AI Cuts Research Costs, Improves Stock Forecasts

Hey, it's Matt. Here’s what’s happening in AI + Wall Street.
Was this email forwarded to you? Sign up below:

ANALYSIS

Tech Giants Double Down on AI Spend After DeepSeek

The big story last week was that a much cheaper AI model from DeepSeek was going to obviate the need to fund massive computing infrastructure.

Well, the biggest companies in the world aren’t pulling back—they’re doubling down.

• Meta (META), Microsoft (MSFT), and Google parent Alphabet (GOOG) are expecting a cumulative $228 billion in capital expenditures in 2025, driven by their investments in artificial intelligence infrastructure. That’s a 55% increase from the roughly $150 billion those companies reported spending in 2024. (Yahoo Finance)

The companies are focused on building out their AI infrastructure with monetization plans to follow. For end users, this means AI models will continue to get better, cheaper and more accessible, which bodes well for adoption.

HEADLINES

OpenAI Says New AI Tool Works Like Research Analyst

OpenAI has launched a new feature called Deep Research, which it claims conducts multi-step research for complex tasks, compiling comprehensive reports in minutes, normally work that would take a human hours. 

Currently available to ChatGPT Pro users for $200/month, with OpenAI planning a wider rollout at $20/month. (The Verge)

The price of research/intelligence/whatever-you-want-to-call-it just keeps dropping. These tools aren’t perfect, but they keep getting better. I can see lots of use cases for this in financial research.

AI Writes Better Emails Than Humans: Allstate CEO

Allstate’s 15,000 customer service agents now use AI to craft claim-related emails, leading to better outcomes and reducing the demand for new hires. (Bloomberg)

Much has been said about AI replacing jobs, but that assumes “work” is finite. As the story above points out, it can free up staff from lower-paid tasks.

Perplexity Adds FactSet Data Integration

Perplexity partnered with AI startup Quartr in November to integrate live and archived earnings call transcripts into Perplexity. (Press Release)

Interesting to see Perplexity, which has mostly been a general AI provider, making a bigger push into the financial domain. As AI models become cheaper and more capable, they’re competing with traditional incumbents across multiple industries.

FUNDRAISING

AI Startup for Advisors, Jump, Raises $20 Million

The automated note-taking and workflow provider is working with many RIAs and is already an approved LPL vendor. (Wealth Management)

I’m always in between Zoom or Google Meet or Microsoft Teams so a tool that can integrate and transcribe meetings with all of these different platforms makes sense for financial advisors.

AI Legal Startup Semeris Raises €4 Million

FinTech Semeris has secured €4 million in funding to scale its AI-powered platform that helps analyze legal documents in debt and structured finance markets, with the company claiming to make document review 10 times faster. (EU Startups)

I’ve written a lot about how AI can turn unstructured data — like PDFs, emails — into structured formats like Excel. Lots of finance is buried in a sea of documents.

RESEARCH

AI Boosts Analysts’ Speed and Accuracy: Study

Investment banks that invest more in AI see their analysts make more frequent and accurate forecasts, according to a new academic study.

The research examined AI investments at 15 major banks by tracking job postings mentioning machine learning and neural networks.

Researchers then analyzed analysts’ forecasts before and after ChatGPT’s November 2022 launch using the Institutional Brokers’ Estimate System, a database that tracks professional stock analysts’ predictions.

The investments led to a 1.4% increase in earnings forecast frequency and 2.3% improvement in earnings forecast accuracy, with even larger gains (2.8%) for metrics like sales and cash flow forecasts.

AI complements rather than replaces analyst skills, with the biggest benefits on complex analytical tasks - particularly for companies with high uncertainty or limited public data, according to the study. Lower-performing analysts improved the most, while investment banks expanded coverage to previously overlooked companies.

I asked one of the study's authors, Devin Shanthikumar, a professor at the University of California, Irvine, for some key takeaways:

For banks:

