Robot analyzing the stock market
In case you have just returned from another planet or been meditating in isolation atop a high mountain for the past year and haven’t seen the news, Artificial Intelligence (AI) is all the rage.
ChatGPT broke out in late 2023 as the new kid on the block with its GenAI (Generative AI) product and the AI industry has not looked back, attracting an estimated half of the global VC funding in Q4 2024 in preliminary data from Pitchbook.
Tech leaders are lining up with the new U.S. Administration to fund AI. The Stargate Project is a new company which intends to invest $500 billion over the next four years, funded by SoftBank, OpenAI, Oracle, and MGX, to build new AI infrastructure for OpenAI in the United States.
If you are an investor of any ilk and like to do your own research, you can update your toolkit and have your own cadre of “AI agents” provide you with the answers to questions you require to help you to make investment decisions to meet your specific needs.
If you leave invest guidance or advice to your advisor, market analysts, investment manager, or fund manager, and you want to know how AI is helping them, read on. If you don’t care about any of this or how AI is profoundly transforming financial services, you can stop reading now.
The AI Era
Invented in the 1970s, neural networks were dismissed as useless for decades. A detailed history of the rise of deep learning can be found in this lecture by Nobel Prize winner Geffrey Hinton, the key people are Hinton, Ilya Suskever, and Professor Feifei Li.
Jensen Huang of Nvidia started general purpose graphics computing (CUDA) in 1999, against great internal and external dissent. At one point an activist investor took a stake in the firm to get the firm to reverse course because they saw CUDA as useless and reducing profit margin.
For the next 13 years, until deep learning took off in 2012, Nvidia owned “100 percent of a zero-billion-dollar market”. It really took off when ChatGPT came out in. The rest is history.
A Generative AI model, like ChatGPT, is known as a Generative Pre-trained Transformer (GPT). It’s a large artificial neural network that’s trained on trillions of “tokens” of training data with one seemingly simple objective: given a randomly sampled sentence, predict the next token.
The vast corpus of digital data input (on the internet) covers all domains of human knowledge in dozens of languages, including our human fallibilities and biases. It allows a GPT to learn a comprehensive “world model”. Unlike a conventional AI/ML (machine learning) algorithm that’s designed to solve a specific problem in classification or prediction, A GPT is extremely versatile and powerful at solving problems in different areas.
With the scale and (lower) costs of vast computational usage, GenAI products can answer questions more quickly and effectively. You simply type your question and within seconds you get back a relatively accurate response.
If you have not already used GenAI already, go check it out yourself right now in the latest version of ChatGPT. Don’t forget to come back here, it’s easy for geeks to go down the rabbit hole and get mildly obsessed with this technology, in which case you may wish to speak to a GenAI therapist, which is proving popular with many.
The training of a GPT follows the pre-training scaling law which describes how the model performance improves when each of the following increases: model size, training data size, and amount of computing.
This scaling shows that “thinking” during the inference time is an extremely efficient way of improving the model performance. A model thinks via multiple rounds of self-review to iteratively improve the quality of the output.
By thinking 15x longer, a model can achieve the same performance as a model that’s 10x bigger in pre-training.
The recent announcement that a team of Chinese researchers has created an open-source Large Language Model (LLM) drove declines in AI-related tech shares on the Nasdaq. The new DeepSeek model delivers high-quality results at a fraction of the cost.
Investors are concerned that DeepSeek’s ability to process results 10-30 times cheaper could pop the AI bubble.
Bin Ren, the ceo of Sigtech, a leading AI solution provider to the financial services sector, disagrees with investor’s concerns on this arguing it is likely to have the opposite effect, and adding, “If anything, DeepSeek’s success shows that we can expect AI models to deliver extraordinary results at an affordable price faster than ever before.
“As tech firms copy the efficiency gains championed by DeepSeek, it will mean that every $1 billion investment in AI in the U.S. will result in an extra 10-30 fold gain in intelligence output.
“Like every great innovation, from electricity to the mobile phone, the marginal utility increases while the marginal cost decreases.”
Ren know what he is talking about. He is the former head of quantitative investment for hedge fund Brevan Howard, and spun out the SIG – Systematic Investment Group into Sigtech to deliver AI solutions to asset managers and funds seven years ago.
Its new offering MAGIC (Multi-Agent Generative Investment Copilots) is a user-friendly investment tool designed to answer all your financial questions, and is the current state of the art.
AI In Capital Markets
There are thousands of companies listed on the U.S. stock market’s various exchanges. As of August 2024, there were about 3,450 different stocks listed on the Nasdaq, 2,240 on the NYSE, and almost 6,500 on the OTC markets.
In recent years, the research analyst industry has been in decline with the reported number of stock analysts at the world’s 15 largest banks having reduced from 4,600 a decade ago to about 3,000 in 2025, a drop of more than 30 percent.
There are more public companies to cover but far fewer analysts. Financial analysis, the role filled by analysts, is the number one AI use case in financial services today.
AI agents are extremely well suited to analyzing public market disclosures, financials, and extant data to crunch numbers to improve decision-making, reduce risk, increase operational efficiency, and ensure adherence to compliance standards. AI Agents make great “co-pilots” to assist investors with analysis, evaluation, and ultimately, making their investment decisions.
An AI agent is an intelligent agent powered by a LLM, that can work on specific assigned task, like a “Central Bank Interest Rate Agent.” An AI agent can use tools like API services, code execution, and web search and browsing, and can be part of a multi-agent environment managed by an agent coordinator designed for specific jobs, like portfolio optimization guidance.
Top financial institutions like Goldman Sachs, JP Morgan, and Blackrock are leveraging AI in highly regulated environments like banking and asset management which require compliance within strict laws and regulations.
