The pricing that came in under its own ask
On 15 July 2026 Csquare priced its New York listing at $21.00 a share. It had spent the roadshow marketing the deal at $23 to $27. Fifty million shares went out a dollar below the bottom of its own range, raising $1.05 billion gross. The stock opened at $20.90 and last traded at $20.37, under the price the buyers had just paid. The ticker is CSQR and the platform is controlled by Brookfield.
The filings are unusually blunt about where the money went. The pricing term sheet Csquare filed with the SEC puts $921.0 million of net proceeds against the revolver, a promissory note and the Series 2024-1 VFNs, repaying all three in full. Essentially the entire raise goes to the balance sheet rather than to growth. Brookfield retains approximately 69.0% of the voting power. At the debut level the company carried a market value of roughly $3.24 billion.
The portfolio behind it is not small. Csquare runs 64 sites across 21 metros with approximately 389 MW of sellable power and more than 1,700 customers, drawn from the prospectus it filed with the SEC and confirmed in independent reporting by The Globe and Mail. This is an established North American colocation platform, and public investors still declined to pay up for it. No source gives a reason for the low pricing and we are not going to supply one.
The same 48 hours, at the other end of the trade
The demand story looked the exact opposite that week. ASML reported its second quarter on 15 July 2026, the same day Csquare priced. Net sales rose from EUR 8,767m to EUR 9,326m quarter on quarter, gross margin rose from 53.0% to 54.0%, and net income came in at EUR 2,918m. Both sales and gross margin landed above the company's own guidance. ASML guided the third quarter to EUR 11.0-12.0bn and raised full-year 2026 guidance to EUR 43-45bn.
ASML is adding capacity to match. The company is putting roughly 30% more into its 2027 low-NA EUV capacity and roughly 30% more into DUV immersion capacity. Chief executive Christophe Fouquet put the customer behaviour on the record: "Our customers, in turn, continue to accelerate their capacity expansion plans." One absence deserves naming honestly. The release carried no net bookings figure at all, with qualitative order language in its place.
TSMC reported the next day, 16 July 2026, and beat its own guidance on every line. Revenue was US$40.20bn, up 36.0% year on year, at a gross margin of 67.7%. Advanced nodes at 7nm and below made up 77% of wafer revenue. Second-quarter capital expenditure was NT$496.00bn against NT$297.22bn in the second quarter of 2025, a 67% increase year on year. Those are the filed numbers, and they are the only ones we will use here.
Our read: one label, two asset classes
Read the two events together and a split appears that neither shows on its own. Say plainly what this is. It is Servola's analysis and our synthesis of separate events. No source draws this connection. ASML and TSMC say nothing about Csquare, Csquare's pricing says nothing about them, and there is no causal link between the results and the IPO. Our claim is narrower and more useful than a cause. It is about what the two, read side by side, reveal about how the market now segments the words "data centre".
The split, in one sentence. Within roughly 48 hours public investors bid up the machinery of AI-scale compute and marked down a 389 MW portfolio of ordinary enterprise colocation. Csquare's focus is deliberately sub-5MW enterprise colocation rather than hyperscale AI capacity, which is exactly what makes the contrast meaningful instead of a coincidence of the calendar. Two very different businesses have been wearing the same two words. This week the tape stopped treating them as the same thing.
Where that lands on your invoice
For an operator this arrives at renewal. The phrase "AI demand" has been doing work in colocation pricing conversations that has very little to do with what most European businesses actually buy. Most of you are buying a handful of racks, some power, some cross-connects and some remote hands. That is sub-5MW enterprise colocation. It is the tier that just got repriced, and it was repriced downward.
The evidence you now hold is public and dated. When a vendor tells you the market has moved and your renewal has to move with it, the market in that same week declined to pay a premium for 389 MW of exactly the capacity you are buying. Investors looked at a 64-site platform with more than 1,700 customers and would not clear $23. Your procurement team can point at a printed price on the NYSE tape from 15 July 2026.
European owners buy from a market with the same two-tier shape. AI-scale capacity is scarce and priced like it. Ordinary enterprise racks are an increasingly ordinary commodity and get bought and sold as one. Csquare is a North American platform and none of its numbers are a European statistic, so treat them as what they are. The transferable part is generic and true anyway: know which tier you are actually buying before you accept a renewal priced as though you were buying the other one.
What to say at the renewal table
Ask the vendor to state, in writing, which tier your capacity sits in: sub-5MW enterprise colocation, or AI-scale. That single answer decides whether an AI-premium increase has any evidence behind it at all. The distinction is rarely offered up front, so ask for it in the terms. If your contract is the former, an AI-premium increase is a pricing decision, not a market fact.
The line to use. Public investors declined to pay a premium for a 389 MW portfolio of the same kind of capacity we are renewing, on 15 July 2026, on the NYSE. That entitles nobody to a discount and it is no kind of threat. It moves the burden of proof onto the vendor, which is where it belongs at a renewal.
One boundary on all of this. Nothing here is a view on Csquare's stock, on Brookfield, or on whether any of it is worth owning. Servola does not give investment advice and this is not it. The listing matters for a single reason. It is a public, dated and unusually clean read on what ordinary enterprise colocation is worth to people who have to pay real money for it.
Read next: AI Labs Are Racing to Go Public Because Private Capital Can No Longer Carry the Compute Bill | The Nvidia Tax Just Found a Second Source
