How He Raised $650M During Crypto’s Biggest Crash (ft. Haseeb)
Summary
Dragonfly’s $650 million fourth fund is a countercyclical bet on financial crypto, not a claim that the market hangover is over. Haseeb says the firm now manages roughly $4 billion and raised amid what Fortune called an “extinction-level event for crypto VCs.” With valuations down and capital scarce, “money goes a long way”: the mandate is to buy durable financial infrastructure while “everything is on sale.”
Haseeb’s foundational thesis is that crypto succeeds when it touches money, while most non-financial visions were an industry-created “fever dream.” Bitcoin, Ethereum, ICOs, DeFi, NFTs, stablecoins, RWAs, perpetuals, and prediction markets all won bottom-up adoption; decentralized social media, DAOs, gaming, and metaverse land were narratives investors talked themselves into. His post-Gensler challenge is blunt: “Where are all the games?… Nobody wants these things.”
The host’s institutional-capture “black pill” gets rejected because incumbent adoption is backward-looking and does not reproduce crypto’s defining properties. NYSE trading 24/7 or institutions tokenizing bonds may be useful, but neither necessarily requires a blockchain nor delivers “global, permissionless, 24/7 programmable finance.” Haseeb compares fears of BlackRock owning crypto to early predictions that Time magazine or Walmart would own the internet: incumbents can migrate existing products online without capturing the new layer’s scarce value.
Haseeb sees patience—and avoiding forced liquidation—as crypto’s most persistent edge. Bitcoin had been near $120,000 only four months earlier, yet the party had abruptly become a “violently hung over” market; his answer is that ICO, DeFi, FTX, and Luna collapses punished those forced to sell while the underlying system kept growing. “Patience is the antidote to almost all problems in crypto.”
Crypto’s scams will never disappear because making money permissionless liberates the worst of human behavior alongside the best, but Haseeb thinks the scam-to-substance ratio has improved materially since 2017. The next speculative revival will not simply resurrect memecoins: like online poker, every game becomes professionalized until amateurs leave and “shark on shark action” remains. He expects a new, probably AI-native game that today’s trench traders will initially dismiss as stupid.
Agentic commerce is a “yes-and” market: approved purchases can stay on credit cards, while truly autonomous agents need crypto-like settlement. Closed AI labs are unlikely to tolerate the liability of agents spending without human approval, and card networks will resist prompt-injected purchases followed by chargebacks. Stablecoins supply the opposite API—irreversible, global, buyer-beware payments—for agents paying other agents, unknown counterparties, or financial markets, especially outside the U.S.
The agent-payment thesis is still far ahead of its evidence. Haseeb describes x402 activity as tiny, with rough figures ranging from “less than a million dollars” to around “a million a day” after discussing farms, and warns that an agent given crypto today may lose it within days. His bet rests on rapidly improving models and human demand to automate procurement and work, not on meaningful current volumes.
On AI, Haseeb expects U.S. frontier labs to widen their lead, eventually become quasi-national-security contractors, and receive government support in a genuine bust. He tentatively places that transition around 2028–29, when military use and economic disruption become large enough to require clearances and Los Alamos-style secrecy. His closing operating principle spans both technologies: “Stay in the game,” because consistent participants often outlast smarter rivals who win early and check out.
Deep dive
1. Dragonfly raised $650 million by staying anchored to financial crypto
Dragonfly announced its fourth fund at $650 million while many peers were downsizing or unable to raise at all. Haseeb called the backdrop difficult for every fund manager, but said Dragonfly was “lucky to have money at a time when money goes a long way.”
The firm manages roughly $4 billion and has backed businesses including Polymarket, Bybit, Monad, MegaETH, Kaito, and Ethena. Fortune’s accompanying headline described an “extinction-level event for crypto VCs,” a headline Haseeb cited as an honest description of the fundraising market.
His explanation for earning another fund was not simply surviving: Dragonfly got one of crypto’s largest questions right. “What’s this space about?” Its answer was money, finance, and an “internet of value”—not decentralized social media, gaming, or reorganizing society around DAOs.
The historical sequence is his evidence: Bitcoin introduced peer-to-peer payments and non-sovereign money; Ethereum added programmable financial contracts; ICOs supplied capital formation; DeFi made the thesis explicit. NFTs, stablecoins, RWAs, perpetuals, and prediction markets extended the same monetary lineage.
2. User demand built financial crypto; investors manufactured the rest
Haseeb stresses that financial applications were adopted “bottoms up.” VCs did not write a manifesto that summoned stablecoins or DeFi into existence; users began using them, and investors noticed that entrepreneurs deserved more capital to accelerate what was already working.
