How Gavin Baker’s insights point directly to the future we’re building at NewTide AI
A Conversation Worth Your Time
Every so often, a discussion surfaces that does more than explain technology—it provides a framework that helps leaders see where the world is actually heading. Patrick O’Shaughnessy’s recent Invest Like the Best interview with Gavin Baker is one of those rare conversations. In the episode, “Nvidia v. Google, Scaling Laws, and the Economics of AI,” Baker offers a clear, grounded explanation of the forces shaping the next decade of artificial intelligence.
For anyone operating in industries where precision, logistics, margin, and operational execution determine success—trading, transportation, energy distribution, retail operations—his insights are especially relevant. Baker’s background as founder of Atreides Management, and previously a leading technology portfolio manager at Fidelity, gives him a uniquely credible vantage point. His understanding of semiconductor economics, compute constraints, hyperscaler strategy, and enterprise adoption patterns allows him to articulate not just what AI can do, but why certain capabilities will matter far more than others.
We strongly encourage listening to the episode in full, because Baker doesn’t speak in abstractions; he speaks in mechanisms, economics, and engineering realities. Those are exactly the kinds of insights that align with NewTide’s perspective on the future of vertical AI platforms like RisingTide.
🎧 Listen here:
https://joincolossus.com/episode/nvidia-v-google-the-economics-of-ai/
What follows is not a summary of the interview, there are plenty of those online if you need one after listening, what we did was try and develop a synthesis of how Baker’s most important themes intersect directly with the platform and solutions we are building for our customers.
The New Scaling Laws Shift the Advantage to Application Platforms
One of the most important observations Baker makes is that the greatest progress in AI over the last two years has not come from simply training ever-larger models. Rather, it has emerged from advances in post-training methods—especially reinforcement learning with verifiable rewards (RLVR) and test-time compute. These innovations allow models to continue improving in real-world conditions, learning from structured tasks, feedback loops, and measurable outcomes.
We have been doing this with our clients in cases where the raw horsepower of the big models themselves is not sufficient to execute key tasks with the precision that true process automation requires. For many use cases simply getting the data flowing and knowing where you want it to go can be fairly easily done with out of the box intelligence that the models already possess. Think automating document processing around BOL’s, Invoices, Orders etc. For more complex tasks like trading intelligence we have found that additional training is required. The key is that you have to get enough data to work with into the right training environment to do that work, which is one reason we have invested in our Data Helm components so heavily. In comparison that might resonate with anyone who is a fan of Malcolm Gladwell’s Outliers book, we do generally see that once you can get 10,000 reps with a good verifiable data you can develop much more specific intelligence around a situation.
This matters enormously for operational industries. Baker’s line—“anything you can verify, you can automate”—captures the opportunity with unusual clarity. Many business domains are full of ambiguous, judgment-driven work; but the ones NewTide serves are full of work that is inherently verifiable. A trade either matches the confirmation or it doesn’t. A bill of lading either reconciles to the order or it doesn’t. A price change is either executed correctly at the store or it isn’t. These are precisely the environments where RLVR thrives.
As a result, the advantage increasingly shifts away from generalized intelligence and toward application-specific platforms that can leverage verification loops to produce reliability at scale. RisingTide’s architecture—Data Helm for clean ingestion, Agent Harbor for domain-specific agents, and Shipyard for managed workflows—was built for exactly this world. Baker’s insights validate that the future belongs to systems capable of applying frontier models to verifiable, mission-critical operations, not to generic AI wrappers with minimal context.
Model Intelligence Is Plateauing—But Usefulness Is Accelerating
Another powerful theme from Baker is his distinction between improvements in model intelligence and improvements in model usefulness. He argues that we may be entering a period where incremental jumps in perceived “IQ” become less noticeable to most users, even as the practical utility of these models rises dramatically.
Usefulness is not determined by benchmark performance; it is determined by the system surrounding the model. How much context can it retain? How well does it follow multi-step instructions? How effectively can it integrate with external systems? Can it execute actions reliably, not just generate text? These questions have little to do with foundational training and everything to do with the scaffolding built around the model—data pipelines, workflow engines, verification processes, domain-trained agents.
This is where vertical AI platforms begin to separate from general-purpose AI tools. RisingTide is not designed to be a chat interface with intelligence. It is designed to be a work system—one that extracts, validates, reconciles, schedules, optimizes, and completes. As frontier models get marginally “smarter,” the practical impact will be experienced most by the platforms that can convert that intelligence into dependable operational throughput. Baker’s framing helps make sense of this shift: we are now entering the usefulness era, not the intelligence era.
