Gianfranco's January 2026 Reading List

What’s worth reading this month—and why does it matter?

Each day, I devote three hours to reading. Occasionally an author writes a piece that feels like a gift—crafted with intention for the reader. It tends to be a piece that encourages a reader to pause, invites them to feel how a strong idea, delivered with empathy, can resonate.

Few writers achieve this.

Each selection did one of three things for me: it sharpened a thesis, challenged a prior, or opened a new line of inquiry. That’s the bar for inclusion.

In this curation practice, I believe the hero of this story isn’t the author of any single essay, nor me as curator, but you—the reader.

If these pieces don’t move you, I haven’t honored the attention you’ve given me. Hold me to that.

Use this as a tasting menu: no set order, categories for context, and a brief note on what each means for builders and capital allocators.Subscribe

AI Systems & Interfaces

• Bubble Trouble 2 - Marc Rubinstein
  • Marc Rubinstein draws tight parallels to 2008: neoclouds like CoreWeave are issuing massive debt backed by GPU collateral under optimistic six-year useful-life assumptions, while hyperscalers quietly signal shorter cycles and rapid obsolescence. A new H100 rental index launched in 2025 is already trending down, exposing compute pricing dynamics the way ABX did for subprime mortgages. Rising defensiveness against short-sellers from Altman, Karp, and others echoes bubble-era rhetoric as credit spreads widen and collateral values come under pressure.
  • Read More

• TPUv7: Google Takes a Swing at the King - Dylan Patel and team
  • Dylan Patel and colleagues outline Google’s push to sell its custom AI chips (TPUs) to external customers, anchored by Anthropic’s commitment to roughly 1 million TPUv7 units. Independent modeling shows these chips deliver 30–52% lower total cost of ownership than Nvidia’s latest Blackwell/GB300 for the same effective AI performance, owing to more grounded specifications, an efficient large-scale interconnect design, and stronger real-world utilization. The competitive presence has already extracted ~30% discounts from Nvidia for OpenAI, while broader adoption by Meta, xAI, and others quietly alters financing models for new AI cloud providers.
  • Read More

• Nvidia v. Google, Scaling Laws, and the Economics of AI - Gavin Baker
  • Gavin Baker frames the AI buildout as an infrastructure arms race: Nvidia sells the picks, Google aims to be the lowest-cost producer of tokens via TPUs, and scaling laws push the true constraints down to compute, power, and supply chains. The discussion ranges from frontier-model dynamics to “power as a bottleneck” and even data centers in space, then lands on a sober warning that many SaaS companies are budgeting and pricing as if software margins will stay untouched—right as AI makes disciplined capital allocation the differentiator.
  • Read More

• The AI user's guide to evals - Justin
  • Justin describes a pattern many encounter: a prompt chain performs beautifully for a few runs, then begins hallucinating, changing languages, or silently losing reliability, eroding trust. His core insight is that the solution lies in evals—adapted software-testing techniques that use careful log review and measurement to catch and fix these issues systematically. His workflow is deliberately unsexy: build an evals dataset (~100 traces in a spreadsheet), patch recurrent failures with prompt tweaks and simple assertions, and postpone “LLM-as-judge” complexity until you’ve exhausted cheap checks—otherwise you’re running a “slot machine” that occasionally pays out productivity.
  • Read More

Operator Playbooks

• The Shape of The Game We Play - Cedric Chin
  • Cedric Chin contrasts business with weightlifting: physical skills demand direct reps, but business exposes operators to rare, regime-shifting “10-year floods” like the post-2022 end of ZIRP, where cheap capital distortions frayed assumptions many internalized over 15 years. Regime changes arrive subtly and delayed, catching those without historical perspective off-guard, while practices hailed as best prove regime-specific and suspect in tighter capital environments.
  • Read More

• Giving your AI a Job Interview - Ethan Mollick
  • Ethan Mollick argues public benchmarks trend upward but falter on writing, judgment, or advice—often contaminated or misaligned—while real differences in model “vibes,” risk tolerance, and task strengths emerge only through repeated, realistic testing. As AIs scale into decision-making roles, treat selection like hiring: run rigorous “job interviews” with your actual use cases, expert blind grading, and probes of consistent attitudes to match the jagged frontier precisely, rather than relying on general scores or vibes.
  • Read More

• Stop Worrying About Survivorship Bias With This One Weird Trick - Cedric Chin
  • Cedric Chin argues that survivorship bias in business biographies is overblown for practitioners: experts like Buffett and Munger treat cases not as sources of direct lessons but as fragments instantiating concepts, building pattern-matching libraries via Cognitive Flexibility Theory to handle novelty in ill-structured domains. By recombining fragments rather than generalizing outcomes, they accelerate adaptive expertise—Buffett used annual reports to illustrate good vs. bad businesses, Munger drew from magazines on forces shaping retail declines—making bias irrelevant when the goal is calibrated sensemaking over universal rules.
  • Read More

Market Structure & Moats

• Robotaxis and Suburbia - Ben Thompson
  • Ben Thompson argues that Google’s Gemini 3—and, more importantly, Google offering TPUs to external partners—changes competitive math: Nvidia’s premium margins now face a credible alternative, and OpenAI can’t count on “best model” status as a permanent shield. Nvidia still benefits from GPU flexibility and CUDA, but concentrated hyperscaler buyers may finally find it rational to route around that software moat, while ChatGPT’s harder-to-dislodge advantage is consumer habit at massive scale.
  • Read More

