Market sizing case example: Tundrix Cooling
Tundrix sells immersion cooling systems for data centers. Its sales team has set a target of 100 megawatts of cooling sold next year, and the CEO wants to know how much cooling new data centers will need and whether the target is realistic. The full solution is below: the structure, every exhibit, the math, a model answer and the mistakes to avoid.
- Type
- Market sizing
- Difficulty
- Medium
- Length
- About 30 minutes
- Format
- Interviewer led
The case
The interviewer reads this aloud. Try it before you read on.
Our client is Tundrix Cooling, which makes immersion cooling systems for data centers: servers sit in tanks of a special fluid that carries the heat away. Tundrix sells into new data centers being built in its home region. For next year, the sales team has set a target of selling 100 megawatts of cooling capacity, which they describe as just 5% of the cooling that new data centers will need. The CEO is not sure. She has asked us to estimate how many megawatts of cooling new data centers in the region will need next year, how much of that Tundrix can realistically sell into, and whether the 100 megawatt target holds up.
Facts the interviewer gives if you ask
- How does a rack's power relate to the cooling it needs?
- Assume every kilowatt of power a rack draws becomes a kilowatt of heat that has to be removed, so cooling demand equals the racks' power.
- Does the market include existing data centers that upgrade their cooling?
- No. Only racks installed in new data centers next year. The client looks at upgrades separately.
- Which racks need liquid cooling?
- The client's engineers say air cooling works up to about 20 kilowatts per rack. Above that, a rack needs liquid cooling.
- Is immersion the only kind of liquid cooling?
- No. The other kind pipes liquid to plates on the chips. The client estimates immersion will win 25% of new liquid-cooled capacity next year.
- How has Tundrix done so far?
- Last year Tundrix won 15% of the immersion cooling sold in the region.
- What does Tundrix earn per megawatt?
- About $400,000 of revenue per megawatt of cooling capacity installed.
- Is the power per rack changing?
- Yes. The newest AI racks already draw about 80 kilowatts, but the client's forecast uses 60 as the typical AI rack next year.
The solution, step by step
How the case runs in the interview, in 5 steps. Each step has the interviewer's question, the exhibit or data shown, and what a strong answer says.
- 1
Structure the problem
The interviewer asks: “How would you estimate the cooling that new data centers will need next year, and how much of it Tundrix can sell into?”
A strong structure
- 1New racks next year by type: storage, general cloud, AI
- 2Power and heat per rack for each type, giving megawatts of cooling demand
- 3Which racks need liquid cooling, and what share goes to immersion
- 4Tundrix's realistic share, against the 100 megawatt target
A strong answer
A sizing tree: new racks installed next year, split by type, times kilowatts of heat per rack, gives megawatts of cooling demand. Then filter to what Tundrix can serve: racks hot enough to need liquid cooling, and the share of liquid cooling that goes to immersion. Then Tundrix's share of that, compared with the 100 megawatt target. A strong candidate says the split by rack type matters because AI racks run much hotter, and flags that a percentage of total demand may be the wrong base for the target.
- 2
Read the exhibits
The interviewer asks: “Here is the client's forecast of new racks next year, and the heat each type of rack gives off. What stands out?”
Exhibit 1
New racks to be installed next year in Tundrix's region, by type
Client forecast of racks installed in new data centers next year, thousands.
New racks to be installed next year in Tundrix's region, by type Rack type New racks next year (k) Storage 40 General cloud 60 AI training and inference 20 Total 120 What to notice
- 120,000 new racks in total.
- AI racks are 20,000, only about one in six.
- General cloud is the largest group by count at 60,000.
Exhibit 2
Heat given off per rack, by type
Typical heat load per rack, kilowatts, from the client's engineering data.
Heat per rackHeat per rack, kWHeat given off per rack, by type Category Heat per rack (kW) Storage 5 General cloud 10 AI 60 What to notice
- Storage racks give off about 5 kW, cloud racks about 10 kW, AI racks about 60 kW.
- An AI rack gives off six times the heat of a cloud rack.
- Only AI racks are above the roughly 20 kW limit for air cooling.
A strong answer
120,000 new racks: 40,000 storage, 60,000 general cloud and 20,000 AI, so AI is only about one in six racks. But the heat per rack differs hugely: about 5 kilowatts for storage, 10 for cloud and about 60 for AI, six times a cloud rack. So AI racks will dominate the heat, and they are the only ones above the 20 kilowatt limit for air cooling. An average across all racks would hide this.
- 3
Do the math
The interviewer asks: “Work out the megawatts of cooling demand, the part Tundrix can sell into, and what that means for the target.”
The interviewer adds
- Heat per rack: storage 5 kilowatts, general cloud 10 kilowatts, AI 60 kilowatts.
- Air cooling works up to 20 kilowatts per rack. Immersion will win 25% of liquid-cooled capacity.
- Tundrix won 15% of immersion sales last year. Revenue is about $400,000 per megawatt.
