🤖 MLOPS: THE BACKBONE OF AI MODELS

AI is taking over the world — and Machine Learning Operations (MLOps) are the picks and shovels driving this gold rush. From the chatbots you interact with to the recommendations in your browser, MLOps brings together the practices and tools that manage the lifecycle of machine learning models, making sure they’re developed, deployed, and maintained smoothly.

💡 Understanding the AI Tech Stack

Before diving into MLOps, let’s zoom out and look at the AI tech stack. The AI application you use daily relies on multiple layers of technologies. Every layer, from hardware infrastructure all the way to applications, is essential to building and deploying AI systems.

1. Infrastructure Layer
 At the base, we have the Infrastructure Layer — the hardware backbone of AI. Companies like Nvidia and AMD provide graphic processing units (GPUs), the essential computing power needed to train AI models. As you may have noticed in the stock market, a select few players have experienced a boom driven by the demand for computing power — fuelled by the rapid growth of massive AI models.
 
 2. Cloud Platforms
 Next up are Cloud Platforms, such as AWS, Google Cloud, and Microsoft Azure. These platforms make the underlying infrastructure (using those powerful GPUs) available to businesses on demand through their data centers. This allows companies to train and deploy AI models without the hefty investment required to build their own infrastructure. According to McKinsey, demand for data centers in Europe is expected to grow to approximately 35 gigawatts (GW) by 2030, up from 10 GW today. To meet this new IT load demand, more than $250 to $300 billion of investment will be needed in data center infrastructure, excluding power generation capacity.
 
 3. MLOps Layer
 Now, let’s turn to MLOps, the focus of our deep dive efforts. MLOps, short for Machine Learning Operations, is critical to turning raw data into useful AI applications. While views on its scope can vary, based on our taxonomy, MLOps encompasses the entire lifecycle: from building and deploying AI models to monitoring and maintaining them. This ensures they perform efficiently, stay updated, and remain secure in real-world environments. We’ll explore MLOps in more detail below.
 
 4. Application Layer
 Finally, the Application Layer is where AI meets the user. These are the tools we interact with daily, from virtual assistants to automated HR platforms. For example:

• SuperHuman helps people reclaim hours by filtering important emails and automating replies.
• Paradox’s AI chatbot streamlines the hiring process, automating everything from candidate screening to scheduling interviews.
• JP Morgan’s COIN AI reviews legal contracts, a task that used to take human lawyers thousands of hours.

5. Foundation Models
 You’ve probably come across foundation models like OpenAI’s GPT or Anthropic’s models. These are AI systems trained on large amounts of data, offering companies ready-made solutions that eliminate the need to build and maintain their own AI models. This changes the role of traditional MLOps. As a developer, you can use foundation models as the base for your applications, with MLOps tools on top, helping you integrate, protect, and manage them responsibly. In that case, foundation models simplify the development process, while a small sub-segment of MLOps companies can be beneficial to ensure they operate efficiently and ethically.

🔎 The MLOps Deep Dive: How We See the Ecosystem

Now that we have a clearer understanding of the AI stack and how MLOps fits within it, we can dive deeper into our focus: MLOps. MLOps is the process and set of tools that ensure AI models are easy to build, run smoothly, and stay up to date in everyday applications. It starts with tools that manage inputs, like raw data, at the bottom and ends with tools that handle outputs, such as predictions, at the top.

Here are the key takeaways from the mapping for each category:

• Data Management: The quality of a model depends on the quality of its underlying data. The Data Management layer handles everything, from collecting raw data to making sure it’s clean, labelled, and organized. It also securely stores data, enables quick searches and even generates synthetic data (i.e., artificially created data that mimics real data). Companies such as Databricks facilitate data cleaning, transformation, and analysis, making sure the data is ready before training a model on it.
• Development & Training: With the data ready to be used, the next step is Model Development & Training. This includes tools and platforms that help with creating AI models or customizing off-the-shelf models for specific tasks. Hugging Face is a great example — they offer developers a library of ready-to-use, pre-trained models that simplify and accelerate the process of building new AI tools and apps.
• AI Model Deployment: Next is AI Model Deployment — where models are put into real-world use, and their performance is optimized. For example, companies like Octo AI facilitate the deployment of AI models and ensure that they run smoothly across various platforms, whether in applications, websites, software, or the cloud.
• Maintaining: Lastly, AI Maintenance is crucial for keeping models running correctly and effectively. On the data side, this means ensuring compliance, quality, and security — such as verifying that data complies with data protection laws before it is used. It involves actively monitoring model performance and tackles critical issues like privacy, security, and ethics. For example, OneTrust helps organizations navigate compliance and privacy regulations, ensuring AI systems not only perform well but also meet legal standards.
• Integrated Platform: In parallel, Integrated Platforms provide all-in-one solutions that simplify the MLOps process. Companies like DataRobot and Dataiku allow organizations to automate everything from data handling and model development to deployment and maintenance.

⛏ MLOps, the pick and shovels of the AI Gold Rush

The record $23.2B in AI funding in Q2 2024 highlights the soaring demand for artificial intelligence. Similarly, we’ve seen steady growth in VC activity focused on MLOps companies, reaching an all-time high in 2023 with 163 deals. This reflects the significant interest from investors in these fields.

