Coinbase Cuts AI Spend by 50% | Kalshi's $40B Valuation & Impending IPO | The Year for SaaS Roll-Ups

Coinbase's AI Spend Reduction and Implications

• Coinbase has cut its AI spend by 50% this quarter, while its usage has increased, particularly with regards to open-source utilization, which is taking away from its frontier model usage, raising questions about whether this is the new normal for the company 10s.
• The topic of AI spend and usage has gained significant traction, but some are skeptical of CEOs from non-AI companies sharing their thoughts on AI performance, viewing it as "performative social media" and instead wanting to see tangible results and numbers 2m6s.
• The discussion around Coinbase's AI spend reduction is seen as relevant because the company is no longer a "bright shining frontier AI company" but rather a common tech CEO sharing a relatable experience of reducing spend by 50% in two months, making it a cost management lesson for other companies 4m37s.
• The reduction in AI spend is attributed to the company's ability to get to grips with its spend and innovate, allowing it to continue generating tokens while cutting costs, with some viewing this as a positive example of cost management 6m14s.
• Despite the positive take on Coinbase's AI spend reduction, the company's recent performance has been lackluster, with a 30% decline in its last quarter, highlighting the challenges it still faces 8m20s.
• The conversation around AI spend and company performance is not just about the numbers, but also about leadership, with some expressing frustration with CEOs who talk about AI without delivering results, and instead wanting to see leaders who can drive meaningful change 10m0s.
• The reduction of AI spend by 50% is seen as a significant move, but its impact on revenue growth is questioned, and it is argued that optimizing AI spend may not help if revenue is shrinking, particularly in the crypto industry 10s.
• The fact that Coinbase cut its AI spend by 50% is viewed as a valuable data point, and it is suggested that every single CFO in the Fortune 500 may have taken notice and asked their CIO to review their own AI spend 2m6s.
• The reduction in AI spend may have implications for companies like Anthropic, which has seen significant growth in its run rate, and it is wondered whether this could lead to a decrease in revenue growth rate, potentially impacting the company's valuation 4m30s.
• The rise of open-source models is seen as a potential threat to the growth of frontier models, and it is acknowledged that the impact of this trend is still uncertain and may become clearer over time 6m40s.
• The example of Coinbase is used to illustrate the point that companies are re-evaluating their AI spend, particularly as they enter the second half of the year, and are realizing that they may have ramped up their spend too quickly 8m20s.
• The increased spending on AI and tokens has not resulted in the expected productivity and revenue lift for companies, including Coinbase, which is now cutting its AI spend by 50% 10s.
• Many portfolio companies that are doing well and are not purely reselling tokens are coming to the conclusion that they are not seeing the lift from net revenue and net productivity that they thought they would get from agentic coding, and this is a conflict that companies will have to deal with 2m6s.
• For software companies, increased AI spend should credibly give revenue lift because they are making more software, and even if a company is not a digital goods company, they should at least be seeing savings from their AI spend 4m42s.
• Companies that are not seeing either revenue lift or savings from their AI spend will be looking at their AI spend with a critical eye, and even high-performing companies will be expected to tie their AI spend to ROI 6m15s.
• The issue of not being able to show enough lift from AI spend will stress even the best companies, and it's time for the next mature phase of token spending and software development, where companies will need to be more thoughtful and intentional about their AI spend 8m30s.

Open Source Models and Their Impact on Frontier AI Companies

• The discussion revolves around the viability of businesses that rely heavily on artificial intelligence, with concerns that open source models could cannibalize their revenue pathways, potentially leading to bankruptcy if they are unable to reach extremely high revenue targets, such as a trillion dollars 10s.
• There is a recognition that even if the bulk of tokens are generated using open source models, the bulk of the revenue will likely still come from state-of-the-art frontier models, indicating a significant business opportunity, but also a need for companies to reassess their cost structures and ambitions 1m42s.
• The conversation touches on the example of Coinbase, which has cut its AI spend by 50%, and the question of whether AI can provide a revenue lift for the company, with a desire to see tangible results and a "magical boost" from AI, rather than just performative statements 4m6s.
• The importance of delivering real results from AI investments is emphasized, with a critique of companies that claim to be part of the "AI age" but fail to deliver, and a call for CEOs to demonstrate the actual impact of AI on their businesses 6m15s.
• The topic of Anthropic's perspective on distillation and the potential for open source companies to steal intellectual property is mentioned, but left for further discussion 2m6s.
• Software companies in the age of AI are either accelerating or becoming irrelevant, and if a company is not getting on board with AI, it will be left behind, with the exception of companies like Coinbase, which is considered more of a financial company than a software company 10s.

