加密代币从根本上是否已经失灵?
- 只有在代币权利被明确写清时,代币与股权才能共存。 一位参与者认为,权利模糊可能让代币以合理价值的10×价格交易。Venice 从 Dragonfly 等机构获得的10亿美元股权融资最初看起来利多,但 Rob Hadick 在播客中的讨论厘清了文件载明的机制,暴露出持有人定价的其实是对 Erik 的信任,而非一条有保障的价值回流路径。
- Jose 倾向于在股权与代币之间建立契约性连接,而 Kevin 则讲述了自己观点的变化。 Jose 主张采用 ACE 这类结构,而不是“信我”(trust me)式的回购承诺。Kevin 表示,自己曾认为所有价值都应归于代币,但链下业务需要股权实体;ACE 的 KYC 转股机制及其他契约安排,能够让持有人捕获价值,而不必只依赖创始人的承诺。
- Grass 的收入和网络经济学,比反弹声浪所暗示的更有利。 讨论将2025年下半年1400万美元的收入与预计2026年上半年1700万美元的收入进行比较,但指出2026年全年收入可达7000万美元,而运营成本约为3000万美元。其基金会结构不存在代币与股权的二元分割:基金会拥有 IP 和硬件,向 Wynd Labs 支付约20%的费用,并通过约850万个不均质且部分冗余的节点提供筛选后的数据,而不只是原始带宽。
- Synapse 的期权逻辑,是搭乘 Hyperliquid 的流动性。 此前的链上期权平台表现挣扎,因为做市商无法在报价的同一场所进行对冲。Synapse 提议在 Hyperliquid 资产上提供期权,包括周末和上线前市场;但该产品仍面临流动性的鸡生蛋问题,当前成交量偏薄,部分参与者更愿意观望或押注 Derive。
- 加密资产的表现看起来好于股票,但讨论仍停留在战术层面。 内存股下跌后,股票指数可能正在盘整,也可能处于派发阶段。Bitcoin 受益于通胀降温和加息预期消退;后续讨论认为 Saylor 无关紧要,因为他不太可能大量出售 BTC,却也缺乏新的买入资金。ETH 从1700美元高位区间升至约1930美元,被认为主要由永续合约驱动,而 AERO 计划迁移主网,存在利好兑现风险。
- Zcash 的 Ironwood 上线,是近期关键检验。 Ironwood 原预计于28日上线,届时已经修补漏洞的 Orchard 池将切换为仅可提取模式。该池中仍有约400万枚 ZEC;若提取缓慢,将令人安心,若资金快速流出,则可能意味着攻击者正在转移资金。
- 区块链的核心逻辑是,生态需要高质量资产和金融产品。 有人称,Solana 目前可以用约10 bps 的滑点,将10万美元兑换进入 Zcash、HYPE、SpaceX 和 meme-stock 市场;新上线资产的滑点则可达到40–50 bps。Robinhood 的链可能争夺这类交易,早期 meme coin 使用量也被视为真实兴趣的证据,尽管其产品与其他 L2 高度相似。
- AI 讨论仍看多内存、neocloud、推理和开源。 一位参与者持有 Mag 7、内存和 neocloud 仓位,并引用 SemiAnalysis 的观点称,随着供给落后于智能体需求,内存价格可能上涨2–3倍。更好的开源模型会削弱 OpenAI 和 Anthropic 的定价权,却可能让推理服务商和 neocloud 捕获更多价值;前沿模型的需求或许仍会持续,因为微小的质量差异会在大量决策中不断累积。Fable 和 Kimi 也分别获得好评,后者被视为 AI 成本可能大幅下降的证据。
1. 加密资产终于跑赢股票——Saylor 影响力消退、ETH 由永续合约驱动,Zcash 面临 Ironwood 检验
- Jason 表示,加密资产难得一次看起来比股票更强。自6月24日或25日 Micron 发布财报以来,内存股已经回吐涨幅;股票指数可能正在盘整,也可能处于派发阶段,他对两种解释都能自圆其说。Bitcoin 的状态好于过去几周,受益于通胀数据降温和加息预期下降。他还观察到一个显著的卖方压力正在减弱:该卖方从一半现金、一半 Bitcoin 的融资方式,转向积累现金,似乎是为了争取时间并降低眼前的不确定性。
- 随后,一位参与者称 Saylor 是“非关键因素”:他不太可能出售巨量 BTC,但在 STRC 尚未恢复、无法重新提供新资金的情况下,也不太可能大举买入。ETH 从1700美元高位区间升至约1930美元,被形容为高度由永续合约驱动。它未必需要立刻回撤,但这更像是为一笔交易建立仓位,而不是纯粹由现货需求推动,后续可能为山寨币反弹铺路。Muneeb 已公开展示了自己的大额 ETH 仓位。AERO 原预计于本月迁移主网,但也可能像 Grass 一样演变成利好兑现行情。
- Zcash 的关键变量是新资金池 Ironwood,预计于28日上线。届时,发现并修补漏洞的 Orchard 将切换为仅可提取模式,相当于对其中约400万枚 ZEC 进行实时审计。讨论该问题的参与者预计,提取速度会慢于许多人的预期,首周甚至可能不到一半。资金缓慢流出将构成利多信号,而极快的资金回撤则可能意味着攻击者正在转移资金。