𝘛𝘩𝘦𝘳𝘦 𝘪𝘴 𝘢 𝘭𝘰𝘵 𝘰𝘧 𝘰𝘱𝘱𝘰𝘳𝘵𝘶𝘯𝘪𝘵𝘺 𝘵𝘰 𝘶𝘴𝘦 𝘈𝘐 𝘵𝘰 𝘮𝘢𝘬𝘦 𝘤𝘶𝘳𝘳𝘦𝘯𝘵 𝘦𝘮𝘱𝘭𝘰𝘺𝘦𝘦𝘴 𝘢𝘯𝘥 𝘣𝘶𝘴𝘪𝘯𝘦𝘴𝘴𝘦𝘴 𝘮𝘰𝘳𝘦 𝘦𝘧𝘧𝘪𝘤𝘪𝘦𝘯𝘵 𝘢𝘯𝘥 𝘦𝘧𝘧𝘦𝘤𝘵𝘪𝘷𝘦. 𝘠𝘰𝘶 𝘥𝘰𝘯'𝘵 𝘩𝘢𝘷𝘦 𝘵𝘰 𝘤𝘳𝘦𝘢𝘵𝘦 𝘢 𝘴𝘵𝘢𝘯𝘥-𝘢𝘭𝘰𝘯𝘦 𝘈𝘐 𝘱𝘳𝘰𝘥𝘶𝘤𝘵 𝘧𝘰𝘳 𝘈𝘐 𝘵𝘰 𝘣𝘦 𝘶𝘴𝘦𝘧𝘶𝘭 - 𝘪𝘵 𝘤𝘢𝘯 𝘣𝘦 𝘶𝘴𝘦𝘧𝘶𝘭 𝘵𝘰 𝘢𝘭𝘭𝘰𝘸 𝘺𝘰𝘶𝘳 𝘤𝘶𝘳𝘳𝘦𝘯𝘵 𝘦𝘮𝘱𝘭𝘰𝘺𝘦𝘦𝘴 𝘵𝘰 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘦 𝘴𝘰𝘮𝘦 𝘰𝘧 𝘵𝘩𝘦 𝘮𝘶𝘯𝘥𝘢𝘯𝘦 𝘵𝘢𝘴𝘬𝘴, 𝘢𝘯𝘥 𝘧𝘰𝘤𝘶𝘴 𝘮𝘰𝘳𝘦 𝘰𝘧 𝘵𝘩𝘦𝘪𝘳 𝘵𝘪𝘮𝘦 𝘢𝘯𝘥 𝘢𝘵𝘵𝘦𝘯𝘵𝘪𝘰𝘯 𝘰𝘯 𝘸𝘩𝘢𝘵 𝘵𝘩𝘦𝘺 𝘥𝘰 𝘣𝘦𝘴𝘵.

For analysts:

𝘈𝘯𝘢𝘭𝘺𝘴𝘵𝘴 𝘢𝘥𝘢𝘱𝘵𝘪𝘯𝘨 𝘵𝘰 𝘈𝘐 𝘵𝘰𝘰𝘭𝘴 𝘪𝘯 𝘵𝘩𝘦𝘪𝘳 𝘸𝘰𝘳𝘬 - 𝘖𝘶𝘳 𝘱𝘢𝘱𝘦𝘳 𝘩𝘪𝘨𝘩𝘭𝘪𝘨𝘩𝘵𝘴 𝘵𝘩𝘢𝘵 𝘈𝘐 𝘪𝘴 𝘮𝘰𝘴𝘵 𝘶𝘴𝘦𝘧𝘶𝘭 𝘧𝘰𝘳 𝘢𝘯𝘢𝘭𝘺𝘴𝘵𝘴 𝘵𝘩𝘳𝘰𝘶𝘨𝘩 𝘸𝘩𝘢𝘵 𝘸𝘦 𝘤𝘢𝘭𝘭 "𝘵𝘢𝘴𝘬 𝘤𝘰𝘮𝘱𝘭𝘦𝘮𝘦𝘯𝘵𝘢𝘳𝘪𝘵𝘺" - 𝘶𝘴𝘪𝘯𝘨 𝘈𝘐 𝘵𝘰 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘦 𝘰𝘳 𝘢𝘴𝘴𝘪𝘴𝘵 𝘸𝘪𝘵𝘩 𝘮𝘰𝘳𝘦 𝘮𝘶𝘯𝘥𝘢𝘯𝘦 𝘵𝘢𝘴𝘬𝘴, 𝘢𝘭𝘭𝘰𝘸𝘪𝘯𝘨 𝘢𝘯𝘢𝘭𝘺𝘴𝘵𝘴 𝘵𝘰 𝘧𝘰𝘤𝘶𝘴 𝘰𝘯 𝘸𝘩𝘢𝘵 𝘵𝘩𝘦𝘺 𝘥𝘰 𝘣𝘦𝘴𝘵 - 𝘵𝘩𝘦 𝘩𝘶𝘮𝘢𝘯 𝘢𝘴𝘱𝘦𝘤𝘵𝘴 𝘰𝘧 𝘨𝘢𝘵𝘩𝘦𝘳𝘪𝘯𝘨 𝘢𝘯𝘥 𝘪𝘯𝘵𝘦𝘳𝘱𝘳𝘦𝘵𝘪𝘯𝘨 𝘪𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘰𝘯. 𝘈𝘐 𝘥𝘰𝘦𝘴𝘯'𝘵 𝘳𝘦𝘱𝘭𝘢𝘤𝘦 𝘵𝘩𝘦 𝘸𝘢𝘺𝘴 𝘵𝘩𝘢𝘵 𝘢𝘯𝘢𝘭𝘺𝘴𝘵𝘴 𝘢𝘥𝘥 𝘷𝘢𝘭𝘶𝘦, 𝘪𝘯𝘴𝘵𝘦𝘢𝘥 𝘪𝘵 𝘤𝘢𝘯 𝘩𝘦𝘭𝘱 𝘵𝘩𝘦𝘮 𝘵𝘰 𝘴𝘱𝘦𝘯𝘥 𝘮𝘰𝘳𝘦 𝘵𝘪𝘮𝘦 𝘰𝘯 𝘷𝘢𝘭𝘶𝘦-𝘢𝘥𝘥.

