Sam McCandlish Co-founder of Anthropic, its chief architect since October 2025 and earlier its chief technology officer; second author of the neural scaling-laws paper Founder · Capital · $10B+ cost 3 headwinds 1

← browse

Founder · Capital · Artificial Intelligence · Technology · Software · $10B+

Sam McCandlish

born Samuel Russel McCandlish

Co-founder of Anthropic, its chief architect since October 2025 and earlier its chief technology officer; second author of the neural scaling-laws paper · b. 1990 · United States (birthplace not established)

class not establishedhousehold not establishedplace not established

OpenAI Anthropic

Cost of failure 3 / 10

soft landingnothing to catch a fall

Headwinds 1 / 10

clear runagainst the current

Two calls made by hand, not formulas. how we score →

Summary

McCandlish trained as a theoretical physicist and finished a Stanford PhD in 2017.

He then moved into machine learning at OpenAI, was second author on the 2020 scaling-laws paper, and co-founded Anthropic in 2021. Forbes puts him at $15.5 billion, almost all a private stake. His family and childhood are undocumented.

Coded record

industryArtificial Intelligence · Technology · Software

talentexceptional

connectionswell-connected

outcome size$10B+ · band 6

path typeFounder · Capital

childhood householdtwo-parent

immigrant generationnone

educationgraduate degree

credential fundingnone

startup capitalangel

took outside investmentyes

kept ownershipyes

public scrutinyjournalistic

ⓘ Hover any row for its definition.

Profiled colleagues

People in this set who share a company with Sam McCandlish.

Subject cooperation: public-only · last reviewed 2026-10-19

Where they worked

OpenAI Researcher (scaling-laws and GPT-3 papers) past Anthropic Co-founder; chief architect (previously chief technology officer) · 2021–now founder still there

How it happened

iPhysics

The Mathematics Genealogy Project lists Samuel Russel McCandlish with a Stanford PhD in 2017, dissertation "Depth Perception in Holography," advised by Eva Silverstein, and Wikidata gives the same degree and advisor. Forbes says he has a PhD in theoretical physics from Stanford. Secondary wiki sites also mention math and physics degrees at Brandeis and a postdoc at Boston University, but none of those pages cite a source for them, so they are left out of the coded record. His parents, where he grew up, how his schooling was paid for and the birth year are not established, and the 1990 here is inferred from Forbes giving his age as 36 in October 2026.

iiInto machine learning

He was a researcher at OpenAI, according to wiki-style profiles that disagree on the dates (2017 or 2018 to the end of 2020 or January 2021). The arXiv listing for "Scaling Laws for Neural Language Models," submitted January 23, 2020, names him second among ten authors, behind Jared Kaplan, with Tom B. Brown fourth and Dario Amodei last, and it lists him as the submitter. He is also the 28th of 31 authors on the GPT-3 paper, submitted in May 2020.

iiiAnthropic

Turning point

Anthropic was founded on January 26, 2021, per Wikipedia, whose infobox lists him among eight founders, including Dario Amodei, Daniela Amodei, Kaplan, Jack Clark and Brown. Forbes calls him cofounder and chief technology officer. Reporting from early October 2025 says Rahul Patil, formerly Stripe's CTO, took over that title and McCandlish became chief architect, working on pre-training and large-scale model training and reporting to Daniela Amodei. What set him apart from other physics PhDs was being a lead author on the paper that gave the company's main bet its evidence, and a group of senior colleagues who left OpenAI together to start a lab. His own account of the decision is not established in the sources used.

ivWhere it landed

Forbes's profile page on October 19, 2026 shows $15.5 billion and a real-time rank of 182, lists him as self-made, and says private investors valued Anthropic at $965 billion in May 2026. A January 2026 Fortune article cited an earlier Forbes estimate of about $3.7 billion for each cofounder and said they had committed to give away 80 percent of their wealth. The $15.5 billion is an estimate of a private stake, and how the founders' shares divide is not established.

vCompany disputes

Per Wikipedia, authors sued Anthropic in Bartz v. Anthropic over books used to train its models. A court held in June 2025 that training on lawfully obtained copies was fair use but that keeping pirated library copies was not, and the company agreed to a $1.5 billion settlement. Music publishers have also sued, and the same article reports a 2026 dispute with the Defense Department over a supply chain risk designation. These concern the company. McCandlish is not described as personally named in the sources used.

Can you replicate their success?

Partly

The early steps are open to a small number of very strong students: a physics PhD at a top department and a move into machine learning while scaling results were still being worked out. The founding step is much narrower. It followed authorship of a paper that became central to how the field plans its compute, colleagues who left together, and investors willing to fund very large training runs. Family background is not documented, so how much a safety net mattered is unknown.

Required conditions

1 Research ability at the level of a Stanford theoretical physics PhD

2 A move into machine learning when scaling results were still being established

3 Senior colleagues willing to leave a lab and start a company together

4 Investors able to fund compute-heavy research at very large scale

The coded evidence

Thirteen groups, every claim sourced

Feeds cost of failure

Parent Occupations

Low

not established

Not established in any source used. The household value in the coded block is a placeholder because the enum has no unknown option.

↗ forbes.com

Parent Education

Low

not established

Not established.

↗ forbes.com

Structural context

founder lens · venture capital

The cost-of-failure score comes from this person's own money and circumstances. But who they were in America carried its own weight, the same way their household or income did, and it shaped how hard the path was just as much. That part isn't in the cost-of-failure number. It's here instead.

tailwindman

Men founded the companies that took nearly all the venture funding and almost every top outcome, a tailwind that never shows up in one person's own circumstances.

↗ techcrunch.com

Among the people recorded here — men: 225. Representation here is who reached these outcomes, not equal odds of reaching them.