- 目前更强势的加密资产仍是 HYPE、Lighter 和 Zcash,风险偏好集中在已经证明过自身 traction 的资产上,至少在 midterms 之前如此。
2. 区块链需要优质资产和有用的金融市场
- 一位参与者表示,Solana 当前的优势在于资产和流动性策略。据称,用户可以用约10 bps 的滑点,将10万美元兑换进入 Zcash、HYPE、SpaceX 和 meme股市场;而较新的资产,滑点可能达到40–50 bps。
- 更广泛的判断是,通用型区块链最清晰的用例,是交易和转移金融资产、以这些资产为抵押借贷,并围绕它们构建产品。如果一条链缺少优质资产,单靠应用本身并没有太多可供使用的东西。Solana 被认为在吸引这些市场方面做得不错。
- Robinhood 被视为这一领域的潜在竞争者。其近期使用情况,包括 meme coin 活动,被看作积极信号:用户愿意跨链并尝试这条链,尽管参与者并不认为 meme coin 交易会重现早期的增长轨迹。
- 另一位参与者则认为 Robinhood 的链不过是使用同一套底层组件的又一个 L2。讨论中没有形成关于代币持有人经济学或未来代币化的更强判断。
3. Synapse 的期权方案,以及交易台为何仍持怀疑态度
- 期权讨论从一次失败的2022年投资说起。问题并不只是需求不足,而是做市商无法在提供报价的同一场所完成对冲。卖出看涨期权后,做市商需要买入永续合约以保持 Delta 中性;中心化交易所则更容易通过统一的交易场所和保证金体系支持这一机制。
- Derive 正在重建自身的链上流动性,同时与 Hyperliquid 和 Lighter 竞争。前桥接团队 Synapse 则转向在 Hyperliquid 之上构建期权。搭乘现有的 Hyperliquid 流动性,理论上可以为该平台已经上线的所有资产提供期权,而不必先建立完整的流动性基础。SpaceX 在 Hyperliquid 上线时,据称期权约占其成交量的5%。
- 潜在产品包括周末期权、Anthropic 这类上线前市场,以及类似 Robinhood 的界面:通过滑块、行权价拐点和盈亏展示,而不是专业交易员使用的、充满 Greeks 的界面。期权带来的用户流转可能少于永续合约,因为买方的下行风险通常仅限于期权费;永续合约交易者则可以反复追加抵押品,最终被清算。零日期权还可以按名义敞口而不只是按期权费收取手续费。
- 投资风险在于,Synapse 能否吸引足够的吃单方流动性,从而让做市商产生兴趣。当前成交量和流动性都被形容为疲弱,形成鸡生蛋问题。市场曾讨论将约80%的手续费用于回购,但这需要治理程序,而且目前手续费极低,实际意义有限。
- 一位参与者表示,近期的集中喊单让他觉得这套交易不对,因此选择观望;另一位则认为,当前真正的问题是哪个模型最终成功的概率更高。如果必须二选一,持怀疑态度的参与者会倾向 Derive。
4. Grass:季节性收入、冗余网络与不存在的股权二元结构
- Grass 的讨论认为,悲观的收入对比具有误导性。市场把2025年下半年的1400万美元收入与预计2026年上半年的1700万美元收入进行比较,但这项业务的收入高度集中在下半年。给出的预测是,2026年全年收入7000万美元,对应约3000万美元的年度运营成本,或者说在再投资前约有4000万美元运营利润。
- 估值讨论更倾向使用调整后 FDV,将 headline 数字下调约30%,以反映由金库和基金会控制的价值。Season 2 参与者原本预计会获得空投,但团队不愿按当时价格发放代币,最终改为现金支付,引发反弹。按项目声称的4×同比增长计算,其交易估值对应不到4×收入倍数;另有一条产品线预计在夏末推出。
- Yan 表示,网络约有850万个节点,但它们并不等价。不同于 Helium,用户通常不需要购买并安装专用设备;Grass 插件在后台运行,利用的是闲置带宽。同一地理区域的数百万个节点会产生冗余,而位于稀缺区域的节点可能更有价值。
- Grass 还在推进纵向一体化,包括自建数据中心和存储数据,因此不必通过庞大的住宅网络反复抓取全部数据。它的价值不只是充当数据管道,而是为那些原本必须拼接并清洗多个数据源的实验室,筛选和整理数据。
- 目前据称已支付约300万美元,而预计收入为7000万美元;项目持续以 USDC 发放奖励,以维持代币供应。Yan 表示,机器人节点或地理位置无用节点退出后,反作弊成本可能下降。
- 该基金会结构被描述为不存在代币与股权的二元分割。基金会拥有 IP、硬件和其他资产,约20%的费用、即每年约600万美元,支付给 Wynd Labs。随后团队被描述为基金会的承包商,而不是底层资产的所有者。这一设计始于2022年,被认为已经解决了约90%的利益一致性问题。
5. 代币与股权:模糊性推高代币,披露则击穿叙事
- Jose 的框架是,当股权实体拿走收入,却只对回购代币作出“信我”(trust me)式承诺时,投资者并不愿意持有代币。他更倾向于 ACE 这类结构,或股权与代币两端之间的其他明确契约关系。
