Meta can buy the next thing. It struggles to build one.

Where each bet sits, and where it came from

Shading marks the zone, marker shape shows origin. Read the red crosses: every one is a new category Meta built itself. The Edge band holds one survivor and two closures.

Instagram and WhatsApp were Edge-stage companies when Meta bought them. They are plotted in Core because that is where they sit now.

The scorecard

Seventeen bets chosen to span the portfolio, not an exhaustive count.

Reality Labs, year by year

Meta kept paying. The segment has lost money every year since 2020 and about $83.6bn in total, and Oculus has been funded since 2014. In Q2 2026 it lost $4.62bn on $431m of revenue, wider than a year earlier, even after VR spending was cut and about 70% of the budget moved to wearables. The money was patient. The next sections look at what it was measured against.

Bets that needed seven years got three

A Beyond bet runs on a 7 to 15 year clock. Here is how long Meta held each new-category bet it built itself.

Metaverse measured from the October 2021 rename to the 2026 VR pullback. Other dates are launch to shutdown. Workplace is left off because it ran about a decade (wind-down announced May 2024, final shutdown June 2026). It is the long-lived exception, and it also ended.

What the first four exhibits say together

• The wins arrived from outside

  Instagram and WhatsApp were bought, and Stories, Reels and Threads were copied onto surfaces Meta already owned. None needed a new user behaviour, so none needed a long clock.

• The losses were held to Core metrics

  Zuckerberg's stated goal was a billion people in the metaverse by 2030. Judged by that, a Level 4 bet looks like failure at year four. The framework says it was never measured against the right rule.

• The money stayed, the rules did not change

  Reality Labs got a decade of capital. It still ran on Core expectations, so funding alone bought no extra time.

• The exception proves it

  Glasses are the one Edge bet that works, built with a partner and allowed to find early adopters first. They now take most of the Reality Labs budget.

• The Core pays for the retreat and the next bet

  2026 capex is guided to $130 to $145bn, mostly AI infrastructure. The Beyond is now a spending line in the Core's budget, which is the same structure the earlier bets failed under.

Core and Beyond, with nothing in between

The book says some companies hold only Core and Beyond, and the danger is that nothing is ready to take over when the Core slows. Meta fits. Its Edge is bought, borrowed or closed, and the Core already shows strain with Gen Z and Millennials, who largely do not join Facebook.

Of the Edge bets Meta started itself, none survives as a business. The one working Edge is a partnership that followed a competitor.

Meta separated the Beyond, then ran it like the Core

Meta got the first move right. It split Family of Apps from Reality Labs. The book says separation is only the start, because the Beyond needs different rules, people, metrics and money. Here is the comparison.

The three stages, and what Meta did instead

The book gives every breakthrough three stages: Chaos, Growth and Scale. Each has its own rules, people, metrics and leadership. The metaverse went from a company rename to a mass-market target.

• Chaos needs trust

  This stage needs people who feel safe to prototype something ugly and say "this isn't working". Project OT went the other way. Reuters reported that Meta captured keystrokes and mouse clicks on US staff devices so AI agents could learn the work. Internal sentiment fell from 74% to 55% favourable. Code changes rose 220% while features reaching users rose 36%, major incidents rose 40% and firefighting time rose 70%. Zuckerberg called off the second wave hours before the first, and about 8,000 people were cut anyway. CTO Andrew Bosworth wrote in an internal memo that Meta "did an atrocious job explaining the vision" and how it would support people's careers through the shift.

  You cannot build the next thing when the builders believe the company is watching them to learn how to replace them. Block pointed AI at the organisation's own artifacts and gave people context and autonomy. That is an Edge bet run as an Edge bet.

  Frontier research needs the same stability. Superintelligence Labs was reorganised four times in about six months, and Andrew Tulloch left for Anthropic after the Muse launch. TBD Lab is the one protected pocket, and it exists because Llama 4 stumbled.

• Growth was skipped

  A Transformational bet should spend years finding a small group of early adopters, then cross to the mainstream one iteration at a time. The book puts early adopters at about 13% of users and calls the gap to the majority a chasm. Meta renamed the whole company around the metaverse before any of that was proven, then judged it by mass-market numbers. VRChat grew slowly from a small crowd and built a real community. Meta had the distribution and not the patience to grow loop by loop.

  The book's warning for founders applies: funding is not proof. Vision picks the hill, and early adopters tell you whether it is the right one.

• Scale was never reached

  The book's Scale stage is where the parent company comes to collect. Meta let Instagram and WhatsApp run with autonomy for years, and both founding teams still left. Supernatural was bought for $400m after an FTC fight, frozen in January 2026, and spun back out to its founders in June. One user wrote after the coaches were let go: "We all felt like it was purchased to kill."

  The book names a better route for what a parent cannot absorb: let it go early, as Alphabet did with Waymo and Verily. Llama open weights was Meta's one asset that ran on the open-source ethos the Scale stage relies on, and Meta dropped it for a closed model. Qwen passed 3 billion downloads by August 2026.

The credit side of the ledger

The Meta Enterprise Platform

Zuckerberg calls the new enterprise platform the next major pillar of the business. Meta hired CJ Desai to run it. He had been MongoDB's CEO for about a year and left the day before its Investor Day, and the stock fell as much as 26.9%. The platform sits on Muse, which has consumer traction, and Desai reports to Zuckerberg. Workplace ended less than four months before the announcement.

The book gives three questions to ask in the first year.

• Is it staged?

  Does it get a Chaos and a Growth stage with a few design partners, or an enterprise launch to the mainstream?

• Is the scoreboard ring-fenced?

  Are KPIs, budget and timeline set apart from the Core from day one, or will it be judged on its first quarterly report?

• Is the clock honest?

  Selling to CIOs against Microsoft, Google and Amazon is a long game. Does the plan say so?

Where the figures come from

• Project OT

  Meta planned to cut some teams by 60% and replace the work with AI, The Next Web, citing Reuters.

• Executive memo

  Meta CTO Andrew Bosworth, internal memo, WIRED, 14 June 2026.

• AI reorganisations

  Meta splits AI division into four new groups, Capital Brief, citing The Information.

• Supernatural

  Meta mercifully spun out Supernatural, TechCrunch. Also The Next Web and Road to VR.

• Horizon Worlds

  Meta shutting down VR version of Horizon Worlds, Fox 10. Reversed in part: Euronews. User numbers: CNBC, NBC New York.

• Metaverse target

  Zuckerberg's June 2022 CNBC interview, via ExtremeTech.

• Quest and Oculus

  Meta stops selling Quest to businesses, SiliconANGLE. Brand retirement: KitGuru.

• Superintelligence Labs talent

  Andrew Tulloch joins Anthropic, Forbes Australia. The $1.5bn offer was reported by The Wall Street Journal and disputed by Meta.

• Muse

  Meta admits Muse's likeness to OpenClaw isn't a coincidence, TechCrunch.

• Workplace

  Meta pulls the plug on Workplace, People Matters.

• Enterprise platform

  Why MongoDB sank today, Motley Fool.

• Smart glasses

  AI glasses shipments rise 263%, ChannelNews, citing Counterpoint Research.

• Voting control

  How Zuckerberg keeps his job, Public Citizen, 2022.