DELIVERY

Too many emails? Tweak your AI Street updates: AI news & analysis on Thursdays, tool deep dives on Sundays, or both. Update your preferences here.

USE CASE

OpenAI, DeepSeek Models Can Handle Tricky VC Math

A month ago, only OpenAI's model could handle complex venture capital calculations like figuring out investor ownership percentages across funding rounds, but now both OpenAI's O3 Mini and DeepSeek can solve these math problems correctly - with DeepSeek solving them nearly five times faster, according to BlueFlame.AI.

ADOPTION

Young Investors Embrace AI Financial Advice: Study

Younger investors are significantly more open to AI-generated financial advice, with 83% of those aged 25–34 saying they would use AI-assisted tools if it reduced costs, according to a new survey by AI fintech firm Aveni and YouGov.

The research highlights a growing generational divide, as only 33% of respondents over 55 feel the same way. The research, conducted in December 2024 with 2,023 UK-based adults, also found that 86% of younger professionals value AI’s role in improving access to financial advice. (IFA Magazine)

I can see a world where a lot more people rely on AI to make their investment choices. You can already do that now with GPT Portfolio.

PODCAST

Alpha Intelligence Podcast: Unpacking DeepSeek

Last week, I launched the Alpha Intelligence Podcast with my co-host, Francesco Fabozzi, featuring in-depth conversations with hedge fund managers, quants, and academics on the future of AI on Wall Street.

For the first episode, we spoke with Tharsis Souza, PhD, former Senior VP of Product Management at Two Sigma, Lecturer at Columbia University, and author of an upcoming O'Reilly book on LLMs' pitfalls.

Our discussion focuses on DeepSeek’s latest model, R1. While DeepSeek claims to have trained R1 for just $6 million without Nvidia’s most advanced chips, the reality is more complex.

• On the podcast, we break down why DeepSeek’s $5.5M training cost might not tell the full story

• We also talk about the big shift in AI compute and what it means for companies betting on infrastructure.

• Learn why "LLMs are the trappiest tool ever for finance people" and how to think about look-ahead bias when using AI in financial applications

• Hear why "the price of intelligence is going to zero" and how to identify opportunities that will become possible as models get stronger

Listen on Spotify or YouTube

MARKETS EDITION

Tracking DeepSeek on Earnings Calls: Zero to 100+ Mentions

Last Sunday, I used Aiera (an AI platform that analyzes earnings calls) to track how DeepSeek suddenly became the talk of the business world. After finding zero mentions before January 20th, it exploded to over 100 mentions post-Davos. I then used Anthropic’s Claude to analyze the transcripts to see how different industries were reacting to this AI breakthrough.

If you’d like to change your settings and start getting these tool breakdowns, sign up here.

WHAT ELSE I’M READING

DeepSeek Is Just the Latest Hedge Fund Innovation

From the telescope to the telegraph to AI, finance has promoted the use of new technology. (Bloomberg)

• Finance is mostly seen as a way to fund innovation, yet it often also subsidizes innovation. This goes back a long way. Cosmologist Edward Kolb observes that although the telescope was immediately recognized as an important tool following its invention in 1608, its first application was in finance. “With the aid of a telescope, one could get the first glimpse of the ships and cargo coming into port, and obtain a crucial hour or so jump on the trading markets.”

Most Investment Firms Still Await AI Benefits

Only 2% of private equity firms expect to realize significant AI-driven value in 2025, while 93% expect benefits in 3-5 years. (World Economic Forum)

Generali, MIT Partner on Insurance AI Study

Generali and MIT will collaborate on AI applications for insurance, as part of Generali's €1.1 billion digital transformation program that includes a €250 million venture fund for insurtech investments. (Life Insurance International)

DATA

Data Centers Need 165% More Power by 2030: GS

Data center power demand is projected to surge 165% by 2030 from 2023 levels, driven largely by AI computing needs, according to Goldman Sachs Research. The shift will require about $720 billion in grid infrastructure investment through 2030, with global data center capacity expected to reach 122 gigawatts, up from the current 59 gigawatts. (Goldman Sachs)

JPMorgan Plans Decade Ahead for AI Power

Here’s a story about JPMorgan Chase's CIO of Infrastructure Platforms, Darrin Alves, on how the bank manages its technology infrastructure. (CIO Dive)