- Venice 是典型案例。项目宣布 Dragonfly 等机构以10亿美元估值参与融资,最初看起来偏利多。随后 Rob Hadick 在播客中解释了文件实际承诺的内容。市场最终将其视为利空,因为持有人意识到,自己依赖的是股权实体履行一项可以被叫停的回购承诺。
- 另一位参与者表示,只要消除造成代币估值超出合理水平10×的模糊性,代币与股权就可以共存。VVV 作为代币本身说得通,但以其曾达到的价格则不合理,除非持有人假设股权最终会实现价值,并将价值回流给代币。在这种模式下,股权获得大部分真实现金利润,而代币需要明确的用途以及更克制的预期。
- 另一位参与者认为,形势好时可以奖励代币,但当业务转弱或寻求退出时,代币就会处于劣势。该参与者还表示,Venice 将一半供应量用于空投,从未直接围绕代币融资,并曾表示目标是尽可能销毁供应量。据称,自主决定的销毁额度最近已提高至约每天2万美元,但销毁时点会影响供应量下降的速度。
- 在这场讨论中,Venice 仍被认为有价值,但核心观点是,如果代币与股权之间的连接更牢固,代币价值很可能会更高。
6. 契约化价值捕获,而非永久信任
- Kevin 讲述了自己观点的变化。他刚进入加密市场时,认为所有价值都应归于代币;对于 AMM 和 MakerDAO CDP 这类纯链上原语,股权实体并无必要。但市场已经证明,大多数有价值的加密业务都需要股权公司,因为它们的现金流来自硬件、服务器、API 请求和法币收入。
- 因此,Kevin 主张为代币设计程序化或契约化的价值捕获机制。他以 MetaLeX 的 ACE 为例:代币可以自由交易,持有人完成 KYC 后则可以成为股东。这类似于稳定币:任何人都可以在链上持有,但要兑换成法币,就必须完成相应的身份验证。若设置一个转换价格,当代币被放弃或公司被收购时,持有人就能捕获代币与股权之间的价差。
- 另一位参与者对 ACE 机制的描述是:项目发行代币,同时将资金融入股权端,代币持有人可以按预先确定的价格转换为股权。基金会或股权实体也可以签署契约,将全部价值或固定比例的价值分配给代币持有人。其目的,是应对代币被放弃、股权却被收购这一反复出现的模式,Vertex 和 Axelar 被列为例子。
- 讨论否定了所有初创公司都应将全部收入用于回购的观点。HYPE 已经制造出强烈预期,认为回购是代币获得价值的唯一途径,但早期公司通常需要将资金再投入增长。如果市场相信多余价值最终会归于代币,投资者或许会更有耐心。Grass 被视为一个例子:项目在建设数据中心的同时,去年将月度运营开支削减了超过100万美元。
- 流动代币也会带来社会和估值成本:创始人需要回应更广泛的持有人群体,而代币市场给出的估值可能低于同等股权公司的估值。一位参与者认为,Grass 如果以股权公司身份融资,估值很可能远高于约4亿美元的代币估值。另一位则认为,私募股权的流动性被高估了,因为存在禁售期、员工要约收购,以及最高可达5%的经纪费用。讨论最后提到一个可能的解决方案,但没有展开说明。
7. AI 交易:内存、neocloud 与开源
- 一位参与者继续持有 Mag 7、内存和 neocloud 仓位,同时承认这些仓位走势高度相关。韩国散户杠杆正在加剧波动;一个流传的统计数字称,每30名成年人中就有1人遭遇追加保证金,韩国市场则被形容为多次以跌停收盘。
- 长期内存逻辑仍未改变。讨论引用 SemiAnalysis 4月的研究,认为随着供给无法追上智能体工作流带来的需求,价格仍可能上涨2–3倍。Meta 宣布出售算力后,neocloud 股价下跌,但该参与者认为这一反应不理性,因为 CoreWeave 等公司的合同规模达到数百亿美元。土地和电力仍然有价值,IREN 被作为例子提及。
- 讨论否定了开源模型的看空逻辑。如果中国模型变得足够强大,OpenAI 和 Anthropic 可能失去定价权,但推理服务商和 neocloud 可以通过直接服务开源模型捕获更多利润。用户将获得更便宜的智能,而该参与者称智能是“终极 Jevons 悖论”。
- 市场对企业投资回报率的抱怨被形容为能力问题:在庞大的成本基础上实现5–10%的效率提升,可能超过在更小成本基础上将 token 支出翻倍的影响。另一位参与者认为,前沿模型仍可能保持优势,因为微小的质量差异会在大量决策中累积。部分任务可以分配给更便宜的模型,但许多用户都表示,每次新的前沿模型发布时,实际效果仍会让他们感到意外。
- 一位参与者称 Fable 的能力强得异乎寻常。另一位表示,使用 Kimi 后很难做空 AI 股票;如果当前趋势延续,类似水平的能力可能在6个月内变得便宜一个数量级。
8. 蒸馏、开源激励与结尾的反事实推演
- 讨论质疑:如果世界上已经有据称达到 AGI 水平的系统,为什么仍然无法阻止模型蒸馏?一位参与者表示,中国模型似乎含有明显的 Claude 痕迹,并疑惑为什么无法识别聚类式查询和 IP 遮蔽,同时也承认不确定其中有多少来自蒸馏,有多少源于中国自身的创新。
- 一位参与者认为,中国不能简单地将模型关闭,因为除非模型处于前沿水平,否则闭源并没有太大优势。更可行的路径可能是推出开源模型,同时采用限制性许可证,要求大规模提供推理服务的厂商分享收入。
