Gavin Baker, founder and CIO of Atreides Management, joins Patrick O'Shaughnessy for their annual deep dive into the rapidly evolving AI landscape. Baker brings his encyclopedic knowledge of technology markets and his unique perspective as both a public markets investor and venture capital participant.

The conversation spans the technical intricacies of AI scaling laws, the geopolitical implications of chip development, and the economic dynamics reshaping entire industries. Baker explains how Gemini 3's release confirmed critical pre-training scaling laws, why Blackwell's complexity created an 18-month deployment gap, and how reasoning models emerged as an unexpected bridge.

They explore Baker's fascinating thesis on data centers in space, the strategic chess game between Google, NVIDIA, and emerging players like XAI, and why traditional SaaS companies risk obsolescence by refusing to embrace AI's lower-margin reality. The discussion concludes with Baker's personal investing origin story, tracing his path from aspiring ski bum to one of technology's most passionate analysts.

Processing AI Updates and Following the Signal

Baker emphasizes using paid AI tiers rather than free versions: 'the free tier is like dealing with a 10-year-old and making conclusions about the 10-year-old's capabilities as an adult'

AI development happens on Twitter with incredible signal-to-noise ratio, requiring close following of key researchers from the four labs that matter

Andre Karpathy's writings require reading 'three times minimum' and any podcast appearances from frontier lab researchers are essential listening

Gemini 3 and the State of Scaling Laws

Gemini 3 proved scaling laws for pre-training remain intact, despite our understanding being 'kind of like ancient British people's understanding of the sun' - precise measurement without comprehension

Two new scaling laws emerged: reinforcement learning with verified rewards and test-time compute, enabling progress during the Blackwell delay period

Andre Karpathy's principle applies: 'with software, anything you can specify, you can automate. With AI, anything you can verify, you can automate'

The Blackwell Complexity Crisis and Market Implications

Blackwell represented the most complex product transition in technology history, going from air-cooled 1,000-pound racks to liquid-cooled 3,000-pound racks consuming 130 kilowatts

Without reasoning models, AI would have stalled for 18 months during Blackwell delays: 'reasoning kind of saved AI because it let AI make progress without Blackwell'

Google's temporary advantage using TPU v6 and v7 (F-4 Phantom era) while Blackwell (F-35 era) remained difficult to deploy at scale

XAI will likely deploy first Blackwell model in early 2026 because 'according to Jensen, no one builds data centers faster than Elon'

Google's Strategic Cost Advantage and Coming Disruption

Google has been 'sucking the economic oxygen out of the AI ecosystem' as the lowest-cost token producer, running AI at negative 30% margins

This strategy becomes unsustainable when Blackwell models make Google no longer the low-cost producer, forcing strategic recalculation

Google's TPU development faces challenges from paying Broadcom 50-55% gross margins on estimated $30 billion in 2027 TPU spending

Data Centers in Space: The Ultimate Infrastructure Play

Space-based data centers offer fundamental advantages: 6x more solar irradiance, 24-hour sun exposure, no battery requirements, and free cooling via radiators

Satellite networks connected by lasers through vacuum would be faster than fiber optic connections in terrestrial data centers

Direct satellite-to-phone communication eliminates the complex terrestrial routing path, providing superior user experience for inference

Implementation requires 'a lot of those Starships' for economical launch capability, with SpaceX being the only viable provider currently

AI ROI Reality and Fortune 500 Adoption

AI ROI has been 'empirically, factually, unambiguously positive' for public companies, measurable through audited quarterly financials and ROIC calculations

C.H. Robinson exemplifies AI transformation: freight forwarding company reduced quote time from 15-45 minutes to seconds while increasing quote rate from 60% to 100%

Fortune 500 companies showing first quantitative AI uplift examples in Q3, with C.H. Robinson stock rising 20% on AI-driven earnings beat

VCs see clear productivity gains with companies achieving 'significantly lower employees for a given level of revenue' compared to two years ago

SaaS Companies' Strategic Misstep with AI Margins

SaaS companies making 'exact same mistake that brick and mortar retailers did with e-commerce' by refusing to accept AI's lower margin structure

Traditional SaaS enjoys 70-90% gross margins while AI companies operate at 40% margins due to compute costs per transaction

Companies like Salesforce, ServiceNow, and HubSpot could run agent strategies but resist due to margin preservation concerns

Microsoft stands as the exception, successfully navigating the transition while others face 'burning platform' scenarios

The Great Game: Competitive Dynamics and Geopolitics

Four American frontier labs (OpenAI, Anthropic, XAI, Google) widening gap versus Chinese open source due to Blackwell access restrictions

DeepSeek acknowledged in technical paper v3.2 that they 'struggle to compete with American Frontier Labs' due to insufficient compute access

China's strategic miscalculation in rejecting Blackwell chips will become apparent by late 2026 when the performance gap widens dramatically

Anthropic's $5 billion NVIDIA deal signals Dario's recognition of Blackwell/Rubin advantages over TPU architecture

Gavin Baker's Investing Origin Story

Baker's path began with childhood fascination with history, progressing from 'pictures of Phoenicians and Egyptians' to daily current events consumption

Originally planned to be a ski bum, river guide, and wildlife photographer before parents requested 'just one professional internship'

DLJ internship involved mailing research reports to clients, but reading the reports sparked realization: 'this is the most interesting thing imaginable'

Early education included One Up On Wall Street, Market Wizards, Warren Buffett's Letters to Shareholders (read twice), and Why Stocks Go Up and Down for accounting fundamentals

Investing appealed as 'greatest game of skill and chance imaginable' where edge comes from 'thorough knowledge of history intersected with current events'