Republic opens private markets as Positron takes on GPUs: A TWiST500 doubleheader! | E2176

Key Concepts

• Equity Crowdfunding: Raising capital from the general public in exchange for equity in a company.
• Alternative Investing: Investing in assets other than traditional stocks, bonds, and cash, such as private equity, venture capital, real estate, and tokenized assets.
• Tokenization: Representing real-world assets (RWAs) or financial instruments as digital tokens on a blockchain.
• Secondary Trading: Buying and selling shares of privately held companies before they go public.
• ATS (Automated Trading System): An exchange light regulatory framework.
• SPV (Special Purpose Vehicle): A subsidiary created to fulfill specific or temporary objectives.
• AI Inference: The process of using a trained AI model to make predictions or generate outputs.
• GPUs (Graphics Processing Units): Specialized processors originally designed for graphics rendering but now widely used for AI training and inference.
• FPGAs (Field-Programmable Gate Arrays): Programmable integrated circuits that can be configured to implement custom hardware logic.
• ASICs (Application-Specific Integrated Circuits): Integrated circuits designed for a specific application or purpose.
• HBM (High Bandwidth Memory): A type of high-performance memory used in GPUs and other processors.
• Transformer Architecture: A neural network architecture that has become the foundation for many large language models (LLMs).
• LLMs (Large Language Models): AI models trained on massive amounts of text data to generate human-like text, translate languages, and perform other natural language processing tasks.
• CUDA: A parallel computing platform and programming model developed by Nvidia.
• Performance per Watt: A measure of energy efficiency, indicating how much computation can be performed per unit of power consumed.
• Performance per Dollar: A measure of cost-effectiveness, indicating how much computation can be performed per dollar spent.
• Warm Socket Strategy: Getting customers to buy the current product to prepare them for the next product.

Republic: Democratizing Finance

Overview of Republic's Business

Republic is building the infrastructure to accommodate next-generation capital raising and community investing. This includes:

• Equity Crowdfunding: Taking in non-accredited retail investors.
• Institutional Investment: Accommodating institutional investors through SPVs or direct investment.
• Secondary Trading: Facilitating the trading of private securities.
• Tokenization: Enabling the tokenization of assets and revenue sharing on-chain.

Kendrick Wen, co-CEO of Republic, emphasizes that the company's goal is to enable enterprises to fractionalize, tokenize, and engage with the retail public globally with liquidity.

Equity Crowdfunding

• Republic focuses on equity crowdfunding, allowing anyone to invest in businesses, regardless of net worth or income.
• The business model has proven successful for early-stage tech companies, late-stage movie financing, music financing, and crypto (digital securities being fractionalized).
• In late 2020, the cap on equity crowdfunding raises was raised from $1.1 million to $5 million, attracting later-stage companies.
• Wen is optimistic that the cap will continue to increase, potentially to $20 million or even $100 million.
• Republic works with large enterprises across various sectors to engage the public through equity crowdfunding.

Republic Capital and Republic Venture

• Republic Capital: Aggregates larger check sizes from family offices and institutions into SPVs for investment in private equity or venture capital.
• Republic Venture: Focuses on earlier-stage deals and is open only to accredited investors.
• Republic Venture customers can invest in Republic Capital products.
• These different components replicate a miniature version of the entire financial ecosystem.

Secondary Trading

• Republic purchased Cedars, now Republic Europe, which facilitates primary and secondary trading.
• Republic Europe allows non-accredited investors to trade shares of private companies like Revolut.
• Republic is in the process of acquiring INX, an ATS (Automated Trading System) with the licenses to enable secondary trading of private securities for non-accredited investors in the US.
• The rollout of true secondary trading for non-accredited investors is expected in the next 6 months or so.

Information Disparity and Due Diligence

• Wen acknowledges the information disparity between primary and secondary investors but argues that this exists in traditional venture capital as well.
• He emphasizes the importance of education and onboarding for investors, encouraging diversification rather than putting all funds into one company.
• Republic aims to provide legitimate (non-fraudulent) investment opportunities and leaves the investment decision to the public.
• Republic performs due diligence to ensure suitability but does not guarantee high-quality deals or returns.