• A window into modern loan origination - Patrick McKenzie
  • Patrick McKenzie explains perpetual futures (perps), crypto’s top trading product (6–8x spot volume), as cash-settled contracts without expiry that settle multiple times daily via funding rates—winners pay losers periodically, embedding interest and anchoring prices near spot through basis trades. While perps solve crypto casinos’ trust and capital-efficiency issues and back much stablecoin collateral, they’re unlikely to spread to traditional finance, where derivatives already manage leverage, liquidations aren’t core revenue, and retroactive deleveraging would alarm regulated players.
  • Read More

• Google, Nvidia, and OpenAI - Ben Thompson
  • Ben Thompson argues that Google’s Gemini 3—and, more importantly, Google offering TPUs to external partners—changes the competitive math: Nvidia’s premium margins now face a credible alternative, and OpenAI can’t count on “best model” status as a permanent shield. Nvidia still benefits from GPU flexibility and CUDA, but concentrated hyperscaler buyers may finally find it rational to route around that software moat, while ChatGPT’s harder-to-dislodge advantage is consumer habit at massive scale. The uncomfortable punchline is OpenAI’s monetization: an ads business could both fund its trillion-dollar compute commitments and deepen its aggregator moat, so refusing to build one turns this moment into a live test of Aggregation Theory—one Thompson says leaves him nervous and excited.
  • Read More

• Perpetual futures, explained - Patrick McKenzie
  • Patrick McKenzie explains perpetual futures (perps), crypto’s top trading product (6–8x spot volume), as cash-settled contracts without expiry that settle multiple times daily via funding rates—winners pay losers periodically, embedding interest and anchoring prices near spot through basis trades. Perps cut capital needs on exchanges by offering extreme leverage (20x–100x vs. traditional 2–4x), enabling frequent liquidations that profitably blow out over-levered positions and automatic deleveraging (ADL) when insurance funds fail in tail events, shifting risk to winners. While perps solve crypto casinos’ trust and capital-efficiency issues and back much stablecoin collateral, they’re unlikely to spread to traditional finance, where derivatives already manage leverage, liquidations aren’t core revenue, and retroactive deleveraging would alarm regulated players.
  • Read More

• More Than Money: Stablecoin Analysis - Knower
  • Knower sees stablecoins (now ~$303B supply) as crypto’s clearest success beyond BTC price, with transaction volume surpassing spot trading and adoption across payments, hedging, remittances, cross-border flows, and merchant use—driven by major issuers (USDT/USDC) plus emerging bespoke/app-specific tokens. Future paths range from dystopian digital-ID surveillance to modest integration into modern rails or a “slow” but steady expansion as digital money digitizes further—bullish overall but tempered by privacy risks and infrastructure challenges.
  • Read More

Capital & Macro

• (Revisiting in 2026) How to Miss By a Mile: An Alternative Look at Uber’s Potential Market Size - Bill Gurley
  • Bill Gurley challenges Aswath Damodaran’s 2014 $5.9B Uber valuation, arguing the $100B taxi/limo TAM and 10% max share assumptions ignore how Uber’s superior features—faster pickups, wider coverage, cashless payments, safety, civility—expand the market well beyond historical taxi use. Network effects (shorter waits, denser coverage, lower prices) plus cross-city scale could multiply TAM 3–6x and capture 2.5–12.5% of car ownership, implying Uber needs only 20–56% share in a $450B–$1.3T opportunity to justify much higher valuations.
  • Read More

• The gift card accountability sink - Patrick McKenzie
  • Patrick McKenzie explains gift card scams’ persistence and weak recourse: retailers outsource issuance to program managers (e.g., Blackhawk, InComm), creating an accountability sink where victims hear “we don’t issue cards, we just accept them”—shifting fraud away from retailer fraud teams. Gift cards receive explicit carveouts from Reg E protections and certain FinCEN/AML rules (e.g., low-value exemptions), balancing innovation and unbanked access against fraud risk. Lighter oversight, breakage economics, and fraud-chain incentives make scammers favor gift cards, leaving victims with police reports instead of chargebacks despite tens of billions in legitimate annual use.
  • Read More

• Is It a Bubble? - Marks
  • Marks argues that a bubble is chiefly a state of mind—envy, FOMO, and “this time is different”—and that the current AI boom, fueled by historically large spend on compute, follows the familiar arc of prior technology manias. He separates a possible bubble in companies’ behavior from one in market pricing, concedes that “inflection” bubbles can finance enduring infrastructure, but warns that uncertain winners and the growing use of debt make the downside non-theoretical. His prescription is character more than forecasting: avoid all-in/all-out thinking, stay selectively exposed with prudence, and keep an eye on the human cost he flags in a postscript—joblessness and the loss of purpose.
  • Read More

• Can we make America feel more affordable? - Noah Smith
  • Noah Smith observes that despite low inflation (~2.7%) and rising real wages, Americans remain deeply upset about affordability—polls rank cost of living as the top issue, with many demanding absolute price declines to heal the 2021-22 inflation scar. Broad deflation via prolonged high rates risks recession, debt spirals, and pricier mortgages, so targeted drops in salient prices like gasoline (via drilling surges, potential Venezuela intervention) and mortgage rates (Fed pressure) offer political relief without full deflation.
  • Read More