Goldman Sachs recently reported using AI to write 95 percent of its S1 filings, the initial registration form for new securities required by the SEC for public companies that are based in the U.S.
Law makers and regulators should be welcoming AI with open arms for the financial services sector. AI solutions are likely much more efficient at processing thousands of pages of laws, regulation, and rules than humans are.
Ask any regular weekend golfer how many rules there are in the game of golf and if he or she has read the Rules of Golf. The 2023 compendium comprise 25 rules (down from 34) with multiple subsections contained within each of them. The section of the book that covers those 25 rules comprises 192 pages and the full rule book, including contents, definitions, and index is 256 pages.
AI solutions are likely much more efficient at processing the Rules of Golf than regular weekend golfers, a high percentage of whom admit to not following the rules, because they don’t know them.
Many books that offer to be your practical guide to U.S. securities laws, acts, rules and regulations (since 1933) are over 300 pages, no doubt one of the reasons for the vast number of securities lawyers at your disposal.
“Generative AI is set to revolutionize financial services by delivering personalized investment recommendations, sharper risk assessments, and more efficient operations. With deep learning architectures becoming increasingly cost-effective, banks and fintechs can now parse massive datasets for real-time insights. The key, though, is making sure we can trust the process—which is where verifiable AI truly shines, says Felix Xu, co-founder and CEO of ARPA Network and Bella Protocol.
Xu adds, “By leveraging cryptographic techniques like zero-knowledge proofs, we can validate AI outputs without revealing sensitive data, enabling faster fraud detection and smoother regulatory compliance. Institutions that embrace these trust-enabling technologies will gain a decisive edge in a fiercely competitive landscape.
“As new language models continue to emerge worldwide, ensuring their security and verifiability will be pivotal to delivering transparent, reliable financial services that keep customers and regulators on board.”
Agentic AI, systems designed to make autonomous decisions, are a key component of algorithmic finance at many leading hedge funds, asset managers, and trading firms like Renaissance, AQR, Citadel, DE Shaw, and Two Sigma.
These firms use AI to automate trading, enhance risk management, and optimize decision-making at speeds and scales that are not possible with human traders. With advances in machine learning and reinforcement learning, agentic AI is expected to continue growing with the segment of players.
Small Language Models (SLMs) are also becoming increasingly important in the AI landscape as companies gain experience with the benefits of SLMs for targeted, efficient, and cost-effective AI solutions for specific tasks while using fewer resources than larger models. A host of tech firms like Google, Microsoft, Samsung, Apple, Mistral AI, and Cohere are reported to be developing SLMs for industries.
SLMs application to smaller, accurate, and more discrete data sets, like domain laws, regulator’s rule books, company disclosures, and extant data, make it a great candidate for specific jobs in regulated financial markets.
AI’s role in the emerging Web3 DeFi community is also an important one to pay attention to says Tom Ngo, executive lead at Metis L2, “AI agents are automating many aspects of DeFi while fundamentally transforming how we think about and act upon market efficiency, risk management, and value creation.
“They will optimize liquidity across protocols, predict market inefficiencies before they occur, and execute complex strategies in milliseconds. Aside from faster trading, it’s also helping to establish an entirely new financial system that maximizes every unit of capital through intelligent automation. This is why we’re building infrastructure that enables AI to operate at the speed and scale DeFi demands.”
AI For Investors
62 percent of adults in the U.S. invest in the stock market and this figure has remained steady over the last few years which is still below the levels in 2007 when it peaked at 65 percent.
Approximately 40 percent of affluent Americans conduct their own research and fully manage their investments using online platforms according to Bank of America. This indicates that a significant portion of affluent individuals in the U.S. engage in self-directed financial research and management.
While comprehensive data on the U.S. adults who perform their own financial research remains relatively limited, the evidence of the trend toward self-directed investing is growing, especially among younger generations, and an important consideration with the greatest wealth transfer in history now taking place.
A recent study by Fidelity Investments found that 71 percent of women own investments in the stock market, with younger women leading the way in embracing investing and taking control of their finances.
With thousands of stocks in the U.S. stock markets, GenAI is increasingly being used by investors to support decisions around self-directed investments though it is not yet clear which of the Top 10 GenAI platforms, or emerging industry AI platforms, investors prefer.
Importantly, 59 percent of high-net-worth U.S investors ($1 million in investable assets) indicate that financial advisors are the most trusted source of financial advice, compared with 38 percent of the general public.
This is an indicator of how important humans are to financial advice, at least in the short to medium term, as is the role that GenAI plays in supporting human advisors with their client’s portfolio optimization and investment goals.
While there is to be excited about with new GenAI models, investors and regulators can be assured the financial services sector has vast experience with algorithmic finance and AI through its evolution over the years.
Markets were computerized in the 1970s, algorithmic finance (algos or quants) emerged in the 1990s, and AI/ML emerged in the 2000s. With the rise of deep learning in 2012, the ChatGPT breakthrough in 2023, the era of AI solutions that are easily usable by everyone in society is here, and this is great news for investors.
These new AI models are optimized and scaled along to deploy using the (massive) computing power required to deliver responses quickly. As this latest evolution of AI scales at pace, the increased transparency of the performance of investments is likely to profoundly change financial investing for many investors, old and new.
We must bear in mind that AI models, like humans, can predict how stocks might perform based on past market behavior, and AI might very well be more efficient, accurate, and unbiased, but the future is often difficult to predict.
The processing of massive amounts of past market behavior and data may not greatly help with accurately predicting the future. The impact of geopolitical and economic communications, policy, and other actions by humans on markets can be massive, and are a significant contributor to market volatility, making it more difficult to predict the future, even for new GenAI.