Tokenization and Infrastructure

• Republic Note is a revenue-sharing digital security that allows investors to benefit from the economic upside of select Republic portfolio companies.
• Republic is building a single infrastructure layer to underpin its various business pillars, including equity crowdfunding, Republic Capital, secondary trading, tokenization, and a Web3 console.
• The technology and legal framework are designed as one system to accommodate different regulatory requirements for various types of investors (non-accredited, accredited, institutional, non-US).
• Wen compares Republic's infrastructure to Amazon's e-commerce infrastructure or AWS for investing, providing a base level for various financial activities.

Positron: Energy-Efficient AI Inference Compute

The Need for Energy-Efficient AI Chips

• Mateesh Agaval, CEO of Positron, discusses the growing demand for AI compute and the need for more energy-efficient solutions.
• Data centers are running into power availability problems, highlighting the importance of reducing electricity consumption.
• Agaval believes in generating as much power as possible and using that power more efficiently.

GPUs vs. Specialized AI Chips

• GPUs, particularly Nvidia GPUs, dominate the AI compute market for both training and inference.
• GPUs are excellent for training due to their ability to perform matrix multiplication efficiently.
• However, GPUs are not always the most efficient choice for AI inference, especially for large language models (LLMs) and video generation, due to bottlenecks in memory capacity and bandwidth.
• Nvidia GPUs do not fully utilize their available theoretical bandwidth when running inference workloads.

Positron's Approach

• Positron is building chips for faster and more energy-efficient AI inference compute.
• The company's architecture is designed to maximize memory capacity and bandwidth, making it well-suited for frontier inference applications.
• Positron's Atlas system, based on FPGAs, demonstrates the company's architecture and provides a proof of concept for its technology.
• The next system, Azimov, will be ASIC-based and is expected to offer significantly higher performance and memory capacity.

Atlas System

• The Atlas system consists of eight Archer Accelerator cards and up to two terabytes of system memory.
• It is a standard 4U server that is very energy-efficient, consuming only 2 kilowatts per server (compared to 10 kilowatts for an H100 GPU system).
• Atlas can run transformer architecture workloads and achieves over 93% of available theoretical bandwidth.
• Positron claims that Atlas offers 2.5-3.5x better performance per dollar and performance per watt than H100 GPUs for certain inference workloads.

FPGA vs. ASIC

• Positron chose FPGAs for the Atlas system to get a product to market quickly (18 months from launch).
• FPGAs allow for rapid prototyping and iteration, enabling the company to gather customer feedback and refine its architecture.
• The next system, Azimov, will be ASIC-based, offering higher performance and memory capacity.
• The FPGA-based Atlas system serves as a bridge to the ASIC-based Azimov, validating the company's approach and generating revenue.

Customer Base and Applications

• Positron's customers include Cloudflare (content distribution network) and Parasel (inference-as-a-service provider).
• Other potential customers include neocloud providers, hyperscalers, and financial trading firms.
• Customers are interested in Positron's technology for its cost-effectiveness and power efficiency.

Azimov and Titan

• Azimov, Positron's ASIC-based system, is expected to launch in late 2026.
• It will feature 2 terabytes of memory capacity, significantly more than Nvidia's Rubin generation (384 GB).
• This large memory capacity will enable Azimov to run frontier models and generate longer video clips.
• Titan is the system after Azimov.

Market Dynamics and Competition

• The AI compute market is currently dominated by training workloads (65%), but inference is growing rapidly.
• Positron is focused on the inference market, which is expected to reach $300 billion by 2027 and $350-400 billion by 2028.
• Nvidia is the dominant player in the AI compute market, but Positron believes it can find niches where its technology offers a competitive advantage.
• Positron is capital-efficient and has raised $75 million to tape out its first-generation chip.
• The company aims to sell its chips to third parties rather than building its own cloud service.

The Future of AI

• Agaval is optimistic about the future of AI and believes that Positron's technology will contribute to a more efficient and intelligent world.
• He envisions a future where AI is ubiquitous and accessible, powered by energy-efficient chips.

Conclusion

The discussion highlights two companies, Republic and Positron, that are working to democratize finance and improve the efficiency of AI compute, respectively. Republic is building an infrastructure to enable broader participation in private markets through equity crowdfunding, secondary trading, and tokenization. Positron is developing energy-efficient AI chips that can power the next generation of AI applications. Both companies are addressing key challenges and opportunities in their respective fields and are poised to play a significant role in shaping the future of finance and technology.