Compute Economics Favor Companies That Turn Tokens Into Revenue
Baker is unequivocal about this point: the companies already seeing the greatest financial returns from AI are those that have embedded it into workflows that produce measurable economic outcomes. His example of C.H. Robinson is illustrative, not because it is unique but because it is replicable. By integrating AI into its quoting workflow, the company shortened response times from tens of minutes to seconds and expanded its ability to quote nearly all inbound requests. The payoff was immediate and material.
The lesson is that AI’s economic value does not come from model ownership or internal research labs. It comes from operational leverage—reducing latency, increasing coverage, expanding capacity, and eliminating costly exceptions. These are precisely the kinds of advantages that matter in energy trading, fuel logistics, aviation supply, and retail operations.
RisingTide’s mission is not to create intelligence for its own sake; it is to build agents and workflows that materially enhance throughput and margin across the make–sell–support cycles of these industries. Baker’s analysis reinforces a simple truth: the winners in AI will be the companies that turn tokens into revenue, not the ones that merely consume them.
The Frontier Battle Creates a Permanent Need for Vertical Translation Layers
Baker also describes the intensifying competition among frontier labs—OpenAI, Google, Anthropic, xAI—and the feedback loops that now accelerate the gap between leaders and everyone else. As these labs ship more frequent checkpoints with ever-increasing capabilities, the idea of enterprises training their own frontier models becomes increasingly unrealistic.
This does not diminish the importance of AI for enterprise operators; it actually increases the urgency for translation layers—platforms capable of integrating frontier intelligence into the systems companies already use. Baker’s view aligns with our own long-standing belief that vertical platforms will form the connective tissue between hyperscaler innovation and real-world workflows.
Enterprises don’t need to compete with frontier labs. They need a partner that continuously absorbs frontier advancements and embeds them into domain-specific workflows with reliable guardrails, governance, and verifiability. That is exactly what RisingTide is built to do.
The Real Threat Isn’t Other Vendors—It’s Fragmentation and Edge AI
Perhaps the most forward-looking element of Baker’s analysis is his discussion of edge AI. With model compression, distillation, sparsity, and hardware improvements, companies will soon be able to run highly capable models locally—on a phone, laptop, or embedded device. This creates enormous opportunity but also complexity. If every device can host its own intelligence, the challenge becomes coordinating these capabilities securely, consistently, and at scale.
This is where vertical AI shifts from being helpful to essential. RisingTide is architected not as a monolithic intelligence layer, but as a coordination system—one that governs data flow, applies policies, routes tasks across cloud and edge, and ensures auditability across the entire lifecycle of a workflow. If the future is decentralized intelligence, the platforms that orchestrate that intelligence gain disproportionate value.
Baker’s point is clear: the great challenge of the next wave of AI is not model scarcity, but system fragmentation. Vertical platforms will be the antidote.
AI’s Largest Dividend Will Come From Reinventing the Make–Sell–Support Cycle
Baker reduces companies to three essential functions: make things, sell things, and support customers. AI, he argues, will transform all three—not through futuristic robotics or speculative use cases, but through continuous reductions in time, cost, and friction.
In the industries NewTide serves, these functions are deeply intertwined. “Making” includes trading portfolios, delivery schedules, inventory positioning, dispatch operations, and site performance. “Selling” includes fuel pricing, wholesale contracting, account targeting, and customer engagement. “Supporting” includes reconciliation, confirmations, invoice processing, regulatory compliance, and logistics coordination.
Vertical AI platforms like RisingTide are uniquely positioned to enhance all three simultaneously because the underlying workflows share common attributes: structured data, verifiable outcomes, time sensitivity, and high financial leverage. Baker’s framework underscores why these domains will be among the fastest to benefit from AI adoption.
Conclusion: Why Baker’s Perspective Matters for NewTide
Gavin Baker’s credibility comes from decades spent studying the dynamics of technological disruption, from the economics of semiconductor manufacturing to the strategies of cloud hyperscalers, to the realities of enterprise adoption. His framework is not an academic exercise; it is rooted in the mechanisms that determine where value will actually accumulate.
His analysis affirms what we believe at NewTide:
General-purpose AI will push the frontier, but vertical AI built with enough depth in the industry it serves will convert that frontier into economic impact.
That is the mission of RisingTide: to transform the most advanced AI capabilities in the world into trusted, verifiable, high-value workflows for trading, logistics, fuels, and retail operations.
If you want to understand the technological and economic currents shaping that future, Baker’s episode is well worth your time.
🎧 Listen here:
https://joincolossus.com/episode/nvidia-v-google-the-economics-of-ai/
