Backtesting ArenaBacktesting Arena
← Back to blog

AI in the Drug-Discovery Race: Shovels, Bets, Giants

Asking which company is well-positioned in the AI drug-discovery race is really three questions at once. We sort the field soberly into shovels, bets, and giants — and show which layer earns regardless of any single trial outcome.

Backtesting Arena·July 1, 2026·5 min read·51 views
AI in the Drug-Discovery Race: Shovels, Bets, Giants

Artificial intelligence is transforming drug discovery — there's little doubt about that. But asking "which company is well-positioned?" is really asking three questions at once. In this field, "well-positioned" means three different things, and the three layers carry fundamentally different return and risk profiles. Blur them together and you can easily buy hope at the price of infrastructure — or the reverse.

Let's sort the field soberly into three layers: shovels, bets, giants.

Layer 1: The shovels — infrastructure earns regardless

In every gold rush, the shovel sellers earn reliably, long before anyone knows who strikes gold. In AI drug discovery, those are the providers of compute, software, and data.

First, the compute layer: specialised hardware and platforms on which the whole industry trains its models. One leading chip company, for instance, has struck a billion-dollar collaboration with a pharma giant (as of 2026) — and the same infrastructure also powers the supercomputers of several AI biotechs. Whoever sells the shovels gets paid no matter which molecule wins in the end.

A layer above sits software: physics-based simulation platforms that researchers license across the industry. The decisive point is the nature of the revenue — recurring licence income doesn't hinge on a single trial succeeding, but on a growing software business. Such a provider often still posts losses, yet its revenue base is structurally steadier than that of a pure drug developer.

And finally the data layer: platforms that pool clinical and molecular patient data and license it to pharmaceutical firms — for example through multi-million-dollar agreements to jointly build AI models. Here too, the seller of the data earns independently of how any single trial turns out.

Illustrative examples of this layer (not a recommendation): a large chip and AI-platform provider (compute), a physics-based simulation-software provider (software), a provider of multimodal clinical-molecular data (data).

The common trait of the shovels: they have nothing clinical to prove. They sell to all the bettors at once.

Layer 2: The bets — pure-plays need Phase III proof

The AI-native "techbio" firms are what most people think of first: platforms whose entire business model is to find better drug candidates faster, using algorithms, lab robotics, and vast datasets. One leading platform, for instance, links lab robotics, a petabyte-scale dataset, and AI models into a closed loop.

This is where the greatest upside sits — and the greatest risk. The honest stocktake (as of June 2026): with few exceptions, the publicly traded AI drug-discovery pure-plays have posted losses since their IPOs, several candidates marketed as "AI-designed" have missed their clinical endpoints, and individual stocks have fallen roughly 90 % from their highs. A cash runway "into sometime in 2027" is a normal sentence in this segment — not a reassuring one.

The methodical crux: a beautiful platform is not yet a medicine. What these firms must prove is not algorithmic elegance but a Phase III result — an AI-found drug that actually works in the decisive, large clinical trial. That's exactly why 2026 is a "show-me-Phase-III" year: the industry's real bottleneck is not compute, but data quality and clinical proof.

That doesn't make the bets worthless — it makes them options: a high payout if a platform delivers, a total-loss risk if it doesn't.

Layer 3: The giants — capital and data as the moat

The third layer is the established pharma and biotech groups that deploy AI at scale without betting their existence on it. Their advantage is unspectacular but robust: balance sheet and proprietary data.

A group with a running cash engine — say in metabolic and weight-loss drugs — and a billion-dollar AI collaboration can use AI to make its existing pipeline more efficient, and survives even if a single programme fails. Other giants pair AI with genetic databases built over decades, or with external data partnerships.

The difference from the second layer is fundamental: for the giants, AI is an efficiency lever, not an existential promise. They don't have to prove that AI alone carries them — only that it improves things incrementally. That's a far lower bar.

