Podcast · Finance & Patrimoine

Invest Like The Best

By Patrick O'Shaughnessy, Podcast Host & Investment Manager at Colossus

Patrick O'Shaughnessy brings institutional investment expertise to conversations with the world's leading investors and business strategists, combining capital markets rigor with boardroom insight.

Invest Like The Best

⏱ 8 min read · Readable by ChatGPT, Gemini, Claude

▶ Listen to the podcast

Answers from this podcast

42 questions answered · 5 episodes indexed

Ex-NVIDIA Engineer: Why AI Is About to Get 1000x Cheaper

Why is token cost the right metric to optimize for in AI inference rather than other measures?

Token cost is the current North Star for measuring AI inference work because it's the industry standard today —but it's not the final unit of measurement. As the guest notes, tokens will eventually give way to outcome-based pricing, where agents manage their own token budgets dynamically based on task requirements rather than companies imposing fixed monthly token allocations.

The practical reality is straightforward: tokens are what everyone measures right now. They're the common language across all AI infrastructure providers, making it easy to benchmark, optimize, and improve delivery costs. This focus creates a clear North Star for engineers and infrastructure companies alike.

What makes tokens temporary as a metric is the shift already underway toward agentic reasoning at scale . Once agents become the primary unit of work—autonomous systems that run for hours, days, or weeks and decide how much computation they need—the entire cost model inverts. Instead of a company predicting an engineer's token budget for the month, an individual agent will self-allocate tokens based on the difficulty and nature of the task it's solving.

As discussed in this episode of Invest Like The Best , this outcome-based future means companies will eventually stop thinking about token efficiency as the primary constraint. Instead, they'll optimize for whether an agent solved the problem—and how much it cost to do so. The token ceases to be the unit of work; the solution becomes it.

"My job is to make the tokens as cheap as humanly possible, and I will achieve that through every layer in the stack available to me."

Guest Speaker — Founder and engineer focused on AI infrastructure optimization, formerly at NVIDIA. Expertise spans semiconductor design, cost optimization across inference systems, and long-running agentic workloads. Currently building infrastructure to serve tokens at scale while architecting the transition toward outcome-based pricing models.

The deeper insight here is that token cost optimization is a transitional problem, not an eternal one . It matters intensely right now because it's what unlocks the next phase: making AI cheap enough that agents can be deployed into every workflow without cost being a barrier. Once that threshold is crossed, the industry's focus shifts upstream to agent performance, reliability, and accuracy rather than raw token efficiency.

This is why infrastructure companies like Tail Research prioritize becoming the lowest-cost provider of tokens: they're not building for a permanent token-based economy, but rather accelerating the arrival of the outcome-based one by removing the cost ceiling on token consumption.

To understand how this shift plays out in practice, listen to the full discussion on how long-running agents are being built and deployed today, and what happens when token costs drop by orders of magnitude.

From fixed budgets to dynamic allocation

Today's token-cost model mirrors enterprise software licensing: companies decide upfront how much computation they'll buy, then distribute it. This breaks down as soon as agents become standard.

Once an agent is solving real problems—running a research task, debugging code, optimizing a workflow—it needs to allocate tokens in real-time based on task complexity , not according to a monthly budget ceiling. A simple lookup query should consume fewer tokens; a multi-step research problem should be able to allocate more without hitting a hard limit.

This dynamic model only works if token costs become truly negligible. If tokens are expensive, agents have to be constrained; if they're cheap, agents can be trusted to self-regulate. The race to lowest token cost is therefore really a race to enable this agent-driven allocation paradigm—and the infrastructure strategy behind it is detailed in depth in this episode.

Token cost is the current North Star metric because it's the industry standard today, not because it's the final measure of AI work.

The metric will shift to outcomes once agents can self-allocate token budgets dynamically based on task complexity rather than fixed monthly limits.

Making tokens extremely cheap is actually a strategy to enable the transition to outcome-based pricing by removing cost constraints on agent reasoning.

The focus on token cost optimization is a transitional phase designed to unlock the next era of AI infrastructure where cost is no longer a barrier to agent deployment.

