Method note: Self-initiated, public information only, unrelated to any client work. Every claim carries a Fact vs. Inference Ladder label: [F] verified fact, [RC] reported claim, [CI] corroborated inference, [SI] single-source inference, [SP] speculation. Figures current as of early July 2026.
Market sizing has a dirty secret: most "48-hour" sizings are 30 minutes of googling and 47.5 hours of formatting. The analyst grabs the first market report a search engine serves, wraps the headline number in a chart, and ships it. This walkthrough shows what the 48 hours are actually for, using a market that is genuinely exploding: liquid cooling for data centers.
The client question, framed as a real one would be: How big is the data center liquid cooling market in 2026, and how fast is it growing? The honest answer takes two days precisely because the fast answer fails the first test any serious number must pass: knowing where it came from.
Hours 0 to 2: Define the market before you size it
Method callout: the definition is the sizing. Every downstream number inherits your scope decisions. For liquid cooling: hardware only (cold plates, CDUs, manifolds, piping), or hardware plus installation and services? Manufacturer revenue or end-user installed spend? Direct-to-chip and immersion only, or rear-door heat exchangers too?
We scope it as: global, calendar 2026, direct liquid cooling for data centers, measured two ways, manufacturer revenue and end-user spend. Holding both measures is deliberate; they answer different client questions, and conflating them is the most common sizing error in circulation.
Hours 2 to 6: The discard phase
Search the market. The first page returns a wall of near-identical reports from commodity research houses, each with a precise-looking headline figure, a CAGR to one decimal place, and a purchase link. Put them side by side and the tell appears immediately: their estimates for the same market in the same year differ by more than a factor of two, with no visible methodology to explain why.
Method callout: a number you cannot trace is not evidence, it is decoration. Run every source through three questions before a single figure enters your model. One, can you see the methodology, or at least reconstruct what was counted? Two, does the publisher have specialist depth in this domain, or does it publish the same template across four hundred markets? Three, does the source have skin in the game, analysts whose reputation rides on this specific coverage area? The commodity reports fail all three. They are not malicious; they are optimized to rank in search, not to be right. Two-times disagreement with no explanation is not a spread to average, it is a signal to discard the whole tier.
What survives the filter is a short list: Dell'Oro Group, a specialist infrastructure research house with named analysts covering this exact market; BloombergNEF for capacity data; company primary disclosures, earnings calls, capex guidance, and product specifications; and practitioner cost benchmarks, used with explicit flags. Four hours spent throwing away the obvious answer is the highest-leverage block in the entire sizing.
Hours 6 to 12: Build the trusted factual base
The specialist read. [RC] Dell'Oro expects the liquid cooling market to roughly double in 2025, reaching close to $3 billion in manufacturer revenue, and to scale toward approximately $7 billion by 2029. This is a reported claim, not a fact, but it is a claim from a house with named analysts, a stated scope (manufacturer revenue), and a franchise staked on this coverage. One credible number with a visible definition beats six incompatible numbers without one.
The physical driver. [F] NVIDIA's GB200 NVL72 rack draws roughly 120 to 132 kW under load; air cooling is not viable at that density, and direct liquid cooling is required. [F] Microsoft confirmed in December 2025 that all future Maia accelerator deployments will use a liquid-default architecture. [RC] Dell'Oro projects leading-edge GPU thermal design power exceeding 4,000 W by 2029. The direction of thermal physics is not in dispute.
The capacity wave. [F] Over 23 gigawatts of data center IT capacity was under construction globally at the end of September 2025, per BloombergNEF, roughly three quarters of it in the US. [F] The four largest hyperscalers guided to approximately $630 billion in combined 2026 capex on their earnings calls, up from $388 billion in 2025. These come from filings and calls, checkable by anyone.
Hours 12 to 24: The bottom-up build
Method callout: never ship a top-down number you cannot rebuild from physical units. Three steps.
Step 1, new capacity. [CI] From the 23 GW under construction and typical 12-to-24-month build timelines, we estimate 12 to 18 GW of new IT capacity energized during calendar 2026. Inference from two facts; labeled.
Step 2, liquid share. [CI] With flagship AI racks liquid-mandatory and at least one hyperscaler liquid-default, we estimate 35 to 50 percent of 2026 additions ship liquid cooled. The mandatory-cooling facts beneath this are solid; the percentage is our judgment and is the model's soft spot.
Step 3, unit cost. [SI] Practitioner deployment guides benchmark liquid cooling at $1,000 to $2,000 per kW of cooling capacity for equipment, with installed direct-to-chip costs quoted higher at the facility level. Single-source-grade benchmarks, flagged, not audited.
The arithmetic. 12 to 18 GW, times 35 to 50 percent, gives roughly 4 to 9 GW of liquid-cooled load added in 2026. At the equipment benchmark that is $4 to 9 billion in the central scenarios before retrofits, which the model deliberately ignores. The range is wide, and it should be; a bottom-up model that produces a suspiciously tight range is usually hiding its assumptions rather than lacking them.
Hours 24 to 48: Triangulate, stress, write
Now check the two independent methods against each other. Dell'Oro's trajectory, roughly $3 billion in 2025 and doubling-era growth, implies manufacturer revenue in the $4 to 5 billion range for 2026. The bottom-up build, run at equipment-level costs and conservative liquid share, lands its lower half in the same corridor; run at installed costs, it points to total end-user spend meaningfully above that, plausibly $6 to 9 billion. [CI] The reconciled finding: roughly $4 to 5 billion in 2026 manufacturer revenue, with total end-user installed spend materially higher, and a credible path to Dell'Oro's ~$7 billion manufacturer figure by 2029. Two methods, independently sourced, failing to disagree. Honest caveat: this triangle has two legs, not three; a supply-side check against vendor segment revenue is the named gap, and closing it is the first task of any follow-on week.
[SP] The labeled speculation for the watch list: NVIDIA's roadmap points toward 600 kW to 1 MW racks in the Rubin era. If that ships on schedule, cooling stops being a procurement line and becomes co-designed infrastructure, and every current forecast is too low. Watch indicator: two-phase cooling moving from pilots to volume purchase orders.
The deliverable, built as a Pyramid: Data, the trusted base above, every claim labeled. Insight, one sentence: the market is $4 to 5 billion in 2026 manufacturer revenue, growing toward a doubling by 2029, and the first page of search results was off by up to 2x because it measures different things without saying so. Implication and Action, written for whoever commissioned it, with a 90-day re-test against Dell'Oro's next update and hyperscaler capex revisions.
Total time: 48 hours. Time spent finding numbers: about four. Time spent deciding which numbers deserved to exist in the model: about eight. That second block is the craft, and it is the one the 30-minute version skips.