By contrast, Web3 gaming, DAOs, metaverse worlds, and utopian non-financial use cases were “a whole industry-wide fever dream.” Decentraland’s billions of dollars against roughly 300 daily active users became the specimen: investors excused missing usage because buyers still assigned value to virtual land.
The post-regulatory pushback is deliberately unforgiving. Chris Dixon argued that regulation and Gary Gensler made non-financial crypto too early; Haseeb’s response was, “Gensler’s gone… Where are all the games?” Even gaming founders now tell him users mainly want better products rather than Web3 elements.
3. Crypto attacks the financial layer the internet was forbidden to change
Haseeb’s PayPal example carries the structural argument: Peter Thiel originally wanted an internet-native currency untethered from banks, but regulators forced PayPal to partner with Wells Fargo. What remained was largely “a thin overlay over the banking system,” not a new monetary substrate.
Banking still follows limited hours because governments ultimately control national financial systems. Private banks may operate them during normal conditions, but crises reveal the state’s authority; controlling that “narrow waist” lets governments regulate how money moves.
Internet-era innovation consequently improved only the edges—mobile account access and check scanning—without replacing the underlying architecture. Crypto “grew like a weed” because it finally competed with financial institutions directly, even as companies touching finance faced subpoenas, threats, shutdowns, and debanking.
4. Incumbent tokenization is not the same product as permissionless finance
Thread Guy’s black pill was that BlackRock, the NYSE, and large Web2 institutions might adopt tokenization inside walled gardens while sending no users or capital toward existing DeFi protocols, applications, or infrastructure.
Haseeb called that “obviously wrong” by analogy to the early internet: incumbents had distribution and developers, yet Exxon, Walmart, and Time magazine did not own the internet merely by arriving online. Their inherited form factors were optimized for the previous world.
The 2017 Enterprise Ethereum Alliance already ran the institutional-arrival script: major financial companies proposed Ethereum forks, tokenized bonds, and blockchain settlement projects, yet Haseeb could not identify a meaningful legacy from the initiative. “They were doing the exact same thing they’re doing now.”
NYSE trading 24/7 is good, he conceded, but currencies already trade continuously and the exchange never needed crypto to change its hours. The real product is “global, permissionless, 24/7 programmable finance”; tokenized bonds are necessary inventory, but not where he expects value to accrue.
5. The downturn rewards anyone who is not forced to sell
Haseeb frames crypto’s cycle as a club switching from free drinks and loud music to lights-on humiliation. Bitcoin had traded near $120,000 roughly four months earlier; now “the party’s over,” participants are going home, and the remaining crowd feels “violently hung over.”
A VC’s crucial advantage is never being a forced seller. Across the ICO collapse, DeFi bust, FTX, the post-Trump hangover, and the Luna collapse, Haseeb argues that forced liquidation—not temporary drawdown—was the recurring mechanism that permanently destroyed returns.
That makes deployment unusually straightforward: “You want to be greedy when others are fearful… Well, obviously. You buy shit.” Dragonfly is concentrating on working financial products, including large investments in Polymarket and Rain, major positions in Ethena and Mesh, and a $75 million round it led for Mesh.
6. Every speculative game ends when professionals solve it
“Crypto will never not be scammy,” Haseeb says, because completely free money enables both donations to causes and instant memecoin rug pulls. Still, he believes 2017’s enormous schemes—including BitConnect—made today’s scam-to-legitimate-project ratio meaningfully better.
The claim that crypto’s reputation has never been worse struck both speakers as ahistorical. Haseeb called the period after FTX “a lower rung of hell”; Thread Guy objected to revisionist accounts suggesting the collapse was not that severe or that confidence remained intact.
Haseeb’s professional-poker analogy explains why retail games die. Early online poker let moderately skilled players profit while amateurs still had fun; improving professionals eventually made every table mechanical and unwelcoming until only “shark on shark action” remained.
Crypto rotates the same way—from ICOs to DeFi farming, NFTs, and memecoins—often with each generation dismissing its successor. Memecoins now have too many professionals and too complete a supply chain; the next game may involve AI, but Haseeb’s “honest non-answer” is that he does not yet know its form.
7. Autonomous agents divide payments into approval and no-chargeback rails
Haseeb rejects a binary contest between cards and USDC. ChatGPT commerce through Shopify or Etsy can preserve the familiar workflow: the human approves a purchase and a credit card settles it. Amazon’s absence reflects platform control, not proof that cards cannot serve agents.