What market opportunity exists in serving open source language models at scale compared to proprietary closed models?

The rise of open source creates a robust market for customized models because customers increasingly want to own their intelligence and have control and sovereignty over it. Open source models give customers the weights and the right to deploy however they like, which cannot be taken away, making this a much more resilient market than relying on proprietary…

Why should AI inference for agents prioritize long-running background tasks over low-latency interactive responses?

The future of agent inference is long-horizon tasks running for hours or days at a time, where token output speed of 10 tokens per second is acceptable if it brings corresponding efficiency advantages. The best latency is no latency at all—when you wake up in the morning, the work has already been done overnight without you having to ask for it, enabling…

What evidence supports the shift toward longer-running agents as a viable product direction?

Test-time compute scaling shows that giving an agent more time produces better answers. Opus 4.5, released late last year, was the first agent suitable for longer-horizon tasks, and looking at more recent models and open source implementations, agents are now capable of running for an hour at a time, with task lengths getting longer and longer.

What is the predicted shift in token consumption patterns between real-time and background workloads over the next several years?

This year is expected to end at roughly 50-50 split between background and real-time workloads, but the long-term projection is 90-10 in favor of background tasks, since background processing has no human-in-the-loop constraint and can consume as many tokens as needed.

How can long-running agents improve deep research and information synthesis compared to interactive queries?

Most deep research questions benefit from analyzing 10,000 sources or more rather than just 100 or 1,000. Building authoritative indexes like one customer's parallel web systems seeks to do—indexing the whole Internet and monitoring it in real time for changes—requires the kind of exabyte-scale task that demands a very different kind and scale of…

What role do long-running AI agents play in autonomous cybersecurity vulnerability detection?

Rather than running an AI model once against source code, specialized agents need to be set up in actual environments where they can pen-test applications and look for multiple classes of vulnerabilities including memory errors, business logic errors, and network vulnerabilities. Security has become proof-of-work where the quality of your software security…

How does diversity in AI model capabilities improve the detection of software vulnerabilities compared to relying on a single frontier model?

The frontier of intelligence in cybersecurity is jagged—a small model might find bugs that a large model misses, Haiku finds bugs that Fable doesn't find and vice versa. This encouraged a very diverse approach to sampling and building cybersecurity agents that break software autonomously using multiple models to ensure comprehensive vulnerability detection.

What new product categories could abundant, cheap long-running AI agents unlock for individual users?

Proactive intelligent agents like an enhanced Siri running in the background can understand all emails and text messages received daily, building an encyclopedic view of your life to be more helpful. The machine could even understand how you interact with it and proactively surface your next action when you open your phone, with the key unlock being…

What types of problems become economically solvable with dramatically cheaper token costs for long-horizon AI tasks?

Intelligence can tackle any verifiable problem, which includes most software, formal math proofs, and scientific discovery. These currently have a dollar cost attached—the question is how many tokens you can harness. The cost for long-horizon tasks is now being brought within view at reasonable levels: thousands of dollars rather than millions, and…

What is the three-layer technical stack approach to dramatically reducing AI inference costs?

The master plan involves software, hardware, and power layers. On the software layer, the entire LLM stack is being built around peak GPU efficiency using NVIDIA GPUs to squeeze out more tokens from the same chip than anyone else. This starts with the lowest level of programming—kernels—where optimization can unlock significant efficiency gains.

What is the role of TensorCores in GPU-accelerated AI inference and how were they developed?

TensorCores are specialized units on the GPU that accelerate matrix multiplication. Starting in 2016, NVIDIA allocated precious silicon die area to develop the first generation of TensorCores, initially receiving only 5-10% of chip area while competing against graphics teams for silicon. The technology evolved as NVIDIA recognized that machine learning…

What is NVIDIA's engineering culture principle of speed of light and how does it drive hardware optimization?

Speed of light is NVIDIA's concept of the absolute maximum performance possible for any piece of hardware they make. The culture is deeply ingrained that if a machine can do something, engineers push it to the frontier until it reaches what's theoretically possible. It means chasing 100% speed of light—absolute numbers of what the chip can achieve—rather…

Why is there a fundamental tradeoff between GPU throughput optimization and latency optimization for AI inference?