AI Spend and Business Outcomes in Software Companies

• The implementation of AI has led to a boost in growth for companies like Box, with Aaron Levy's efforts being recognized as effective in tying AI to the company's revenue and business model, although some of his approaches may be seen as too focused on AI 2m6s.
• There is skepticism about the goal of implementing AI if it is not tied to a company's business, and Rory Jason is cited as being consistent and ahead of the times in his views on AI, including his preference for Sam and Open AAI over Anthropic 4m30s.
• Dario, a figure associated with Anthropic, has criticized Chinese models for allegedly stealing their work through distillation of their models, which is seen as hypocritical given that Anthropic's own models were trained on other people's IP 6m40s.

Intellectual Property and Legal Concerns in AI Model Distillation

• The allegation is that Chinese open-source companies are breaching the terms of service of Anthropic by sending millions of prompts, recording the answers, and using that data to train their own models, effectively competing against Anthropic 8m50s.
• The main issue at hand is whether the use of Chinese models by US companies is illegal, with the interesting aspect being that it is clearly in breach of Entropic's terms of service, but this is a contractual problem between Entropic and the Chinese model companies 10s.
• The situation could lead to copyright issues and trade secrets acts, taking it above the level of contractual and into actual legal issues that the government might take an interest in, with Entropic potentially arguing that these are strategic US assets that need to be protected 2m6s.
• There is a possibility that the US government could put the full weight of its power behind Entropic, making what was a contractual dispute between two parties into a government-backed issue, with the potential for legislation that bans the use of Chinese models by US companies if they have been proven to distill using US foundation model technology 4m30s.
• The goal of Entropic and its supporters may be to ban the use of Chinese models by US companies, citing national security concerns and the theft of US intellectual property and data, with the potential for tariffs or other penalties on US startups that use Chinese models 6m15s.
• The idea of banning Chinese models is not implausible, with some arguing that it is almost inevitable given the support for such a move from influential figures such as Sam and Dario, as well as people around the administration 8m40s.

Regulatory and Policy Considerations in AI Model Usage

• The issue is an example of regulatory capture, with two separate issues being conflated: whether Chinese models should be banned due to distilling US prompts and getting a leg up, and the broader implications of such a ban on the tech industry 10m20s.
• The discussion revolves around the potential banning of Open AI and Anthropic due to distillation and national security concerns, with the argument that these companies should not be banned for being "naughty" but rather pay a fine for any intellectual property theft, similar to what Entropic had to pay, 10s.
• The concern about national security is also mentioned, with the example of Huawei being banned in the US, and the possibility that Frontier models could be seen as equivalent, leading to a potential ban, 1m42s.
• The argument is made that open-source models running in the US on US inference pose no risk, with the code being open to inspection and no backdoors, but this argument requires a government entity willing to listen and understand the nuances, 2m6s.
• A middle ground is suggested, where instead of banning Open AI and Anthropic, companies could make Fortune 500 companies uncomfortable with using open-source models, citing security concerns, which could lead to a de facto ban, 4m10s.
• The possibility of banning non-US, Chinese-based companies that have been found "guilty" of distillation is considered plausible, with US open-source companies potentially benefiting from this, 6m15s.
• Entropic's decision to reach out to the Senate Banking Committee, specifically Liz Warren, is seen as an attempt to get something done, with the implication that this is a serious issue, 8m20s.
• The question of whether these companies should be banned is posed, with the response being that they should not be banned, but rather pay a fine for any wrongdoing, and that national security concerns should be addressed separately, 10m30s.

