Two green steel plants absorbed 26% of this year's proptech venture funding

Start with the average round, because that is where the fewer-but-bigger story lives. About $8.7bn across 794 deals ^[1][5] is roughly $11.0m a round ^[1], against $8.5m for last year's $12.3bn across 1,446 ^[3][7][2]. Now take out two steel plants. Stockholm's Stegra took about $1.6bn in a June private equity deal that made Wallenberg Investments its majority owner ^[11], Madrid's Hydnum Steel took $695m in August at a $3.1bn valuation ^[12], and the $6.4bn left across the other 792 deals averages $8.1m ^[4], which is below 2025 rather than above it.

On a run rate, eight months of 794 deals annualizes to about 1,191, some 18% fewer than last year, while $8.7bn annualizes to $13.05bn, about 6% more ^[5][6]. That is the "roughly match or slightly exceed" Crunchbase describes ^[4]. Set both against 2019 and they halve: 54% of the dollars, just under half the deal count ^[7].

The part of this that is genuinely AI applied to buildings is smaller than the framing carries. Bedrock Robotics' $270m Series B at $1.75bn ^[14] and Nesto's $216m Series E at $1.47bn ^[15] add to $486m, or 5.6% of the year's total ^[9], and Bedrock has put 77% of its lifetime funding into that single round ^[12]. This is probably wrong, but the more interesting reading is that proptech has stopped being an allocation and become a residual bucket: the top of the table is two industrial decarbonization assets plus a 14-year-old hospitality software company that raised $300m at $2.5bn in January ^[13], which groups them by adjacency rather than by shared thesis. Five deals took 35.4% of the year's dollars ^[8], and four of the five were outside the United States ^[10].

The mechanism is straightforward. At 6% to 7% money ^[8], against 15-year mortgages that touched 2.5% during the pandemic ^[9], anything whose revenue is a function of transaction volume gets underwritten differently, and Crunchbase's read is that the strain lands on generic real estate software and on later-stage companies without exceptional growth ^[18].

Exits point the same way. EquipmentShare priced 30.5 million shares at $24.50 in January for about $747m of primary proceeds, and the $859m total implies roughly $112m, or 13% of the offering, went to existing holders ^[16][10]. That was the only significant IPO in the space all year, while M&A is described as robust ^[16][17], so the marks are coming from strategic buyers rather than from a public bid.

What would break the thesis: if September through December lift the ex-steel average above $8.5m, then the round inflation is real and my taxonomy complaint is noise. And if one of this year's acquisitions prints a disclosed price above the target's last private round, the repricing is narrower than the deal count suggests.