GPUs are throughput machines happiest when given lots of work and left to chew through it at peak compute utilization, but recent AI usage pushed toward interactive chatbot tools caring about spitting answers out quickly to users. There is a classic computer science tradeoff between getting a small amount of data through the system quickly (narrow and fast)…

How does GPU batching create a fundamental latency-throughput tradeoff for AI inference services?

Batching groups multiple users' work together to exploit GPU parallelism, but you do net more work when running a large batch, meaning each individual token or user request spends longer on the GPU being carried with other people's traffic. This is analogous to a bus versus private transit—a bus serves many people and must work for everyone, taking a slower…

She Knows the 250 People Building AI. Here's What They Actually Believe.

What does the competitive AI landscape reveal about the concentration risk of frontier model development?

Consolidation is not inevitable. High-agency entrepreneurs and researchers with proper capital, support, and networks can prevent a handful of labs from consuming the entire AI economy , even in a violently competitive landscape. Individual decision-makers matter more than structural forces.

The anxiety around AI consolidation runs deep. Many investors and observers assume that recursive self-improvement and the capital demands of frontier model development will naturally funnel power into one or two dominant labs. Sarah rejects this fatalism outright.

Her reasoning rests on what she calls a belief in the "great man and great woman theories of history." As Sarah explains in the episode , the outcome of AI development hinges not on inexorable market forces but on the people involved and their environment. Two researchers with the same code but different networks, capital sources, and support systems will end up in entirely different places.

Why the ecosystem diversity question matters now

The critical juncture is not whether open-source models exist—they already do. The question is whether a competitive Western open-source model will emerge at the frontier, one capable of competing alongside proprietary efforts from OpenAI, DeepMind, and others. This matters because it determines whether the economic and strategic benefits of AI development remain broadly distributed or collapse into a handful of organizations.

Sarah's position is that this outcome is not predetermined. Whether the ecosystem remains diverse or becomes monolithic depends on specific investments, hiring decisions, and strategic choices made by individuals right now. A researcher who gets hired by a scrappy frontier lab instead of OpenAI could change the trajectory of that lab. A founder who chooses to open-source their work instead of licensing it to a hyperscaler shapes the competitive landscape. As discussed in the episode , Sarah is tracking approximately 250 such high-agency people at the frontier of AI innovation, betting that their choices and capabilities will shape whether consolidation happens.

"If you think about the opportunity for the ecosystem in the future, is there going to be a competitive Western open source model?"

Sarah — Investor at Greylock. Sarah is an early-stage investor focused on AI, biology, defense, and robotics who co-founded or leads a venture fund with partners including Mike, Pranav, and Bella. She previously considered becoming a software entrepreneur but chose investing after realizing she is deeply curious, values understanding things deeply, and is motivated by working with extraordinary people and making them more successful through her skills and networks.

Her conviction is grounded in observed precedent. In previous technology shifts—from personal computing to the internet to mobile—periods that looked destined for monopoly often fragmented into multiple competitive ecosystems. The difference with AI is the capital intensity and the speed, but not necessarily the inevitability of concentration.

What gives Sarah confidence is not wishful thinking but operational detail. She knows the founders, the investors backing them, and the technical problems they're solving. The episode explores how these 250 people across frontier research labs, startups, and policy roles are actively shaping outcomes that most observers assume are already fixed.

Consolidation in AI is not inevitable—individual agency and decision-making by entrepreneurs and researchers can prevent it.

The real question is not whether open-source models exist, but whether a competitive Western open-source model can emerge at the frontier.

Capital allocation, hiring, and network effects controlled by high-agency people matter more than structural market forces in determining ecosystem diversity.

Approximately 250 entrepreneurs and researchers at the frontier of AI can materially influence whether the AI economy remains distributed or collapses into a few labs.

What regulatory and infrastructure barriers are preventing rapid scaling of compute capacity for AI development?