Economic and National Security Implications of AI

• The idea of banning certain AI models is considered unnecessary if the code is open for inspection, the weights are available, and there's no telemetry back to China, as this would pose no danger 10s.
• The US economy is heavily reliant on the AI boom, with 40% of the S&P 500 tied to it, and any disruption to this could have significant impacts on the stock market, 401ks, and the overall economy 2m6s.
• The comparison is made to the oil situation in the Persian Gulf, where the US is so addicted to oil that it would do everything to protect it, and similarly, the US may take measures to protect the AI industry, even if it means making decisions that might not be in the best interest of the overall economy 2m6s.
• The suggestion that the US government should back all data centers, as previously mentioned by Sarah Frier, is discussed, and it's noted that while it may have been a controversial statement, it may also have some truth to it 4m42s.
• The potential consequences of propping up the AI industry, including trade restrictions and higher prices for AI inputs, are considered, and it's argued that this could have negative effects on the rest of the economy, with the losers being those who won't have access to cheap intelligence 6m15s.
• The example of IBM and MS DOS is used to illustrate the potential consequences of protecting certain companies or industries, and how this could stifle innovation and limit the growth of other companies 8m30s.
• The counter-argument is presented, acknowledging that while protecting the AI industry might be considered "dumb" in hindsight, it's also possible that governments may make decisions that prioritize national security over economic efficiency 10m50s.

Market Competition and Pricing in the AI Industry

• The history of Large Language Models (LLMs) has seen significant changes, with the industry likely leaving the oligopolical age and entering a new era where companies compete aggressively on price, rather than features, which can lead to massive price erosion 10s.
• In an oligopolical market, companies like Anthropic and OpenAI have dominated, with similar pricing and a focus on competing on features, but as the market evolves, price competition becomes more prominent, which can be beneficial for innovation in the short term but potentially detrimental in the long term 2m6s.
• The importance of competition is highlighted by the example of Coinbase, which has cut its AI spend by 50% due to the presence of open-source providers, demonstrating that competition can drive down prices and improve services 4m42s.
• Microsoft is facing significant challenges, with its stock down 16.5% in a single month, the worst performance since 2000, and its strategic position in the AI market is being questioned, particularly with regards to its reliance on OpenAI and lack of a standalone model 8m30s.
• The conversation around Microsoft's AI strategy and its investment in OpenAI has sparked debate about the company's direction and the potential risks and benefits of its approach, with some arguing that competition is essential for driving innovation and improvement in the industry 10m15s.

Microsoft's AI Strategy and Market Position

• Microsoft's core software business lacks a compelling AI product, and the company is being impacted by Co-work and Claude Code, which are becoming major players in the industry, serving individual knowledge workers and developers, respectively 10s.
• The market is looking at Microsoft's situation and realizing that the company doesn't have a strong AI growth story, despite its equity ownership in Open AI, and is instead relying on selling inference to Open AI for growth 2m6s.
• Microsoft's Azure growth is decelerating, which is troubling, as the company needs to accelerate growth to meet market expectations and maintain its stock price, with a guide of 37% growth being seen as a failure compared to the previous 40% growth 4m42s.
• The deceleration of Azure's growth is a significant issue, as it is a key factor in Microsoft's business and should be accelerating, especially in an era where companies are expected to be running multiple agents 24/7, and any slowdown in Azure's growth should be reflected upon 6m15s.
• Microsoft's 30% ownership interest in Open AI is notable, but the company itself lacks a state-of-the-art frontier model, which is a significant difference from other companies like Google, which has its own standalone model and product to sell 8m10s.
• The potential IPO of Anthropic in the coming months is expected to be highly volatile, with even small news items having a significant impact on the stock's price 10m30s.

Kalshi's Growth and the Future of Prediction Markets

• Kalshi is reportedly raising a new round at a $40 billion valuation, having previously raised a round in May at $22 billion, with the company recently announcing $2 billion in revenue, and this significant increase in valuation may be a sign of the "casinoization of society" and a risk-on mentality from consumers 10s.
• The growth of Kalshi can be attributed to the trend of sports betting, which has become a huge industry, with the company generating around 70% of its revenue from sports betting, and its success in this area is likely to continue as long as the industry expands and the company can take a disproportionate market share 2m6s.
• For Kalshi to become a $100 billion company in the next 12 months, either sports betting would need to continue to expand and the company would need to take a large market share, or the company's non-sports betting business, such as its crypto perpetuals, would need to become much bigger than expected 4m30s.
• The non-sports betting side of the business, including financial products and prediction markets, may have potential for growth, but it is unlikely that predicting election outcomes or other non-sports events will become a huge business, as the number of people interested in betting on these events is relatively low 6m40s.
• Other companies, such as Poly Market, are also operating in the online betting and prediction market space, and have received significant investment, including a 20% ownership stake from the Intercontinental Exchange, which suggests that there is potential for growth in this area 8m20s.
• New products, such as FOMO's PES, which allows consumers to bet on stock prices, are also being developed, and these products may contribute to the growth of the online betting and prediction market industry 10m30s.