Fully autonomous activity is different because major labs face extraordinary liability. An agent might be prompt-injected, buy unwanted goods while its owner sleeps, or execute 50 transactions in two minutes; repeated chargebacks would push card networks to prohibit spending without a clear approval record.
Open-source agents should move faster because their users accept more risk—the software equivalent of people injecting experimental peptides before regulators declare them safe. The autonomous-payment API therefore becomes “no chargebacks” and “buyer beware”: precisely the properties Haseeb assigns to stablecoins.
That rail also expands the addressable transaction set. An agent could pay another agent, an unidentified counterparty anywhere in the world, or purchase a financial asset—cases credit cards cannot reliably cover. Haseeb expects early adoption outside the U.S., where cross-border commerce and B2B stablecoins are already more relevant.
8. Today’s agent-payment numbers are tiny, but capability is compounding
Haseeb repeatedly cautions that the thesis is early. He called x402—the Coinbase-adopted payment protocol—increasingly dominant but economically insignificant, describing volumes as “less than a million dollars” before also giving a rough “million a day” figure while separating out farming activity.
The more important question is what agents are actually buying, and his answer is that current use cases remain small. “If you give your model crypto… you should not expect to see that crypto in a few days,” because present systems are still too vulnerable and unreliable.
His optimism rests on rate of change: generalized agents only recently became credible, and model quality is improving quickly. People will automate procurement and pieces of their own jobs wherever possible; economic activity will migrate to agents “whether we want them to or not,” often before observers recognize it.
9. Frontier AI may become national infrastructure, but not tomorrow
Thread Guy characterized China and the U.S. as frighteningly close in AI; Haseeb called that “a very mid-grade take.” Chinese open models may trail by only months, he said, but they allegedly distill U.S. systems while frontier labs increasingly compound proprietary synthetic and internal data.
His benchmark specimen was Alibaba’s Qwen: excellent on older AIME mathematics contests but sharply worse on new ones, which he interpreted as evidence of benchmark gaming. Actual production coding, in his experience, leaves Chinese models “just way worse,” even though they have improved markedly in six months.
China remains the only formidable challenger and is far ahead of Europe or Japan, but Haseeb expects the gap between open and closed models to grow. Measurement will become harder as benchmarks lose meaning, while the leading U.S. labs accumulate private data that followers cannot reproduce.
Once military use “goes vertical” and AI moves GDP by something like 10% annually, he expects U.S. controls to harden—perhaps around 2028 or 2029, though “I don’t think it’s 2027.” Labs would resemble government contractors: sensitive workers hold clearances, secrecy approaches Los Alamos, and competition becomes prohibitively capital- and regulation-intensive.
10. A genuine AI bust would invite government buying, not routine VC support
Thread Guy cited the reported government use of Claude in the operation to capture Maduro and the public dispute involving Anthropic. Haseeb said it was not surprising and guessed—not with certainty—that Anthropic ultimately caves because the U.S. government is “the biggest enterprise in the world” and too valuable a customer to lose.
He resisted vague claims that Washington will “backstop” OpenAI. That could mean equity, bonds, compute purchases, nationalization during insolvency, or merely supportive words; without specifying the mechanism, the phrase explains little.
OpenAI, in his telling, remains far from financial distress: it continues raising at higher valuations, and Nvidia declining to invest $50 billion would not itself create a collapse. The company could simply reduce compute commitments.
If a genuine bust arrived, however, Haseeb thinks government intervention would be “almost no question.” Washington could buy bonds, invest in compute, or pre-buy demand on the credible argument that frontier capacity is strategically necessary even before consumers or enterprises will fund it.
11. Longevity beats brilliance when early winners retire
Reflecting on Kyle Samani’s departure, Haseeb recalled entering crypto beside fund managers he regarded as unreachable titans. Kyle had remained one of the last peers “bruising every single day,” fighting for portfolio companies while other early leaders became quieter or checked out.
Naval Ravikant’s lesson was unsentimental: “The smart people win early and check out.” Stay long enough and seemingly unbeatable rivals retire, grow lazy, disappear, or implode; disciplined operators often do not defeat them directly because “they will beat themselves.”
Talent leaving for fintech worries Haseeb, and talent joining banks would leave him “in tears.” Moving into AI is different: he regards it as the century’s most important technology and sees its advances as additive to crypto, smartphones, and energy rather than a zero-sum competitor.
His final invocation was simply, “Stay in the game.” Crypto is not the only worthwhile technology, and understanding AI will make participants better at crypto; after roughly a decade of grinding, Haseeb is “not going anywhere” because he believes the industry still has much further to run.