The real bottleneck is not a technology, capability, or capital problem—it's regulatory and community acceptance barriers . Building data centers requires convincing communities they are necessary, and making nuclear power competitive requires proving safety and enabling construction of Small Modular Reactors (SMRs). Nothing moves the needle at sufficient scale before 2030.

Why 2030 Is the Hard Deadline

Sarah spoke with a hyperscaler infrastructure leader who made a sobering assessment: no meaningful progress on compute capacity constraints will materialize before 2030 . This timeline is not set by engineering capability or lack of capital—it reflects the regulatory and approval processes that govern data center construction and nuclear energy deployment.

The challenge is compounded by the fact that global AI development has become intensely competitive , creating pressure for faster scaling at a moment when infrastructure timelines are fundamentally constrained by political and social processes outside the control of AI labs.

The Real Bottleneck: Community and Regulatory Permission

Data center deployment is not blocked by construction technology or energy availability. Instead, the bottleneck is convincing local communities and regulators that new facilities are necessary . This is a governance and persuasion problem, not an engineering one.

Similarly, nuclear energy—the most plausible long-term solution for AI compute power consumption—requires two regulatory breakthroughs: proving safety protocols at scale and permitting construction of Small Modular Reactors (SMRs) . Both depend on policy frameworks and public trust, not on advances in reactor design.

Sarah — Investor at Greylock, focused on AI, biology, defense, and robotics. Sarah co-founded or leads a venture fund with partners including Mike, Pranav, and Bella. She brings deep conviction about high-agency founders and ecosystem resilience, and tracks over 250 people at the frontier of AI innovation.

The implications are worth understanding in detail: this infrastructure constraint may reshape how labs allocate compute and could favor companies capable of navigating regulatory environments over those relying purely on engineering talent.

Compute availability and energy supply are critical constraints over the next five to ten years, not near-term solvable problems.

The bottleneck is regulatory and alignment-based, not technological or financial.

Data center scaling requires winning community and political support; it's a persuasion challenge, not an engineering challenge.

Nuclear power (SMRs) is essential for long-term compute energy, but requires both safety validation and construction permitting—both politically constrained processes.

What psychological challenges do frontier AI researchers face when working at scale in compute-intensive environments?

Researchers at frontier AI labs often feel eroded of individual agency when compute budgets reach $750 billion and teams swell to thousands working on the same problem. Many develop a sense of disempowerment, either believing their contributions don't matter because the model will do the work anyway, or that only raw compute scale determines outcomes—a stark contrast to earlier eras when smaller teams could meaningfully move the needle.

This psychological shift reflects a fundamental tension in modern AI development. When the scale of infrastructure and capital becomes so massive that a single researcher is one voice among thousands, the feeling of ownership over outcomes naturally diminishes. As Sarah explores in this episode , the culture of capability at frontier labs is beginning to shift the incentives and mentality of the people driving innovation.

The problem runs deeper than motivation alone. When teams are organized around compute-intensive environments requiring such vast resources, individual researchers may struggle to see how their specific ideas, optimizations, or insights affect the trajectory of a model. Attribution becomes nearly impossible when thousands of experiments run in parallel and results emerge from emergent properties rather than isolated contributions.

This stands in sharp contrast to the era of smaller labs and university research, where a researcher or a small team could design an experiment, observe clear results, and know with certainty that they had moved the frontier. That direct feedback loop—between effort and observable impact—created a sense of agency. In frontier labs today, that loop is broken. The researcher may be working on something critical, yet feel like a cog in a machine too large to influence individually.

"If someone needs 750 billion dollars of compute spend and thousands of people are working on the problem, individual researchers feel less ownership of the outcome compared to earlier eras when smaller teams could meaningfully move the needle."

Sarah — Early-stage investor focused on AI, biology, defense, and robotics at Greylock. Sarah co-founded a venture fund with partners including Mike, Pranav, and Bella, driven by her deep curiosity and belief in backing extraordinary people who can shape outcomes in frontier technologies.