- 支持开源的激励来自多个方向:Nvidia 和主要科技公司都在投资开源模型,而真正明确受到威胁的,只有两家前沿实验室,以及正在逼近前沿能力的开源模型。前沿模型提供商也可以向推理或算力纵向整合,因此,开源较旧模型、同时保留下游价值,可能是理性的选择。
- 结尾的反事实推演是,如果没有 AI 热潮,加密市场可能会成为唯一的游戏,因为 GDP 走弱将带来更多货币支持。参与者开玩笑说,meme coin 可能会登上 FIFA 广告,并庆幸 NFT 在被铺满世界杯之前就已经退潮。
完整逐字稿
Is there a world where that token-and-equity model split can make sense, or does it almost always lead to misaligned incentives?
1. The Token vs Equity Debate
Yeah, no. I just think people don't want to hold tokens in things that have an equity entity that captures revenue.
They can coexist where the ambiguity that gives the token a 10× higher valuation than it deserves does not exist.
When times are good, the token can get rewarded, but once things go south, or if you're trying to get an exit, then it goes to [__].
If that's the case, we need to figure out a structure where the token can have some kind of programmatic or contractual value capture.
All right. Welcome back to the Delphi Podcast, the show about markets, crypto, AI, and unfiltered opinions. I'm your host Kevin Kelly and joined usual by the typical cast of characters. We got Jose, Yan, Satoris, and Jason.
Since a lot's been happening in the markets, we've had some trades unwind, and we've had crypto showing a little bit of green on the screen. I'd love to start off by getting each of your takes on the current state of markets. Maybe, Jason, we'll start with you.
For once, crypto's actually looking a little bit better than equities, right? When I look at equities, to your point, you mentioned the unwind. You're kind of seeing it with the memory names, right? Micron, after they had their great earnings, which was June 24 or June 25, had a super volatile day. Since then, you've seen an unwind across a lot of the names and things that have been driving a lot of the bullish price action over the last couple of months.
I'm starting to get a little bit concerned there. When I look at equity indices, I think you could look at this in one of 2 ways: it's either consolidating or distributing. It's one of the 2, and when I zoom out, I think you can make the case for both right now.
