franchise. And it offers the potential to enhance competition and create significant value for merchants and customers.
Credit Card
- Ending loans held for investment up $6.6 billion, or 4%, year-over-year; average loans held for investment up $6.8 billion, or 5%, year-over-year
- Purchase volume up 5% year-over-year
- Revenue up $417 million, or 6%, year over-year
Consumer Banking
- Ending loans held for investment up $3.8 billion or 5% year-over-year; average loans held for investment up $3.4 billion, or 5%, year-over-year
- Ending deposits up $24.1 billion, or 8%, year-over-year
- Auto loan originations up $1.7 billion, or 22%, year-over-year
A huge portion of the integration will involve Capital One bringing Discover up to speed on the technology infrastructure it has spent years modernizing; however, running a payment network is new to Capital
Days after winning regulatory approval for its blockbuster acquisition of Discover Financial Services, Capital One Financial said that its expectations for what the integration will cost haven’t changed. The $35 billion transaction has been and will continue to be costly, but Capital One Chairman and CEO Richard Fairbank said that the $1.5 billion estimate for integration expenses during 2027 remain intact — except shifted out by about six months to account for the deal’s longer-than-expected regulatory review. Fairbank told that he thinks this transaction is different from other acquisitions, where the goal is “to take two companies, squash them together and rip out the costs.” “I think that Discover brings us a growth platform, both on the network side and with respect to their card franchise, that allows us to preserve the best of what they do, leverage a lot of Capital One’s capabilities that we bring and build something really special,” Fairbank said. Upon the closing of the Discover merger, the combined company will have $660 billion of assets. Capital One will own a massive chunk — estimated to be between one-fourth and one-third — of the subprime card market. And it will operate Discover’s payment network, instead of having to use Visa’s or Mastercard’s — an element of the transaction that Fairbank has called “the holy grail.” But the road to getting there isn’t completely nailed down. A huge portion of the integration will involve Capital One bringing Discover up to speed on the technology infrastructure it has spent years modernizing. Meanwhile, Discover will take Capital One “back to the world of data centers,” Fairbank said. He added that his bank has experience ramping up the tech stack of a credit card company. However, running a payment network is new to Capital One, and “very complex and very high stakes,” Fairbank said. Another key facet of the merger is Capital One’s effort to increase acceptance of Discover’s payment network internationally. Fairbank characterized this spending as a long-haul type of investment, measured in “a whole bunch of years.” Kyle Sanders, an analyst at Edward Jones, thinks it will take several years for the merger’s benefits to manifest themselves, and that near-term integration challenges “will present obstacles.” Last week, the deal earned the approval of the Federal Reserve and Office of the Comptroller of the Currency, but regulators ordered Discover to pay more than $1 billion in fines and restitution in connection with the company’s earlier overcharging of merchants. When asked about recent regulatory developments, Fairbank said that Capital One knew risk management would be “a big investment,” but the company hasn’t changed its outlook on how much those efforts will cost.
TD announces a new Layer 6 office in New York City to . drive the bank’s ability to deploy advanced Machine Learning solutions
TD Bank Group announced it will open a new Layer 6 office in New York City. As TD’s AI research and development center, Layer 6 has driven TD’s ability to deploy advanced Machine Learning solutions since it was acquired by TD in 2018. Currently operating from its head office in Toronto’s MARS Innovation District, the new Layer 6 office will now grow to more closely support TD Bank, America’s Most Convenient Bank®, as well as TD’s other U.S. based operations. It will also allow TD to take advantage of an expanded pool of world-class talent and further cement its leadership and competitive advantage driving innovation in banking through AI. The Layer 6 office will formally open later in 2025 with a mixture of an initial 20 data scientists, applied machine learning scientists, GenAI implementation specialists, and others, who will sit at TD’s New York office, One Vanderbilt. “Our U.S. expansion of Layer 6 underscores our commitment to deepening our presence in New York City and investing in the future of innovation,” said Leo Salom, President and CEO of TD Bank. “The new Layer 6 office establishes a strong foundation for advancing our GenAI capabilities and bringing critical expertise and delivery in-house.”
Q1 2025 Homeownership Program Index (HPI) reports the number of entities offering homebuyer assistance programs increase by 55, offering more ways to qualify buyers and close loans in a tough market
Down Payment Resource (DPR) released its Q1 2025 Homeownership Program Index (HPI) report, which saw the number of entities offering homebuyer assistance programs increase by 55 year-over-year. The number of programs increased by 43 during the first quarter, bringing the total number of available programs to 2,509 — the highest recorded by DPR. That marks a 2% increase from Q4 2024. Of the programs, 952 programs (38%) are available to repeat buyers, 240 programs (10%) do not have income restrictions and 19 programs support first-generation homebuyers, an increase of 16% over the last quarter. Lenders can use down payment assistance (DPA) to lower a homebuyer’s loan-to-value (LTV) ratio by an average of 6%. The average benefit is $18,000. “Rates are still high and prices keep climbing, but we’re seeing expanded program offerings, new providers and greater flexibility in how funds are used — not just for down payments but also to cover closing costs, lower the rate or meet other buyer needs,” said Rob Chrane, founder and CEO of DPR. “More programs now include manufactured and multi-family homes, opening new paths to affordability and steady income. For lenders, that means more ways to qualify buyers and close loans in a tough market.” ”Other homebuyer assistance” programs increased 35% from the previous quarter, below-market-rate (BMR)/resale-restricted programs, which offer housing at prices lower than the open market, with restrictions on resale to ensure affordability for future buyers, typically low-to moderate-income households, increased 18% and grant programs grew 7%. Other stats:
- 80% of DPAs in Q1 were deferred payment programs, a 3% increase from the previous quarter. Deferred payment loans, which are often forgivable, mean that borrowers don’t make monthly payments, and the balance is typically due when they sell or refinance or the loan matures.
- Over half (53%) of DPAs in Q1 offered partial or full forgiveness over time, as long as the homeowner meets certain requirements, such as maintaining primary residency.
- Of the programs, 990 (39%) were offered through local housing finance agencies (HFAs), a number that was virtually unchanged from the previous quarter. Nonprofits accounted for 21%, a 2% increase over Q4 2024. State FHAs represented 18%.
- Manufactured housing programs saw growth, increasing from 914 in Q4 2024 to 971 in Q1 2025. For multifamily housing, a total of 833 programs were available, marking a 3% increase from Q4 2024. Of these, a growing number of programs support purchasing three-unit homes (562) and four-unit homes (536).
- A total of 20 programs offered special funding to surviving military spouses, an 18% increase from the previous quarter, while energy efficiency programs grew by 17%. Other incentive programs included 69 for educators, 56 for protectors (jobs focused on safeguarding people, property or information), 50 to assist military veterans and 50 for Native Americans.
- Of the 2,509 homebuyer assistance programs, 81% of programs are funded, 10% of programs are inactive, 4% of programs have a waitlist for funding and 5% of programs are temporarily suspended.
- Of the programs, 74% in the database are for down payment or closing cost assistance, 10% of programs are first mortgages, 3% of programs are Mortgage Credit Certificates (MCCs) and 13% are other program types.