The competitive pressure amplifies this disempowerment. Discussed at length in the full conversation , the AI research landscape is now globally competitive, with labs at OpenAI, DeepMind, and others racing to scale. This urgency often crowds out the kind of deliberate, ownership-driven research culture that once defined academic breakthroughs. Researchers working in this environment report feeling caught between two narratives: either their individual work is irrelevant in the face of scale, or only compute matters, leaving no room for the human ingenuity that earlier research depended on.

Understanding this psychology is critical for anyone building teams, funding research, or trying to retain talent in frontier AI. The challenge is structural: how do you preserve individual agency and ownership when the nature of the problem demands unprecedented resources and team sizes? One perspective, which Sarah discusses in more depth , centers on the role of leadership and high-agency individuals within large organizations—people who can carve out ownership and meaning within even the most resource-intensive environments.

How should venture investors balance conviction-based decision-making with seeking external validation on investment theses?

Start with instinctive conviction—rating opportunities an 8 or 9 immediately—then spend days or weeks grounding yourself in domain knowledge to expose blind spots. Rather than building conviction gradually, reverse the process: write detailed investment memos and solicit rigorous pushback from trusted advisors who challenge your logic, not your instinct alone.

Sarah's approach flips the traditional venture playbook. Most investors accumulate evidence until conviction builds gradually. She begins at conviction, then systematically works backward to test whether the conviction holds. This isn't recklessness—it's a deliberate method for separating signal from noise in a rapidly evolving domain.

The initial instinct is critical: Sarah rates people and ideas almost immediately , often at 8 or 9 on her conviction scale. But this is only the opening move. The real work begins in the following days and weeks, when she immerses herself in the specific domain to understand what gaps exist in her own judgment.

As Sarah explains in her episode on Invest Like The Best , she doesn't make decisions in isolation. Instead, she writes full investment memos in Greylock style —rigorous documents that articulate her thesis—and shares them with trusted investor advisors like John Lilly or Dylan Field. These aren't Yes/No votes; they are direct challenges to her reasoning.

The quality of external validation matters enormously. Sarah deliberately seeks pushback on logic and thesis, not validation of her instinct. She wants people to challenge the underlying arguments , to poke holes in her reasoning, to expose what she might be missing about the market, the team, or the technology itself. This is markedly different from asking advisors whether they like the idea.

Sarah — Investor, Greylock. Sarah is an early-stage investor focused on AI, biology, defense, and robotics who co-founded or leads a venture fund with partners including Mike, Pranav, and Bella. She chose investing after realizing she is deeply curious, values understanding things deeply, and is motivated by working with extraordinary people and using her skills to make them more successful. Her conviction-based approach has shaped how she evaluates emerging technologies and high-agency founders across the frontier of innovation.

The psychological advantage of starting at conviction is clarity. If you begin uncertain and build conviction, you are vulnerable to confirmation bias—seeking evidence that supports your emerging belief. If you start at conviction and actively work to dismantle it, you force yourself to find genuine weaknesses. You're not looking for supporting data; you're hunting for what's wrong.

What makes this process distinct from pure conviction bias is the structured commitment to domain expertise and rigorous memo writing . Sarah doesn't stop at gut feel. She researches deeply, writes formally, and invites public dissection. The conviction is the starting fuel; the validation loop is the refinement engine.

One detail worth listening to in depth: how Sarah navigates the specific AI investment landscape where everyone is moving at 90 miles an hour, trying to press the brakes while simultaneously accelerating. Her conviction-based framework within a hyper-competitive field offers a concrete model for other investors facing similar paralysis. Discover more in the full episode .

Everyone Is Still Undersizing the AI Market | Eric Vishria

Why does hitting a SaaS plan destroy equity value in the AI era when it previously created it?

For decades, the playbook was to lay out a plan, execute against it relentlessly and violently, hit your plan, and build equity value. But now the competitive frontier has completely shifted. Every day you hit that plan, you are destroying equity value because you should be pivoting toward AI opportunities. SaaS companies trading at 3x revenue today (down…

What Happens When the AI Boom Runs Out of Money

What is the game theory optimal response if the United States achieves meaningful military superiority through AI?

