What jobs would a pharma AI supercomputer create, and what do they pay?

A full build is costed at 35 specialists on 150,000 to 220,000 euros. Phase 1 in 2026 starts smaller, with 15 to 20 hires. Two skill sets get named: HPC architecture at 120,000 to 220,000 euros, and computational chemistry combined with AI at 130,000 to 180,000 euros. Germany's entire HPC talent pool sits at roughly 200 people, while tech companies offer 250,000 and up. Moving from wet lab work into computational roles brings a 40 to 60 percent salary jump. By 2026 German pharma will need more than 100 of these specialists. Hardware is the straightforward part. Filling the roles is where it gets hard.

Is it cheaper for a pharma company to own a supercomputer or rent cloud compute?

It depends how hard you run it. For 24/7 workloads, on-premise comes out five to seven times cheaper than cloud. Bayer's current cloud bill runs around 12 to 15 million euros a year and keeps climbing, which puts break-even on an owned machine at three to four years, with pure saving after that. Manufacturing adds a second lever. A 1 to 2 percent yield improvement on biologics is worth hundreds of millions, enough to pay for the machine on its own. Against that sits the sticker. Eli Lilly spent 110 million euros on hardware, 300 to 400 million once ten years of running costs are added.

Why is Bayer looking at this now rather than two years ago?

The balance sheet moved. Debt stands at 32.7 billion euros, down 3.2 billion year over year. Q3 2025 EBITDA came in at 1.5 billion euros, up 21 percent. Nubeqa and Kerendia grew 63 percent against the prior year, which absorbs the Xarelto patent cliff, and Lynkuet launched in November 2025. At the same time the pipeline has filled with cell therapy, gene therapy and precision oncology in late-stage trials, all of it computationally hungry. My bet on timing: plan in Q4 2025, decide in Q1 2026, deploy Phase 1 at the end of 2026 and scale fast through 2027.

Should Bayer Build Its Own Supercomputer?

Should pharma companies build their own AI supercomputers, or rent them?

Bayer, AstraZeneca, and several Top 20 pharma firms are evaluating in-house AI compute. The economics rarely justify it: the workload is bursty, the talent for utilisation is scarcer than the hardware, and the regulatory risk of in-house model training around clinical data is meaningfully higher. For most pharma functions, renting compute (Azure, AWS, Google Cloud life-sciences SKUs) and investing in AI-fluent humans beats a Ferrari sitting in a garage 90% of the year.

Am I answering a question that nobody has asked? I don't know but I wouldn't be surprised if my boss Bill Anderson asked this question long ago...

Eli Lilly just dropped €110 million on a supercomputer with 1,016 NVIDIA Blackwell Ultra GPUs. Add 10 years of running costs, and you're looking at €300-370 million total.

Meanwhile, Bayer's been strategically building cloud-first AI (AWS, Google Cloud, PRINCE platform) while paying down debt. Plot twist: Q3 2025 results look really good. Debt down €3.2B year-over-year, blockbusters launching, margins expanding.

So I asked: Should Bayer upgrade from the company car to our own Ferrari?

(Btw - we have far fewer company cars at Bayer than companies of our size usually have. For business travel in Germany we are obliged to travel by train since Deutsche Bahn is using 100% green energy to power its trains.)

Spoiler: The finances say maybe yes. And finding those 35 brilliant minds to run it? Challenge accepted. 😎

The Numbers (Or: Why This Ferrari Idea Isn't Crazy)

Lilly's Ferrari: €110M sticker price, €300-400M over 10 years. Discovers molecules in 5 minutes vs. a year. One of pharma's largest AI supercomputers.

Bayer's Momentum (We're Just Getting Started):

Debt: €32.7B, down €3.2B YoY – crushing it! 🎯
Q3 EBITDA: €1.5B, up 21%
Nubeqa + Kerendia: up 63% YoY, offsetting Xarelto's patent cliff
Lynkuet: Just launched November 2025
Pipeline: Cell therapy, gene therapy – computationally hungry stuff

The Reality: €300M equals 2-3 blockbuster drugs. BUT if it makes us discover molecules 3x faster and saves hundreds of millions in manufacturing? That's not an expense – that's a weapon.

When the mortgage is shrinking €3B+ annually and products are printing money? Maybe it's time to think bigger.