Investors see MLOps as key to democratizing AI, enabling businesses across all sectors — healthcare, finance, logistics, and more — to harness AI’s potential. By simplifying complexity and reducing the costs of AI development, MLOps allows companies of all sizes to create specialized, high-impact models tailored to their unique needs.
 This is why the MLOps space is one to watch. It’s not just expanding; it’s building the very infrastructure that will power AI’s revolution across all industries.

Meme of the Month

*LLM — Large Language Model

In case you missed it…

General Technologies 🚀

🌌 Microsoft & OpenAI: Orion Model Coming Soon?
 Despite rising tensions between the two companies, Microsoft is gearing for the launch of OpenAI’s highly anticipated “Orion” model. If all goes well, we could see the latest model from OpenAI by the end of the year! Read more here.

🛰 Beyond Text: AI Can Now See & Hear!
 The chatbot era is evolving. With new voice and video capabilities, AI is redefining communication, moving past text-based chat like ChatGPT. Discover more here.

🔊 SpaceX’s Spectacular Rocket Catch.
 In October, Elon Musk’s team flew the world’s most powerful rocket back to its launchpad and captured it with two massive robotic arms. Check out the impressive video here — it’s a major leap toward sending humans to Mars.

🎧 What We’ve Been Listening To This Month
 — NY Times’ Hard Fork: Musk’s Role in the Election and AI’s Impact on Youth Mental Health here. 
 — TechCrunch Industry News: Catch up on the creator economy here — if you missed last month’s newsletter.

Sustainability 🌍

⛏ Massive U.S. Lithium Reserve Could Meet 9x Global Demand, New Study Finds
 A groundbreaking study reveals that the Smackover formation in Arkansas holds enough lithium to supply global demand nine times over, reshaping the future of U.S. energy independence and EV battery production. Read more here.

🔎 The ‘Missing Middle’: Climate Investing’s Biggest Challenge and What It Means for Scaling Solutions
 Despite record interest at NYC Climate Week, a critical funding gap known as the “missing middle” threatens the scale-up of climate solutions, creating unique challenges for investors and innovators striving to bring impactful technologies to market. Find out more here.

Blockchain & Crypto 💸

⚖️ Regulation

• SEC granted an “accelerated” approval to NYSE to launch options trading for multiple spot BTC ETFs
• Italy plans to raise capital gains tax on crypto to 42% from 26%
• Bitnomial sues the SEC over securities label on its XRP Futures
• Crypto.com sues the SEC after receiving a Wells Notice
• Mango DAO agreed to settle SEC charges involving the sale of unregistered securities
• U.K. set to introduce stablecoin regulations within months — Circle Executive

🏦 Financial Institutions

• Stripe acquires stablecoin platform Bridge for $1.1B
• Securitize partners with Zero Hash to enable purchasing of BlackRock’s BUIDL via USDC
• Visa launched their own Tokenised Asset Platform that is aimed to help banks issue fiat-backed tokens
• Stripe reinstates crypto payments for US businesses
• Fidelity plans to launch their fund tokenised fund
• Franklin Templeton launches their tokenised fund on Aptos
• Swift will begin live banks trials of digital asset transactions in 2025

🔥 Top Stories

• Buenos Aires launched QuarkID, a blockchain digital ID for public services, with zkSync developers Extrimian and Matter Labs.
• Kraken shared their plans to launch their own Layer 2, Ink
• Worldcoin World App 3.0, a “super app” with Mini Apps for user-friendly transactions, identity verification, and privacy-focused features.
• Uniswap Labs announced the launch of Unchain, a proprietary Layer 2 solution
• Tether.io has now crossed $1.2B in fees per 90 days. As a comparison: BlackRock, the world’s largest fund manager, had an operating income in Q3 this year just north of the $2B mark. BlackRock has around 19,800 employees, Tether has 150 approx.

🔎 Research
 
 📄 Grayscale published their views on potential implications of US election outcomes on digital assets markets
 📄 a16z shared their latest State Of Crypto 2024
 📄 Alexander Lange (Inflection) shared a state of VC, warning on effects of market concentration
 📄Arthur (DeFiance Capital) published a new investment thesis, arguing for the renaissance of DeFi protocols
 📄 Alex Thorn (Galaxy) published a “policy” scorecard on the positions of Biden, Harris and Trump on crypto
 📄 Patrick Mayr (Cherry Crypto) demonstrated the real-world benefits of DAOs with the example of HairDAO

Videos
 
 📹 Bankless invited a16z to talk about their latest crypto report In their latest weekly rollup,
 📹 Bankless explored whether crypto markets have already accounted for the potential impacts of a new U.S. president.
 📹 Bankless invited Omid Malekan (Columbia Business School) to talk about fundamental differences between ETH and SOL

Life Sciences 🔬

🎖 AI wins big at the Chemistry Nobels
 The Nobel Prize was awarded to three scientists (two from Google) who used AI to crack a decades-old mystery: understanding how proteins fold into three-dimensional shapes. Learn more here.

💉 Ozempic and similar drugs aren’t just tackling obesity
 They’re set to revolutionize global health. As costs drop, billions could benefit. Read more here.

🧫 Exosomes are making waves as a supposed cure-all
 From hair loss and skin aging to tackling long COVID and even Alzheimer’s. But here’s the catch: the science hasn’t fully caught up with the hype. Dive deeper here.

🤖 From Treatment to Prevention
 AI is shaping the future of proactive healthcare. Find out more here.

Not subscribed yet?

Subscribe Here