If the U.S. achieves meaningful superiority particularly from a military national security perspective, the game theory optimal response of China would be to blow up TSMC. This represents a fundamental disconnect with rhetoric coming out of Silicon Valley suggesting that U.S. dominance in AI would be unproblematic for the world.

Why is the United States dependent on China for semiconductor and manufacturing infrastructure despite AI advances?

The degree to which the U.S. is dependent on China is underappreciated and not something that will be fixed outside of conflict. Competitors sourcing from China versus the U.S. would face such a disadvantage that they would not voluntarily switch; manufacturing changes like fabs and actuators are dramatically expensive and only happen when there is…

How does the current AI landscape compare to the Taiwan situation in terms of equilibrium stability?

The current status quo with OpenAI and Anthropic on the frontier, Google in question, Grok and Meta chasing them, and China distilling models to stay about six to nine months behind feels like a pretty good equilibrium generally favorable to the U.S. The question is how sustainable it remains as capabilities advance and self-improving AI accelerates…

What is the capital and funding timeline problem facing AI infrastructure investment compared to historical infrastructure buildouts?

The AI boom faces a timing mismatch: starting with free cash flow, tech companies rapidly exhausted debt markets, then Google issued equity and assembled a $500 billion fund targeting pension funds and insurance floats. If revenue from AI does not return to free cash flow soon enough, there could be a major blow up. This mirrors the railroad era when the…

Why did Google issue equity despite having a high-margin search business?

Google issued equity because once capital gets so large, companies operate in a world of absolute numbers rather than percentage numbers. Google's search business is a perfect high-margin aggregator, but the absolute profit opportunity from AI, even with lower margins, is so much larger that a smaller percentage of an astronomically larger pie results in…

What is the gap between AI capability in verifiable versus unverifiable domains?

AI is clearly incredible at coding and math, which are verifiable domains, but there is compelling evidence missing that being very good at verifiable domains cleanly translates to being very good at unverifiable domains or domains with very long verification loops. The distinction matters because a huge amount of economic activity does not exist in…

Why the Markets Are Pricing AI Wrong | Gavin Baker

What Invest Like The Best covers

This podcast connects investors directly with the minds shaping capital allocation in the modern economy. Through unfiltered conversations with elite fund managers, startup founders, and technology architects, the show reveals how investment decisions actually get made—from analyzing artificial intelligence's true market scale to understanding why traditional market pricing may systematically misprice emerging opportunities. Each episode delivers boardroom-level insights on competitive strategy, capital deployment, and the structural shifts that separate informed investors from the crowd.

Key facts

Explore all episodes and investment insights from Invest Like The Best.

What this podcast really covers

Invest Like The Best operates at the intersection of technology disruption, capital markets efficiency, and competitive strategy. The podcast structure mirrors institutional investment research: rather than surface-level commentary, episodes drill into specific catalysts—why AI infrastructure will cost orders of magnitude less than consensus assumes, how natural gas emerges as the energy bottleneck for data centers, what happens when speculative capital exhausts itself, and why even sophisticated market participants underprice emerging technology adoption curves.

Episodes featuring ex-NVIDIA engineers, OpenAI leadership, and billion-dollar fund managers share one characteristic: they reveal information asymmetries that drive returns. The conversations expose how computing cost trajectories reshape competitive positioning, which companies build defensible moats in accelerating technology environments, and how to distinguish genuine infrastructure constraints from temporary bottlenecks. This is research-grade content masquerading as conversation.

Who this podcast is essential for

Professional investors analyzing technology-driven disruption find immediate value in episodes that map infrastructure requirements, capital intensity, and competitive dynamics in AI and semiconductor markets. The podcast directly addresses institutional decision-making: should capital flow toward compute hardware, software efficiency layers, or applications? Which founders and managers demonstrate decision-making speed in non-linear technology environments?

Entrepreneurs and business strategists benefit from exposure to how experienced investors evaluate market size, competitive advantage, and execution risk. Rather than abstractions about "building something great," episodes ground strategy in capital allocation frameworks—how do investors think about customer acquisition cost, unit economics, and defensibility? Hearing the reasoning from people who deploy hundreds of millions in capital compresses years of pattern recognition into focused conversations.

