Why Google Might Be the Best Company in History

Alphabet at a Glance

Capital expenditure nearly tripled in three years — mostly AI infrastructure. The core engine is large and growing fast enough to fund speculative bets without threatening the company's health. That capacity is the precondition for everything that follows.

Source: Alphabet Inc. Form 10-K, FY2025 (filed Feb 4, 2026).

The Moonshot Portfolio: "Other Bets"

Alphabet reports every business outside Google Services and Google Cloud under one segment called Other Bets. The name sounds boring. What's inside it isn't — some of the company's most ambitious, longest-shot projects.

Source: Alphabet Inc. Form 10-K, FY2025.

The honest reading: this is not a hidden profit center. Revenue is flat, the loss is widening — driven largely by a $2.1B stock-compensation charge tied to Waymo's rising internal valuation — and Other Bets is well under 0.5% of Alphabet's total revenue. Alphabet isn't quietly making money here; it is openly spending it, on the belief some of these bets will eventually matter far more than today's numbers suggest.

Google is not a conventional company. We do not intend to become one.Founders' 1998 letter to shareholders — still cited in Alphabet's 2025 filings

Isomorphic Labs

Founded in 2021 as a spin-out from Google DeepMind, with a narrow mission: apply DeepMind's AI breakthroughs to the commercial work of designing new medicines. Led by Demis Hassabis as CEO alongside President Max Jaderberg, headquartered in London with offices in Cambridge, MA and Lausanne. It operates as its own company within the Alphabet ecosystem — outside investors, its own governance — not an internal Google division.

From AlphaFold to a drug design engine

The starting point is AlphaFold, DeepMind's system that solved a fifty-year grand challenge in biology: predicting the 3D shape a protein folds into from its genetic sequence alone. That shape determines a protein's function — central to understanding disease and designing drugs. DeepMind made its predictions for nearly all 200 million proteins known to science freely available.

Isomorphic has since built its own Drug Design Engine ("IsoDDE"), going meaningfully beyond AlphaFold 3: it more than doubles AlphaFold 3's accuracy on difficult, unfamiliar protein-drug structures, models subtle effects like a protein reshaping around a drug molecule or a hidden binding pocket opening only when needed, performs roughly twice as well at predicting antibody structures, and estimates binding strength without the expensive crystal-structure data physics-based methods traditionally require. In one internal test, it correctly identified a real, recently discovered drug binding site from raw sequence alone — no prior knowledge of where to look.

In plain terms: Isomorphic isn't just reusing AlphaFold. It built a purpose-made engine for the actual work of drug design — steps that used to require slow, expensive lab work and now increasingly happen first in software.

Commercial validation

This technology is now tested against real pharmaceutical partners, at real commercial terms — combined potential payments across all three exceed $2 billion, striking given Isomorphic has not yet brought a single drug of its own to market.

Funding

Google DeepMind: History and Structure

Isomorphic cannot be understood without its parent — the research lab whose breakthroughs made it possible, and arguably the more improbable story.

Games, audio, protein biology — that range is itself part of the story: a lab organized around a general method, not a single application. Which is exactly what made spinning off a dedicated drug-discovery company a natural next step, not a departure.

Demis Hassabis

The person both organizations have in common — and his path to running them is unusual enough to tell on its own.

He was a chess prodigy, a master by 13, and captained his college chess team while studying computer science at Cambridge. At 17, still a student, he designed the video game Theme Park, a commercial success. After graduating he worked as lead AI programmer at Lionhead Studios before founding his own studio, Elixir Studios, producing several more award-winning, AI-driven games.

In 2005, at the height of that career, he walked away entirely. A long-standing fascination with the human brain led him back to academia — a PhD in cognitive neuroscience at UCL, focused on memory and imagination, then postdoctoral research at Harvard and MIT. He's since described this as the moment he stopped thinking like an engineer and started thinking like a scientist: form a hypothesis, build an experiment to test it.

That combination — a scientist's curiosity, an engineer's instinct to actually build something — led him back to AI, and to co-founding DeepMind. The lab's original pitch to investors was disarmingly direct:

Step one, solve intelligence. Step two, use it to solve everything else.Demis Hassabis, on DeepMind's founding pitch

It read as an outlandish claim at the time. Fifteen years, a world-champion-beating Go program, and a Nobel Prize in Chemistry later, it reads as a plan that was simply followed, one deliberate step at a time — with Isomorphic Labs as its most literal, most commercial expression yet.

Hassabis on Research and AGI

Hassabis has been unusually explicit about how he and his teams choose which problems to work on — a framework that explains why DeepMind's bets look so different from a typical corporate research lab's. In a 2026 published conversation, he laid out three criteria:

Beyond that, DeepMind looks for "root node" problems — ones whose solution unlocks an entire new field. Protein folding was exactly that: solving it opened up drug discovery and disease research broadly, which is precisely what Isomorphic now exists to exploit commercially. His ambitions go further — a twenty-year dream of a working simulation of an entire living cell, accurate enough that most drug discovery experimentation happens in software, wet labs reserved only for final validation.

The Bet

The bear case is straightforward: Other Bets loses money, the losses are growing, and even Isomorphic's clearest wins exist alongside the fact it hasn't brought a single drug to market yet. Read uncharitably, this is an advertising company subsidizing expensive side projects that may never pay off.

But the more interesting fact is how rare this experiment is. Very few public companies could simultaneously fund a Nobel Prize–winning research lab, a serious drug-discovery company with three blue-chip pharma partners and outside sovereign investors, and a self-driving car business now raising billions on its own commercial merit — all while losing money on most of it, year after year, without shareholders forcing a retreat. Alphabet can, because its core advertising business is that good.

That, more than Search, is the real argument for "best company in history." Not that Alphabet makes the most money — but that it may be the best-funded, longest-patience science experiment a public company has ever run in the open.

The market's current price, and most analysts' targets, are effectively a bet that Other Bets repeats the trick Google Cloud already pulled off — just as fast. This paper argues that bet is reasonable, given the company's history and financial capacity — not that it's already been won.