Why This Could Be Brilliant

Here's where the Ferrari metaphor breaks down: A real Ferrari is expensive, prestigious, and mostly about looking good going fast. A supercomputer? That's an expensive machine that can actually shortcut its way to billions in savings and competitive advantage. Here's why:

🚀 Speed = Billions

Molecular design: months → hours
Manufacturing optimization: real-time instead of quarterly
Cell therapy modeling: weeks → days
When racing blockbusters to market, speed isn't just nice – it's worth billions in first-mover advantage

💰 Economics + Manufacturing Gold

On-premise 5-7x cheaper than cloud for 24/7 workloads
Current cloud bill ~€12-15M annually (and growing)
Break-even in 3-4 years, then pure savings
Hidden gem: 1-2% biologics yield improvements = hundreds of millions in profit
The supercomputer could pay for itself in manufacturing optimization alone

🔒 Data Sovereignty

Your hardware, your data, your control
Easier GDPR/regulatory compliance
When you've got billions in IP and litigation provisions, ownership matters

🔬 Future-Proofing

Cell therapy, gene therapy, precision oncology in late-stage trials NOW
This isn't "maybe someday" tech – it's need-to-have for the science we're already doing
Pipeline loaded with computationally intensive work ready to feed a supercomputer

The Challenges (Nothing We Can't Handle)

€300M Is Real Money: Could develop 2-3 drugs instead. Counter: what if it helps us develop them faster?

GPU Refresh Cycles: 2026 GPUs need refreshing by 2029. But we'd own the advantage – trade-off worth it.

Cloud Works: AWS/Google delivering. Counter: at scale, ownership becomes cheaper.

Talent Competition: Germany's HPC pool: ~200 people. Tech offers €250K+. But the best want interesting problems. "Build pharma's largest supercomputer"? Pretty interesting.

German Regulations: Energy mandates, compliance. But Bayer's handled regulations since 1863. We're good at this.

Where to Park The Ferrari? (We've Got Great Options)

Leverkusen: HQ proximity, existing industrial infrastructure, political support (8/10) – safe and smart

Wuppertal: €1.4B R&D modernization happening anyway, perfect integration timing (7.5/10) – scientifically ideal

Berlin: Best talent pool in Germany, PRINCE platform already there, startup ecosystem (7/10) – talent magnet

Honestly? All three work. We've got the infrastructure, the political support, and the sites. The hard part isn't where – it's when. And Q3 2025 results suggest "when" might be sooner than expected.

So... Buy The Ferrari?

My Take: Timing is getting really interesting.

Phase 1 (2026): €30-40M initial, €120-150M over 10 years. Specialized HPC, hire 15-20 people, prove 3x faster OR €10M+ savings.

Phase 2 (2027-2028): Scale when Phase 1 delivers. Additional €150-220M. Total: €270-370M.

Why This Works:

+ Financial momentum: Debt down €3.2B YoY

+ Pipeline ready: Cell/gene therapy needs compute NOW

+ Manufacturing upside: Hundreds of millions in savings

+ Competitive necessity: Can't fall behind Lilly/Novo

Real Question: Not "can we afford it?" (we can). It's "can we afford NOT to?"

Prediction: Plan Q4 2025, decide Q1 2026, deploy Phase 1 end of 2026. Scale fast in 2027.

The Talent Reality

The Mission: 35 specialists at €150-220K competing against tech giants. German HPC pool: ~200 people.

Why We'll Win: Best people follow interesting problems. "Build pharma's largest AI supercomputer to save lives" beats "be Google employee #47,293." Bayer's turnaround story – blockbusters launching, betting on AI – is compelling.

Hot Skills: HPC Architecture (€120-220K), Computational Chemistry + AI (€130-180K). Wet lab → computational = 40-60% salary jump.

Evolution: 2020 = buzzword. 2025 = arms race. 2026 = German pharma needs 100+ specialists. This is the challenge I love.

Final Take

Lilly bought the Ferrari. Novo's shopping. Bayer's evaluating.

After 9 years at Bayer: we're stronger than some people think. Debt dropping €3B+ annually. Blockbusters growing 60%+. Pipeline loaded with compute-intensive science.

Hardware? Straightforward. Talent? That's where it gets fun. Our transformation story is real, our science is cutting-edge. I am sure that "Help build pharma's largest AI supercomputer" attracts exactly the minds we need.

Whether Phase 1 in 2026 (my bet) or conservative path, pharma R&D is an AI arms race. Winners combine right hardware with right minds.

That combination? That's what gets me excited to come to work.

Should Bayer build one of pharma's largest supercomputers?

Please answer in the comments!

If you're an HPC architect, computational chemist, or AI engineer who wants to work on something that matters – let's talk. 🚀

What's your take? Why does it make sense for Bayer to have a supercomputer?

Or is this nuts and just part of the AI hype?

(I heard people back in the day saying that this whole internet thingy is just a big hype and only a toy for computer nerds :P )

Please let me know in the comments what you are in favour of. Should Bayer build its AI Ferrari?

#Pharma #AI #TeamBayer #MoreThanCareer #DSO #DrugDiscovery

Ready to take control of your pharma career?

Book a free getting-to-know call. I will tell you honestly where you stand, what the market looks like for your profile, and what your next move should be.

Book a Free Getting-to-Know Call