IPO Market Trends and Bending Spoons' Strategy

• The market's direction and TAM expansion are being discussed, with Perks being a brilliant way to expand a TAM to a mega market, and the conversation is taking place on June 30th 10s.
• SpaceX's IPO has not frozen the AI IPO market, as Bending Spoons, a company that owns AOL, Evernote, and other 20-year-old software companies, is going public at a $20 billion valuation on July 1st, showing that the IPO market is still active 2m6s.
• The volatility around SpaceX's IPO has made OpenAI and Antropic potentially nervous about going public, but as long as the IPO market remains up, they will likely proceed with their plans, despite the risk and high volatility 4m30s.
• Bending Spoons' IPO multiple is around 8-10 times forward revenue, which is considered a healthy multiple, especially for a company that is the antithesis of an AI company, and this multiple is higher than that of many single-product B2B SaaS companies 6m40s.
• The valuation of Bending Spoons is seen as a clever way to make money, as the company has acquired and repackaged various tired consumer products, raised prices, cut costs, and is now being valued at a high multiple, with some wondering if the valuation is a little frothy 8m50s.
• Despite potential concerns about the valuation, it is believed that Bending Spoons will do well in the medium term, as the company has identified over a thousand potential targets for acquisition and can maintain outlier growth rates for a longer period 11m20s.
• The question of whether a company can maintain outlier growth for five plus years to justify a premium is debated, with the conclusion that if they can execute at their current level, it's justified for five years due to the numerous targets available, such as struggling software companies that can be improved with smart management and revenue arbitrage 10s.

Acquisition Strategies in the B2B Software Space

• There's a chance for several companies to take struggling software companies, improve them, and do revenue arbitrage by packaging them together into something high growth, with the example of Marquetto being a company that threatens its customers and has a poor product but still has a sticky customer base 2m6s.
• If one were to choose three targets for acquisition as the CEO of a B2B company, they would look for companies with nine figures in revenue, a sticky customer base, and potential for improvement, such as Marquetto, which has 300 million of revenue left but is decaying due to poor management and lack of features 4m30s.
• The idea of buying up B2B companies with sticky customer bases and improving them with better management and features is discussed, with the example of taking a company like Marquetto and giving it better leadership to retain its customer base and grow revenue 6m15s.
• The problem of private equity firms putting mediocre executives in charge of acquired companies, which can prevent them from reaching their full potential, is mentioned, and the possibility of buying these assets at a reasonable price is discussed 8m40s.
• The marketing automation space, including companies like Marquetto, and other markets like SEO optimization, with examples like Seamrush, are mentioned as areas where there is potential for improvement and growth 11m30s.
• The idea of buying and bundling tools like Seamrush, which was acquired by Adobe for under two times its revenue, to offer AI optimization to customers is proposed as a potential business strategy 10s.
• Companies like Pedro Duty, which has a commanding market share but lacks AI-enabled incident resolution, are seen as potential targets for acquisition and integration with AI-enabled products to drive reacceleration 42s.
• Many companies, including those with customer success teams that have become overly aggressive, have given up and have broken cultures, making it easier to turn them around if they have a sticky customer base 2m6s.
• The "bending spoons" model, which involves buying and revitalizing companies, is seen as a viable strategy, with potential targets including companies that have been acquired by private equity firms but are not being run effectively 2m6s.
• Constellation, a company that has been successful in the past, may need to reboot its model to continue to be successful, and its approach to acquiring and managing companies may not be effective in today's market 4m30s.

