Invest1 publisher3 min readPublished
$3.3bn of disclosed cash minted $26bn of new unicorn value in August
Crunchbase counted 29 additions to its unicorn board in August, worth about $63bn, and across the nine AI, software and semiconductor entrants with disclosed terms, every dollar of new cash set roughly eight dollars of paper.
The Investor · Invest desk

What happened
- Twenty-nine companies joined the Crunchbase Unicorn Board in August, adding around $63 billion of value to the board.
- The highest-valued new entrant was XPeng Robotics, a China-based humanoid robotics business valued at more than $6.3 billion.
- Semiconductors was the second-largest sector with five new unicorns, while robotics and financial services added three each and data centers, security and energy two each.
- The United States accounted for 16 of the 29, China for four, and Nigeria and Indonesia each added one, their first new unicorn of the year.
- Nine companies left the board, three by going public with Unitree Robotics the most notable, and six by acquisition, including Hugging Face, OpenRouter and Airtable.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- contradiction An allocator reading the sector count as an allocation signal gets the shelf wrong: the two semiconductor names with disclosed marks carry $5.5bn and $5bn, and only River AI matches them among the seven AI and software entrants.
- constraint A mark set with four cents of cash per dollar of valuation, as at HubX, gives the next investor almost nothing to price against, which pushes the first real price discovery out to the follow-on round.
- exposure HappyRobot's $1.2bn post-money against 150 disclosed enterprise customers puts $8m of implied value behind each account, making account churn rather than model quality the thing that moves its next mark.
- decision Underwriting a $5bn valuation on a one-year-old company means choosing a founder's track record over a revenue line, and that choice cannot be revisited later without a markdown that the fund has to explain.
Nine of the 29 entrants arrive with both a round size and a valuation attached, and the ratio between them is the number worth carrying around: roughly $3.29bn of disclosed cash against $25.95bn of new marks, which is 12.7 cents of cash per dollar of paper, or about 7.9 dollars of valuation per dollar in [4]. That leaves out DEEPX, the eight-year-old South Korean edge-inference chip designer, which took about $29m as the first tranche of a Series D from existing investors and carries no stated mark [17]. River AI is the thick end of the distribution, $1.1bn from a round led by General Catalyst and AMP PBC against a valuation of about $5bn, 22 cents on the dollar [5][6]; HubX is the thin end, a $50m Series A led by Point72 into a four-year-old Turkish consumer app developer at $1.25bn, four cents [16][5].
Three companies took $2.2bn of that $3.29bn, about 67% of it [12]: River AI, a model training and deployment platform less than a year old whose named asset is xAI co-founder Igor Babuschkin [5]; Lumilens, a two-year-old photonic interconnect maker already deployed inside data centres, which raised $700m at $5.5bn [4]; and Source Foundry, one year old, founded by Stanford researchers, which took $400m at $5bn from a round led by the hedge fund Situational Awareness [6]. The largest cheque written into anything resembling application software was Wispr Flow's $280m Series B, led again by Menlo Ventures, at a $2bn valuation for a five-year-old voice-writing and transcription business [13][14].
The pedigree trade sits alongside the hardware: Pragmatik Labs, under a year old, took a $220m seed at $2bn led by Gaorong Capital and HongShan, its founder Junyang Lin having left Alibaba this year after working on Qwen [12]. More than a third of the month's entrants were under three years old [2], so the underwriting is on people and roadmaps rather than accounts. Where an operating number does exist it is thin against the price: HappyRobot put 150 enterprise customers behind a $1.2bn post-money [15], and CodeRabbit, at $1.5bn on a $143m round, is spending more than $10m of it keeping its tools free for open-source projects, about 7% of the raise on distribution [14][10].
My view, with the counter in the same paragraph: the disclosed cash says training infrastructure and the silicon under it, not applications, is where late-stage allocators are concentrating, or rather where they will pay for optionality on the data centre buildout, since $700m into a two-year-old is a bet that a buyer exists later [4]. The counter is that $63bn added to a board is not $63bn transacted, and 12.7 cents per dollar is exactly what you would see if prices were being set by the smallest cheque that clears a preferred round with a large headline number on it [4]. The acquisition column is the test: if August's six buyers print prices at or above the last private marks, the entry prices were cheap; if they print nothing, the $63bn measures what was announced rather than what anyone paid [10].
What to watch
- Whether August's six acquirers disclose prices, since Hugging Face's and OpenRouter's terms would show if entry marks held.
- DEEPX's Series D beyond the roughly $29m first tranche, and whether a valuation is stated at all this time.
- River AI's next round, because a sub-one-year-old at about $5bn needs a second, unrelated buyer to make the first mark real.