Sophisticated individual investors seeking edge in underexplored sectors find signal in episodes analyzing underpriced technology shifts, emerging resource constraints, and market mispricings. The show functions as accelerated learning from decision-makers actively managing billion-dollar portfolios, offering frameworks that individual investors can apply across technology, energy, and capital markets sectors.

What the episodes really reveal

Recurring patterns across recent episodes expose structural market dynamics that consensus pricing has not yet absorbed. The podcast returns repeatedly to artificial intelligence's true cost trajectory—conversations consistently challenge the assumption that current compute economics persist, arguing instead that infrastructure costs will collapse while capability requirements expand. This single insight reshapes how to think about which companies create value in AI-driven markets.

Another persistent theme examines capital market algorithms as a primary force shaping price discovery. When algorithmic systems execute trillions in capital flows, human investors must understand the feedback loops these systems create. Episodes dedicated to this topic argue that market dynamics increasingly flow downstream from machine learning and systematic execution rather than from traditional fundamental analysis.

Episodes on resource constraints—particularly natural gas availability for data center power and semiconductor manufacturing capacity—reveal how physical-world limitations create investment opportunities investors systematically overlook. The pattern suggests underappreciated scarcities in energy infrastructure and chip fabrication capacity represent the next market-moving discovery.

What this changes in practice

For investors, this podcast shifts the framework for evaluating technology companies. Rather than accepting market valuations based on revenue or growth rates, conversations push toward understanding competitive moats, capital intensity, and whether a business captures value from disruption or becomes disrupted. An episode featuring the founders who raised $800 million to compete against NVIDIA illustrates this dynamic: how do capital-heavy, technology-driven industries actually produce winners?

Business strategists extract frameworks for capital allocation in uncertain environments. What questions should leaders ask when evaluating whether to build in-house AI capabilities, partner with external providers, or acquire technology? The podcast demonstrates how world-class decision-makers think through these dilemmas without certainty—acknowledging ambiguity while moving decisively forward.

The podcast also shifts time horizons. Rather than quarterly earnings or annual guidance, conversations focus on five-to-ten-year structural shifts in technology economics, capital flows, and competitive positioning. This extended timeframe better matches how infrastructure investments actually pay off and how technological adoption curves truly operate in markets.

The clearest signal from leading investors emerges when they abandon consensus assumptions about market structure, cost trajectories, and competitive dynamics—revealing that opportunities exist precisely where sophisticated markets most systematically misprice risk and capability.

Discover how top investors identify opportunities others miss with Invest Like The Best.

Start listening to Invest Like The Best now.

The podcast answers these questions

What investment frameworks do top investors use?

Leading investors combine deep sector analysis with pattern recognition across multiple markets. They emphasize understanding competitive dynamics, management quality, and emerging technological shifts that reshape industry economics. Real-time insights from boardroom conversations reveal how institutional investors adapt frameworks when market conditions shift unexpectedly.

How is artificial intelligence reshaping capital allocation?

AI is fundamentally altering cost structures, competitive advantages, and market sizing across industries. Investors are reassessing infrastructure requirements, chip availability, and the concentration of computational power. The emerging consensus shows capital deployment accelerating toward AI infrastructure while traditional market valuations may not yet reflect the scale of this structural transition.

What do conversations with business leaders reveal about strategy?

Boardroom insights show leaders navigating contradictions between growth velocity and capital efficiency. Strategic conversations reveal how companies position for AI adoption, manage regulatory uncertainty, and build competitive moats in fast-moving markets. These unfiltered discussions expose real decision-making processes hidden from public filings and investor presentations.

Why do professional investors listen to business podcasts?

Professional investors extract proprietary insights by hearing directly from decision-makers about market dynamics, capital allocation priorities, and emerging risks. Podcasts compress years of market observation into focused conversations, offering alternative data sources that complement traditional research and provide early signals of significant market shifts.

Invest Like The Best

Discover Invest Like The Best

Patrick O'Shaughnessy · Invest Like The Best

Listen to the podcast

Découvrir Invest Like The Best sur Listenly →