Reengineering and Growth in Acquired Companies

• The approach of private equity firms, which often bring in new managers who are not experienced CEOs, may not be effective in driving growth and change in acquired companies, and a more radical approach may be needed to drive reacceleration 6m10s.
• The "bending spoons" approach, which involves taking bold action to drive change and growth in acquired companies, is seen as a more effective strategy than simply optimizing and pressing the buttons, especially in companies that are pre-AI and need significant transformation 8m20s.
• The concept of generating new revenue from AI and re-engineering a company is a bigger task than just optimizing existing processes, and unless this is achieved, cost-cutting measures may not be successful 10s.
• Jason's B2B company may face a harder managerial task than Bending Spoons, as it requires more innovation and a significant re-engineering of the company, rather than just raising prices and optimizing 42s.
• To be successful in this space, a company needs a substantial checkbook, and buying a company like PageDuty at a premium could be a significant investment, with the board having a fiduciary duty to consider such an offer 2m6s.
• Any public company in decline that receives an offer with a premium of 15% or more has a fiduciary duty to take it seriously, and management may be aligned to accept such a deal to bail out of a struggling company 4m10s.
• The idea of rolling up smaller companies, like Bending Spoons did, could be a viable strategy, starting with smaller deals and building up to larger ones, and there are many companies that could be potential targets for this approach 6m15s.
• Companies like Patriot and Ahrefs could be potential targets for acquisition, with Patriot having a strong customer base and Ahrefs having a large customer count, despite facing challenges and having cheaper competitors 8m30s.

Chimath Palihapitiya's AI Startup and Market Concerns

• The goal of finding a home for venture-backed companies that are not large enough for a standalone IPO, but still have value, is a key aspect of this strategy, and there are many companies that fit this description and could be potential targets 10m40s.
• Chimath Palihapitiya has raised $135 million for his AI startup, 8090, which is a software factory platform that enables teams to collaborate with AI to handle the full software development cycle, and he is now the CEO of the company 2m6s.
• The platform allows for collaboration on new builds, code refactoring, and governance, and it is considered a great and super interesting market with a lot of competition, but also a lot of potential for growth and innovation 2m6s.
• There are concerns that Chimath Palihapitiya may not be fully committed to the startup, as he is a wealthy individual with multiple interests and may not be working on the project 100% of the time, which could potentially lead to the company running out of energy 4m30s.
• The fact that Chimath Palihapitiya is a CEO who is also a wealthy investor and may have other commitments and interests raises questions about his ability to dedicate the necessary time and effort to the startup, and some people may be hesitant to invest in the company due to these concerns 6m15s.
• Despite these concerns, Chimath Palihapitiya is given credit for taking the risk and trying to make the startup a success, and it is acknowledged that he is a talented and experienced individual who has had significant success in his career as an investor 2m6s.

Startup Funding Challenges and Growth Expectations

• The current state of the startup market makes it too easy to start a company, but the opportunity cost of cash is real, and growth rates of 1.5 million to 5 million may not be enough to raise a good series, which is a correct reflection of the current venture market 10s.
• A founder was turned down due to their growth rates not being sufficient, and the decision was based on the opportunity cost of cash, with the founder finishing the year at 1.5 million and projected to finish next year at 5 million 42s.
• The statement about the growth rates was met with agreement, but it was suggested that the message could have been phrased more carefully to avoid sounding obnoxious, and that a little bit of compassion from capital providers can go a long way 2m6s.
• The discussion also touched on the topic of engagement and rage on social media, with the point being made that rage is still a form of engagement, and that taking down a tweet that sparked controversy may have been a mistake 4m30s.
• A story was shared about two portfolio companies that were growing at great rates but were not quite at the level of top performers, and how VCs were not being honest with the founders about their chances of raising a successful round, with a tool being mentioned that can provide founders with a more realistic assessment of their odds 6m15s.
• The tool, an AI pitch deck generator, uses benchmarks from iconic and benchmark to provide founders with an honest assessment of their chances, and it was suggested that VCs should be more honest with founders about their prospects 8m20s.
• Founders are often not honest with themselves about their company's growth rate and its implications for raising funds, and they may be advised to converge on profitability or raise a lower amount if their growth rate is not compelling, such as 50%, 10s.