A strong answer
Demand: storage 40,000 times 5 kilowatts is 200 megawatts, cloud 60,000 times 10 is 600, AI 20,000 times 60 is 1,200, total 2,000 megawatts. AI is 17% of racks but 60% of the heat. Only AI racks are above 20 kilowatts, so liquid-cooled demand is 1,200 megawatts, and immersion's 25% is 300 megawatts. The team's 100 megawatts is 5% of 2,000, but it is a third of the 300 megawatts Tundrix can actually sell into. At its 15% share Tundrix would sell 45 megawatts, $18 million, against a target of $40 million.
Every number, worked out19 calculations
Every number, worked out Number Worked out Result Total new racks (k) 40 + 60 + 20 120 AI share of racks (%) 20 ÷ 120 × 100 16.7 AI heat versus cloud heat per rack 60 ÷ 10 6 Storage heat (MW) 40,000 × 5 ÷ 1,000 200 Cloud heat (MW) 60,000 × 10 ÷ 1,000 600 AI heat (MW) 20,000 × 60 ÷ 1,000 1,200 Total cooling demand (MW) 200 + 600 + 1,200 2,000 AI share of heat (%) 1,200 ÷ 2,000 × 100 60 Immersion market (MW) 1,200 × 0.25 300 Target as share of total demand (%) 100 ÷ 2,000 × 100 5 Target as share of immersion market (%) 100 ÷ 300 × 100 33.3 Tundrix at 15% share (MW) 300 × 0.15 45 Base case revenue ($M) 45 × 0.4 18 Target revenue ($M) 100 × 0.4 40 Stretch at 20% share (MW) 300 × 0.2 60 Stretch revenue ($M) 60 × 0.4 24 Liquid demand not served by immersion (MW) 1,200 − 300 900 AI heat at 80 kW per rack (MW) 20,000 × 80 ÷ 1,000 1,600 Increase at 80 kW (%) (1,600 − 1,200) ÷ 1,200 × 100 33.3 - 4
Brainstorm the options
The interviewer asks: “What could make the real number bigger or smaller than your estimate?”
A strong answer
Drivers, grouped. Demand: the AI rack forecast itself (delays in new builds, supply of chips and power), rising power per AI rack (at 80 kilowatts instead of 60 the AI heat would be a third higher), more cloud racks crossing 20 kilowatts. Technology mix: immersion's 25% against chip-level liquid cooling, customers' comfort with tanks of fluid, serviceability. Tundrix's share: new customers, partnerships with data center builders or server makers, price. Scope: upgrades of existing data centers, which are outside this estimate, and other regions. A strong candidate names the two biggest swings: the AI rack count and immersion's share.
- 5
Make the recommendation
The interviewer asks: “The CEO wants to know: is the 100 megawatt target realistic?”
A strong recommendation
The 100 megawatt target is not realistic, because it is 5% of the wrong base. New racks need 2,000 megawatts of cooling, but only the 1,200 megawatts from AI racks needs liquid, and immersion's 25% of that is 300 megawatts. Hitting 100 megawatts would take a third of the immersion market, more than double Tundrix's 15% share. Set a base case of about 45 megawatts ($18 million) and a stretch of 60 megawatts ($24 million at a 20% share). To grow further, target the builders of AI data centers, consider a chip-level liquid product for the other 900 megawatts, and size the upgrade market. Risks: the AI rack forecast and immersion's share are the two biggest swings. Next steps: validate the forecast with the top builders and check immersion's win rate in recent tenders.
The model answer
A strong candidate's whole answer, as it sounds out loud. The marked phrases are explained below.
- 1Scope and sizing treeConfirm new builds next year, then racks by type, heat per rack, liquid filter, immersion share and Tundrix's share.
- 2Read the exhibitsSee AI racks are one in six by count but six times the heat of a cloud rack.
- 3Sizing math2,000 megawatts total, 1,200 from AI, 300 for immersion, 45 for Tundrix at 15%.
- 4What could move itAI rack count, power per rack, immersion's share, Tundrix's share and out-of-scope markets.
- 5RecommendationThe target is 5% of the wrong base, set 45 to 60 megawatts, grow through AI builders and new products.
Thanks. So Tundrix sells immersion cooling into new data centers in its region, the sales team has a 100 megawatt target for next year, which they call 5% of new cooling demand, and the CEO wants to know how big the demand really is, how much Tundrix can sell into, and whether the target holds up. Can I check the scope? Thanks, so new data centers only, next year, in the home region.
Here is how I'd build it. Start with the number of new racks next year, split by type. Multiply each by the heat a rack gives off to get megawatts of cooling demand. Then narrow it to what Tundrix can actually sell: racks hot enough to need liquid cooling, and the share of liquid cooling that goes to immersion. Then apply Tundrix's realistic share and compare with the target. I'd split by rack type rather than use one average, because I'd expect AI racks to run much hotter than the rest. I'm also a little wary of the 5%, because a small share of a big number can be a large share of the right number.
