Most Saturday mornings I read the weekly funding digests before I read anything else. It's a habit left over from twenty years of operating roles, and it survived the move to the investing side of the table. On August 15 the number at the top of one of them was $4.48 billion, and for about ninety seconds I felt good about the state of the market.
Then I read what was underneath it.
Of that $4.48 billion, roughly $4.23 billion was venture firms raising money from their own investors. Accel closed $3.55 billion across four early-stage funds, another $365 million went into Team8's third fund for enterprise AI security, K2 Global disclosed $200 million in fresh commitments, and Mirae Asset held a first close at $118 million. Capital that actually landed in a Web3 company's account that week, meaning primary rounds where a founder got money to hire and ship, came to about $17 million across three deals: River Markets at $8.5 million, Entravel at $7.5 million, Memebook at $1 million.
Seventeen million dollars, out of a $4.48 billion headline. Four-tenths of one percent.
I've written before about crypto venture funding splitting into a barbell. What I'm looking at now is narrower and more useful than that. What our industry quotes about itself sums three unrelated kinds of money, and only one of them is capital a founder can spend this quarter. Right now the bridge that matters runs from a Web3 team to an allocator whose written mandate says artificial intelligence.
What Three Weeks of Digests Actually Showed
I went back and did the same arithmetic on the two weeks either side of that one.
August 1 to 8: fourteen deals, just over $3 billion disclosed. Index Ventures raised $2 billion across three funds, fresh off its Wiz payout. South Park Commons closed a $575 million fourth fund aimed at physical AI. White Star Capital closed $250 million. Against that, the companies: Yellow Card raised $40 million to connect banks to stablecoin processing, Vangrid $9 million for smartphone-based DePIN, JPYC $6.3 million, Yooldo $1 million. Call it $56 million.
August 15 to 22: nine deals, about $1.3 billion disclosed. Two corporate transactions accounted for nearly all of it: ZeroStack took $1 billion of Memecore tokens onto its balance sheet in exchange for shares and warrants, and Ripple Prime closed $275 million of senior notes with institutional buyers. Set against those two line items, classic venture into operating projects that week totalled roughly $20 million, split between NeoSoul at $11 million, Beldex at $8 million, and twyne at $2.5 million.
Add the three weeks together and you get about $8.8 billion in headlines and something close to $93 million in primary rounds. Roughly one percent. Read the top-line number and you'd reasonably conclude the market is flush. Go one level down to the third line item and you land somewhere closer to the truth.
Three Kinds of Money, One Headline Number
The confusion isn't anyone's fault, exactly. The trackers are honest about their categories — you just have to read past the summary to see them. But once you separate the buckets, they behave nothing alike.
The first bucket is fund closes. That's limited partners committing capital to general partners: Accel, Index, Team8, K2, Mirae, South Park Commons, White Star. This money is real, and it's also a promise about the next three to five years rather than a wire this month. It's dry powder measured on the day the powder was purchased.
The second bucket is balance-sheet engineering: senior notes, token-for-equity contributions, treasury placements, and the occasional acquisition. Ripple Prime's $275 million is corporate debt raised inside a regulated wrapper, ZeroStack's billion is an accounting event, and eToro's $231 million purchase of TradeZero is corp dev. None of it funds a new team.
The third bucket is what most people think the headline means: primary venture rounds into operating companies, at seed, pre-A or Series A. Money that becomes salaries and audits and go-to-market.
A fund close is a promise made to someone else's LPs. A round is money in your account. The digests add them together, and your payroll doesn't.
Why Does a $200 Million "Crypto Fund" Own Neuralink?
This is the detail I keep turning over. K2 Global's $200 million appeared in a crypto funding digest, filed under its crypto fund. The mandate disclosed with the raise covers AI, robotics, semiconductors, compute infrastructure, and national security technologies. The portfolio companies named alongside it are Neuralink, xAI, SpaceX, Shield AI, Tenstorrent, and Agility Robotics.
Not one crypto company in the list.
I'm not calling that a bait and switch. I think it's the honest shape of allocation in 2026, and the digests are simply catching a firm that historically wrote crypto checks doing what its LPs are now paying it to do. Accel's four new funds are explicitly framed around AI and the broader tech cycle, with Israel among the geographies in scope, which happens to be the ecosystem where a good share of my own network lives. Securing autonomous agents is the stated purpose of Team8's third fund, South Park Commons went to physical AI, and every one of those mandates was written by people who watched the last two crypto cycles from close enough to hold a considered opinion about them.
So when a Web3 founder tells me the dry powder is out there, I believe them. I just want them to look at whose mandate it's sitting under. As Investment Director at ChainGPT Labs I sit on the AI-adjacent side of this deal flow most weeks, and the shift in the room is unmistakable. The question a generalist partner asks is no longer "what's your token design." It's "what does this do that an AI company couldn't do without you."
