Gavin Baker
InvestorManaging Partner at Atreides Management
Boston, Massachusetts
Writes $250K – $50M checks · typically $5M
About
Gavin Baker spent eighteen years at Fidelity, running the OTC Portfolio from 2009 to 2017 and helping build the firm's venture investing arm, before founding Atreides Management in 2019 as a crossover firm investing across public and private technology and consumer companies. He is one of the most technically detailed public voices on the AI build-out, organising his view around what he calls watts and wafers — electrical power and semiconductor capacity as the two constraints that decide who can build. His private positions cluster on that thesis: AI networking silicon, processor IP, inference accelerators, storage and battery materials.
Beyond investing
- Education
- Studied at Dartmouth College (BA, Economics and History)
From Gavin's own website — common ground for a warm intro.
Experience
- Atreides Management, LPCIO & Managing Partnercurrent2019 — presentBoston
- Fidelity InvestmentsPortfolio Manager, Fidelity OTC Fund1999 — 2017
Skills & more
From Gavin's professional profile.
Investments
Per curated public sources.
Public positions
He reasons from physics and supply chains rather than from narratives. The recurring move is to find the input that cannot be scaled quickly — a wafer, a gigawatt, a permit, a DRAM bit — and ask who gets paid because of it. He is unusually willing to publish the full argument rather than the conclusion, including the condition under which he would be wrong.
2026 · AI infrastructure
He frames the next phase of AI around two physical constraints — watts and wafers — arguing these determine capacity far more than model quality does.
The organising idea of his current portfolio.
2026 · Bubble debate
He argues this build-out is not the dot-com fibre build because it is funded from operating cash flow against near-full utilisation rather than from debt against unused capacity, and names rising private credit as the indicator that would change his view.
A falsifiable version of the bubble argument with a named tripwire.
2026 · AI applications
He contends that AI application companies have on net destroyed economic value so far, and that durability requires either a data moat or control of the tokens.
One of the most contrarian positions held by a major AI investor.
2026 · Power
He expects the acute power shortage to ease around 2027 and 2028 as new supply arrives, but views permitting and zoning as the slower constraint behind it.
A dated, checkable forecast.
2026 · Capital access as a moat
He argues frontier labs deliberately manage burn and pricing to preserve investor goodwill, treating access to capital as a competitive asset in its own right.
Reframes lab economics as a financing question rather than a product one.
2026 · Valuation
He points to an internal contradiction in AI-adjacent valuations — semiconductor equipment trading at high multiples while memory trades in the mid single digits — and reads it as a late-cycle signal.
A rare specific valuation critique from within the trade.
2025 · Memory
He argued a memory shortage would follow from AI's data-movement demands and built positions accordingly.
His most concrete recent supply-chain call.
▶All 8 tracked positions
2026 · Nvidia valuation
He has argued publicly that the bear case on Nvidia does not survive contact with the underlying demand data.
A directly checkable position on the most-watched stock in the trade.
Prediction record
The memory-shortage call was right and profitable. The open calls are dated and checkable: power easing around 2027–2028, wafer discipline preventing an overbuild, extended GPU useful life, and the application layer staying net value-destructive without moats. His Nvidia position is expressed with leverage, which he has been public about.
AI's data-movement requirements would create a memory shortage
Correct2025 · Quarters
He increased memory positions in late 2025; those names delivered very large gains through 2026.
Reported from 13F positioning and press analysis rather than a dated public forecast.
The acute power shortage eases around 2027–2028 as new generation comes online
Open2026 · 2027-2028
Unresolved.
He separates generation from permitting, and expects the latter to clear more slowly.
Wafer capacity discipline prevents an AI overbuild
Open2026 · Years
Unresolved.
Depends on foundry capital-expenditure decisions he does not control.
The AI application layer remains net value-destructive without data moats
Open2026 · Years
Unresolved and contested.
He is positioned in infrastructure rather than applications, so the call is not disinterested.
Splitting inference phases could extend GPU useful life to a decade or more
Open2026 · Years
Unresolved; depreciation schedules across hyperscalers have been lengthening.
A significant claim, because it changes the financing maths for the whole build-out.
The Nvidia bear case fails against the demand data
Open2026 · Quarters
Unresolved.
He holds a levered position, which he has disclosed.
Controversies & responses
Where Gavin has drawn public criticism — with what happened, their response, and the criticism itself, side by side.
▶Concentration and leveraged exposure2026
- What happened
- Reporting has described him expressing his Nvidia conviction through leveraged call options rather than shares alone.
- The response
- He argues the demand data does not support the bear case and has been explicit about his reasoning in public.
- The criticism
- Options-expressed conviction magnifies both the upside and the drawdown, and a manager broadcasting a levered position invites the charge of talking his book.
- Where it stands
- Ongoing; the thesis is unresolved.
▶An unusually public investor2026
- What happened
- He is among the most active large-fund managers posting detailed technical analysis on social media, which moves attention in the names he discusses.
- The response
- He publishes his reasoning in full rather than only his conclusions, which makes the arguments checkable.
- The criticism
- Critics of investor-commentators note the difficulty of separating analysis from positioning when the analyst holds the position.
- Where it stands
- Ongoing and openly acknowledged.
Podcasts & interviews
featuring Gavin BakerColleagues at Atreides Management
Network9
Investors connected to Gavin in public investor directories — a warm-intro map.
Frequently asked questions
- Who is Gavin Baker?
- Gavin Baker spent eighteen years at Fidelity, running the OTC Portfolio from 2009 to 2017 and helping build the firm's venture investing arm, before founding Atreides Management in 2019 as a crossover firm investing across public and private technology and consumer companies. He is one of the most technically detailed public voices on the AI build-out, organising his view around what he calls watts and wafers — electrical power and semiconductor capacity as the two constraints that decide who can build. His private positions cluster on that thesis: AI networking silicon, processor IP, inference accelerators, storage and battery materials.
- Where does Gavin Baker work?
- Gavin Baker is Managing Partner at Atreides Management.
- What check size does Gavin Baker write?
- Gavin Baker typically writes checks of $250K – $50M.
- What companies has Gavin Baker invested in?
- Investments tracked here include AcuityMD, Decart, Enfabrica, Lightbits Labs, Positron AI, SiFive, Sila Nanotechnologies, Skyryse.
- Where did Gavin Baker go to school?
- Gavin Baker studied at Dartmouth College (Bachelor's Degree - AB, History and Economics).
- What podcasts and interviews feature Gavin Baker?
- Tracked appearances include "Watts and Wafers — Invest Like the Best", "Gavin Baker on AI investing: follow the money".
Same school
Also attended Dartmouth College494
+ 488 more
Shared employers
Also worked at Fidelity Investments183
+ 177 more