The methodical core

LayerWhat it isWhere the return comes fromMain riskWhat it must prove
ShovelsCompute, software, datarecurring, industry-wide revenuevaluation, macro cyclenothing clinical — sells to all
BetsAI-native drug platformssingle trial wins, licencesclinical failure, cash burnone Phase III result
GiantsBig pharma deploying AIexisting products plus efficiencypatent cliffs, inertiaonly that AI helps at the margin

The transferable lesson — and the reason this is more than a stock list: nobody can reliably predict which drug candidate will clear every clinical hurdle. Accept that, and you're left with two methodically clean paths. Either position yourself in the durable layer of the value chain — the shovels and giants that earn regardless of any single trial outcome. Or deliberately buy the basket rather than the single name: a broadly diversified sector index that contains the unknown winner anyway.

The mistake the methodically honest lens avoids: paying an "AI premium" for a clinically unproven platform, as if hope were already proof. Infrastructure always earns. Bets have to deliver first. Giants can afford to wait.


This is not investment advice. The companies referenced are illustrative examples of each layer, not a recommendation of individual securities. Figures are as of June 2026 and subject to change; this sector is more volatile than most.

FAQ

What does "well-positioned" mean in the AI drug-discovery sector? Three different things, depending on the layer: infrastructure providers ("shovels"), AI-native drug platforms ("bets"), and established pharma groups deploying AI ("giants"). The three carry fundamentally different return and risk profiles and shouldn't be lumped together.

Why are infrastructure providers seen as more robust than the pure-plays? Because they sell compute, software, or data to the whole industry, so they earn independently of which single drug ultimately passes its trials. They have nothing clinical to prove.

What is a "Phase III result" and why does it matter? Phase III is the large, decisive clinical trial before approval. For the AI-native layer, the real proof is not an elegant platform but an AI-found drug that actually works in Phase III. Until then, the business model remains unproven.

How can you invest if you don't know the single winner? Two methodically clean paths: position in the durable layer (shovels, giants) that earns regardless of trial outcomes — or deliberately buy a broadly diversified sector basket that contains the unknown winner anyway, rather than betting on one name.

Try it yourself

Run the backtest with your own parameters and time ranges.

Run backtest →

More on this topic

Market Analysis

Bitcoin and stocks: how tight is the link really?

Backtesting Arenatradingstrategies.work

On Nov 5 the Nasdaq-100 fell 6.4 % and bitcoin 5.6 %. Over the year the Nasdaq is up 12 % and bitcoin down 16 %. Daily data back to 2015 shows why both are true: the link to tech stocks is six years old, comes and goes, and hits harder on the way down. The trend bitcoin still makes on its own.

MacroBitcoinRisk management
Nov 13, 20261 min
Market Analysis

Bitcoin's 2026 Fall Was One of Its Mildest – Yet It Sent the Price Further Back in Time Than Ever

Backtesting Arenatradingstrategies.work

Measured by drawdown, the 2026 bear market was the mildest of bitcoin's big cycles: −53 % against −77 % to −93 % before. Measured in time, it was the worst: the low matched a price from five and a quarter years earlier, further back than any low since 2010. And against its own history, bitcoin was not cheap at any low since 2017. Three numbers for every bear market, and what each tells you before you call bitcoin a bargain.

DrawdownBitcoinMethodology+1
Oct 15, 20261 min
Market Analysis

How Often Does Bitcoin Crash 30% – and Is It Getting Rarer?

Backtesting Arenatradingstrategies.work

Since 2011 Bitcoin has fallen 30 percent within a month 19 times. How often that happens today, how it fits with falling volatility, how far the drop usually went, and what that means for the size of your position.

DrawdownRisk managementBitcoin+1
Oct 5, 20261 min
Market Analysis

Stocks in Wartime: Why the Recovery Statistics Only Contain the Winners

Backtesting Arenatradingstrategies.work

"Stocks have recovered from every war" is true — for one country. After the April 1940 high the Dow needed four years and nine months to get back to even, in 1914 the exchange was closed for four and a half months, and St Petersburg, Vienna, Tokyo and Shanghai are missing from the statistics because their exchanges ceased to exist. Across 39 markets, a diversified investor is down in real terms after 30 years in 12 % of cases.

MethodologyBacktestingDrawdown+1
Sep 21, 20261 min
📬

Don't miss new blog posts

One short email per new post — strategies, backtests, market analysis. No spam, unsubscribe with one click anytime.

By subscribing you accept our privacy policy. We use Resend for delivery. Double opt-in confirmation required.

Comments (0)

Join free to post comments.

Sign up →

No comments yet. Be the first!