What Actually Cleared, and Why
Look at the handful of teams that did raise primary capital in those three weeks. There's a pattern, and it isn't subtle.
NeoSoul raised $11 million for infrastructure that lets traders configure autonomous AI agents. Another $9 million went to Vangrid for a decentralised spatial intelligence network pitched squarely at the physical-AI era. Yellow Card took $40 million to let banks process stablecoin payments, with Standard Chartered's venture arm and Sony's innovation fund in the round. Every one of those stories is legible to an allocator who has never opened a block explorer, and two of them have a named institutional counterparty attached.
Beldex is the interesting exception. Eight million dollars for a privacy L1 — about as crypto-native a story as exists — and the money followed a decision to stop shipping end-user products and start shipping developer infrastructure, an SDK and a wallet extension. Even the exception cleared by becoming something a non-crypto buyer could evaluate.
Of the companies we've backed at PRIM3, the ones getting the easiest meetings this year are the ones a bank's product team can describe without saying the word blockchain. Kima Network is the clearest example I have. What it does underneath is cross-chain settlement, which is genuinely hard engineering, but the sentence you actually say in the meeting is about moving money between systems that don't talk to each other, and that sentence travels.
Where This Reading Breaks Down
I want to be careful here, because there are at least three ways I could be overstating the case.
Dry powder does eventually deploy, and the deployment schedule is most of the reason the two numbers diverge in the first place: a fund that closed this August writes its first cheques sometime in 2027 and its last around 2031, which means the $3.55 billion Accel just raised will surface as company rounds spread across four or five vintages rather than as a wire this autumn. Some meaningful share of it will land on teams that look crypto-native by the time it does. Timing lag is not absence. I'd also point out that I have been on the wrong side of exactly this arithmetic before, because if I'd run the same calculation in the middle of 2020 I would have concluded the market was finished roughly eight months before it did the precise opposite, and that memory is the main reason I'm hedging this piece as hard as I am.
Three weeks in August is also a thin, seasonal sample from one tracker with one methodology. And it systematically undercounts the bucket I care most about, because undisclosed rounds don't carry a number into the total. Blueprint Finance closed a strategic round led by Polychain with Bullish, BitGo, FalconX and Keyrock participating — no figure published. Ripple backed two RWA tokenization providers, ZILO and Licuido, with no figure. ZIGChain took investment from Nomura's Laser Digital, no figure. The real builder number for those weeks is higher than $93 million, possibly by a lot. What I don't think changes is the ratio's order of magnitude.
The third thing I'd push back on if someone said it to me: an AI mandate isn't hostile to crypto. Team8 raising to secure autonomous agents is a bridge, not a wall. Agents that transact need settlement, identity, and payment rails, which is the whole argument I made about the two payment rails agents can use earlier this summer. The mandate has moved. It hasn't closed.
What Founders Should Do With This
Three practical things, and I'd argue for all three in any pre-seed conversation I take next month.
Stop quoting the headline number in your deck. When a partner reads "$5 billion raised in crypto last month" on slide four, they know exactly which bucket that came from, and it reads as either sloppiness or spin. Give them the third bucket instead, and give it to them honestly, because being the person in the room who describes their own market accurately is worth more than any slide you could put in its place.
Then build the round for a generalist. Not because crypto-native funds are gone, but because the marginal cheque is increasingly coming from someone whose thesis document says AI. That means revenue, or a named institutional counterparty, or a wedge that a non-crypto partner can restate to their IC without your help. The founder survey data this year already pointed the same direction: most teams now want equity in the structure, not a token-only deal, and a meaningful share are already generating revenue at seed.
And price the raise for eighteen months, not twelve. If the crypto-native seed cheque is scarcer than the headlines suggest — and I think it is, for now — then the cost of being wrong about your runway is existential, while the cost of raising slightly more at a slightly worse valuation is annoying. I've watched that trade go badly in both directions across the portfolio, and only one of them kills you.
Questions I Keep Getting
Is crypto venture funding actually down in 2026? Deployment into new companies is thinner than the headlines imply, but committed capital is not. The gap between the two is the point of this piece. Fund closes, corporate debt, and treasury transactions all get counted as "crypto funding" alongside primary rounds.
Should a Web3 founder pitch generalist AI funds now? If the product has a defensible answer to what it does for an AI-native buyer, yes. That's where the newest mandates are written. If it doesn't, a generalist partner will pass faster than a crypto-native one, so know which conversation you're in before you take the meeting.
Does this mean crypto-native funds stopped investing? No. Coinbase Ventures, Ethereal Ventures, Polychain, Haun Ventures and others were all active in these same three weeks. They're writing smaller cheques into fewer, more legible companies, and a good number of those rounds went out without a disclosed figure.
Everyone in this market is reading the same digests. Very few people are reading the second paragraph, and that's where the actual market is.