Founder-VC Dynamics and Realistic Growth Expectations

• Harry's tweet about turning down a founder with a growth rate of 1.5 to 5 not being good enough to raise a series A is discussed, and it is suggested that founders with such growth rates should be honest about their prospects and potentially change how they project future revenues, 2m6s.
• The importance of finding investors who believe in a company's potential, even if its current growth rate is not exceptional, is emphasized, and founders are advised not to run a lengthy fundraising process, but instead give investors a short amount of time to review their company and make a decision, 4m42s.
• The challenges of getting investors' attention in a crowded market are highlighted, with examples of companies like Higsfield, which has achieved significant revenue growth in a short period, and the need for founders to find investors who are willing to take a bet on their company's long-term potential, 6m15s.
• The idea that a growth rate of 1.5 to 5 may not be sufficient to raise funds in a competitive market, where companies with higher growth rates, such as 1.5 to 15, are more likely to attract investors, is discussed, and founders are advised to understand the facts and the consequences of their growth rate, 8m30s.
• Founders can still build amazing generational companies even with initial growth rates of 1.5 to 5, as there is only a modest correlation between initial growth rate and overall outcomes, and companies like Procore have grown slowly at first but become huge outcomes later on 10s.

AI Agents and Their Disruptive Potential in Software

• Most investors will say that the opportunity cost of cash is too high and will look for high-growth investments, which can make it difficult for companies with slower growth rates to raise funding, and even if investors want to invest, they may be limited by their job requirements 2m6s.
• The discussion touches on the topic of Claude Tag, a product that allows Claude to be a fully present member of a Slack channel, and Jason is asked to share his thoughts on it, with the context being that Claude Tag is an autonomous agent that can be focused on specific areas such as legal 10m30s.
• Claude Tag has the potential to be a significant product for traditional software, as it can run across multiple platforms including Salesforce, HubSpot, and others, and its importance to Anthropic could make it a major deal 14m20s.
• The conversation also mentions that some companies may not be suitable for venture capital and may need to adjust their plans accordingly, and that it's okay if a company is not a venture asset, as long as they can still achieve their goals 5m40s.
• The concept of an autonomous agent that can run 24/7, take in data, build analytics and dashboards, and operate without human intervention has the potential to disrupt the software industry, making companies like Salesforce and HubSpot seem like "dumb databases" if they do not adapt, 10s.
• The idea of such an agent raises questions about its potential impact on existing companies and products, such as Slackbot, and whether it could become a significant player in the industry, potentially even surpassing existing solutions, 42s.
• Salesforce's decision to support this concept, despite having launched their own version, may be due to the fact that they own Slack and need to allow access to the agent in order to remain a cross-platform communications platform, 2m6s.
• The ability of an AI to capture context, or the way people actually work and interact with software, is a key aspect of its potential to automate tasks and disrupt the industry, and Slack provides a unique entry point for this, 4m30s.
• The potential for Enthropic, a company mentioned in the context of this concept, to have a significant impact on the industry, possibly even surpassing the revenue of all public software companies combined by the end of the year, is a topic of speculation and wonder, 8m40s.

Materiality and Strategic Priorities in AI Companies

• The fact that some companies, like Coinbase, are cutting their AI spend by 50% raises questions about the materiality of certain developments in the AI space and whether they are worth the time and investment of major players, 10m20s.
• The definition of materiality is typically considered to be 10% of revenue, and if a company cannot generate at least $10 billion of revenue, it may not be considered significant, with this concept being relevant to companies like Salesforce and Anthropic 10s.
• Anthropic may not be interested in integrating with other apps or generating revenue if it's less than $10 billion by the end of the year, and the company's priorities may not align with those of its leaders or other stakeholders 1m2s.
• There was a notable incident involving a person leaving the Figma board, which was related to Anthropic, and it seemed that Anthropic was not aware of the significance of this event to others, leading to a sense of surprise and apology 2m6s.

Closing Remarks and Lighthearted Reflections

• The conversation touches on the idea that large companies or "elephants" can have a significant impact on smaller entities or "little people" when they make moves or decisions, with a quote about elephants and little people being trampled 4m30s.
• The discussion concludes with a lighthearted tone, referencing a British beach and a sports event, with the participants expressing their appreciation for the conversation and looking forward to future interactions 6m15s.