Looking at the forecast, there are 120,000 new racks: 40,000 storage, 60,000 general cloud and 20,000 AI. So AI is only about one in six racks. But the heat chart shows storage at about 5 kilowatts, cloud at about 10 and AI at about 60, six times a cloud rack. So I'd expect AI racks to dominate the heat even though they are the smallest group by count.
Can I confirm the exact heat figures? Thanks, so 5, 10 and 60 kilowatts. And I'm assuming every kilowatt of power becomes a kilowatt of heat to remove. Storage is 40,000 times 5 kilowatts, which is 200 megawatts. Cloud is 60,000 times 10, which is 600 megawatts. AI is 20,000 times 60, which is 1,200 megawatts. In total that is 2,000 megawatts, and AI racks, at 17% of the count, are 60% of the heat.
Now, which of that can Tundrix sell into? Thanks, so air cooling works up to about 20 kilowatts a rack. Storage and cloud racks sit well below that, so only the AI racks need liquid cooling, which gives 1,200 megawatts. Immersion is expected to win 25% of liquid-cooled capacity, so the immersion market is 300 megawatts. That is the real base for Tundrix's target. The team's 100 megawatts is 5% of 2,000, but it is a third of the 300 megawatts Tundrix can actually reach. Last year Tundrix won 15% of immersion sales, which on 300 megawatts is 45 megawatts. At $400,000 a megawatt that is $18 million, against $40 million for the target.
What could move this? On demand, the AI rack forecast is the biggest swing, since it drives 60% of the heat and all of the liquid demand. Power per AI rack is also rising. At 80 kilowatts instead of 60, AI heat would be a third higher. On the technology mix, immersion's 25% against chip-level liquid cooling matters a lot, and it depends on how comfortable operators are servicing servers in tanks of fluid. On Tundrix's share, new builders and partnerships could move it. And there are markets outside this estimate, like upgrades of existing data centers and other regions. I'd test the AI rack count and immersion's share first.
So my answer to the CEO is that the 100 megawatt target is not realistic, because it is 5% of the wrong base. New racks need 2,000 megawatts of cooling, but only the 1,200 megawatts from AI racks needs liquid, and immersion's share of that is 300 megawatts. Hitting 100 megawatts would mean winning a third of the immersion market, more than double the 15% Tundrix won last year. I'd set a base case of about 45 megawatts, which is $18 million, and a stretch of 60 megawatts, a 20% share and $24 million, if Tundrix wins new builders.
To grow beyond that, Tundrix should focus its sales team on the builders of AI data centers, consider a chip-level liquid product to reach the other 900 megawatts of liquid demand, and size the market for upgrading existing sites. The main risks are that the AI rack forecast slips or immersion loses share to chip-level cooling. Next steps are to validate the forecast with the largest builders and check immersion's win rate in recent tenders.
Why this answer works
Segment when one group behaves differently
AI racks are 17% of new racks but 60% of the heat and all of the liquid demand. An average per rack would have hidden that.
Size the addressable market, not the total
Tundrix can only sell into immersion on racks that need liquid cooling, 300 of the 2,000 megawatts. Targets should be set against that base.
A small share of the wrong number
5% of total demand sounds modest, but it is 33% of the market Tundrix can reach. Always restate a target as a share of the addressable market.
The phrases that do the work
“a small share of a big number can be a large share of the right number”
Hypothesis. Flagging the risk in the target's base before doing the math shows judgment.
“AI racks, at 17% of the count, are 60% of the heat”
Segment insight. Contrasting count and heat shows why the split by rack type matters.
“That is the real base for Tundrix's target.”
The so-what. Moves the conversation from total demand to the addressable market.
“the 100 megawatt target is not realistic, because it is 5% of the wrong base”
Answer first. Answers the CEO's question directly and says why in the same sentence.
Common mistakes on this case
- 1Using one average power per rack across all 120,000 racks, which hides that AI racks are 60% of the heat.
- 2Treating all 2,000 megawatts as Tundrix's market, when only the 300 megawatts of immersion on liquid-cooled racks is.
- 3Accepting the 5% framing of the target instead of restating it as 33% of the immersion market.
- 4Mixing units: forgetting that 20,000 racks times 60 kilowatts is 1,200,000 kilowatts, so 1,200 megawatts.
- 5Ending with a total market number and no answer on whether the target is realistic or what it should be.
Now run this case out loud
Soreno's AI interviewer gives you this case, shows each exhibit when you ask, follows up on your math and scores each part.
Questions people ask
Is Tundrix Cooling a real company?
No. Tundrix Cooling is an invented client, and every number is case data written for practice. The numbers reconcile, so you can check your math against the solution.
How long does this case take?
About 30 minutes out loud, the length of one interview case. It is rated medium.
How should I use this solution?
Try the case first. Read the prompt and the facts, write your structure and work the numbers, then compare each step with the walkthrough and the model answer.
Can I practice this market sizing case out loud?
Yes. Soreno's AI interviewer runs this case with you, shows the exhibits when you ask, follows up on your answers and scores each part.