Going back to crypto, I think Bitcoin looks better than it has for the first time in weeks. A lot of that is maybe the seller overhang starting to abate—not in terms of the actual numbers, but I think he clearly realizes he messed up and is trying to raise as much cash as possible. Before, he was doing a half-and-half cash-Bitcoin type of raise, and now he's just accumulating cash, probably to buy himself some time and remove that immediate uncertainty. Maybe that's why Bitcoin is doing decently well.
Obviously, inflation prints came in cool, which is good. That pulls away rate-hike expectations, which I think we were dubious about anyway. Generally, things look okay for me on the crypto side. A lot of the names that have been doing well in crypto are continuing to do well: HYPE, Lighter, and Zcash, which is back to its pre-exploit levels almost entirely.
There is some risk appetite in the names that have proven themselves. I'm feeling decent about the 3 names that have been doing well in crypto continuing to do well, at least until midterms.
The market definitely looks a lot better than it's been in a while. You're seeing more and more things do well. Just anecdotally, I think my watch list had basically the majors and then HYPE, Zcash, and Venice 2 months ago. Now it's grown a lot. There are a lot more green shoots happening, so it's definitely an interesting market.
I think Zcash getting back to the pre-bug level is pretty bullish for it. Ironwood, which is the new pool, goes live on the 28th, I believe—13 days from now. Then you'll start to get that real-time audit as the Orchard pool, the one where the bug was found and patched, is put into withdrawal-only mode.
You'll start to see the funds move out of there. I think it'll be slower than people might think. There are around 4 million Zcash in there. You'll probably see half of it move within the first week. I don't even know if you'll see half of it move within the first week, really. I think it could be slower than people would expect.
If it's slow, that's also a bullish sign. The main risk is that you see a drawdown super fast. Then you could think, “Oh, maybe somebody is actually moving a lot to Ironwood from here who exploited it.” We'll see how that goes.
There's a lot of interesting stuff happening on-chain, too. Solana has been doing really well with all the stock stuff they've been getting on the chain. I think they've been doing a really good job.
Isn't that just a few wallets? I saw some posts.
I don't know who the people actually buying this are. I just mean from a liquidity standpoint. On Solana right now, on-chain, you can swap $100K into Zcash, HYPE, SpaceX, and all the meme stocks, and you get maybe 10 bps of slippage. On some of the newer listed stocks, you get up to 40 or 50 bps of slippage. That's cool. It's pretty good.
At the end of the day, when it comes to general-purpose blockchains and all that, it's pretty clear that the main use case people like is using them to trade and move financial assets around, borrow and lend against them, and all of that. If that's the main use case, your job as an ecosystem is to get the best assets and build products around them.
That's why, if you look at so many other chains out there, they just don't have good assets. It doesn't matter what applications you build on them if there aren't good assets to do stuff with. I think Solana has done a good job getting all that.
I feel like Robinhood is a pretty clear ecosystem that could compete in that same domain. I think they should go after the same things Solana is. We'll see how that shakes out.
2. Synapse, Options & Hyperliquid
Now that they got some good usage this past week, with Robinhood and a lot of meme-coin stuff—which some people like to talk down on—it is what it is. You didn't see meme-coin trading take off on a lot of other hype chains, right? That's always a signal. Even if I don't think any of us here really believe that the meme-coin game is going to play out the way it has in the past, and buying the first meme on Robinhood is going to go to the millions, it's still a good sign overall that people have wanted to bridge there and that there's genuine excitement for a lot of people to use the Robinhood chain.
I think it's so uninteresting as a thing, though. It's just another L2—
For sure.
—doing the same exact primitives. There's nothing new.
Actually, speaking of that, do you want to talk about Synapse? Synapse is probably one of the more interesting setups right now. I guess we'll skip the whole market chat, but just on the topic of options, we invested in something on the options front back in 2022, and the basic thesis for why options didn't really play out on-chain was because—
They've never played out, right? There have been so many attempts.
Yeah, there have been attempts throughout the years. Market makers can't provide liquidity. The setup was never enabled for market makers to provide liquidity in the same venue where they hedged.
When they sell a call, they have to long the perp to be delta-neutral. Typically, on centralized exchanges, they were able to do that all in one place. You didn't have this portfolio margin, but in on-chain land, that never really existed.
So, we invested in something, and basically, the reason it didn’t work was the difficulty around the liquidity side for the DEXs, right?
So, it’s like during the GMX heyday. It was a similar GMX-type pool, and at the time, GMX was great, but in the end, better systems were invented. So, Derive now is doing that, but they’re trying to rebuild all of their liquidity on-chain themselves. Right? They’re creating a perp DEX that’s all in one venue, but the difficulty is, as we know, you’re then competing with Hyperliquid and Lighter, and it’s very difficult to get substantial liquidity.
Synapse, the former bridge that the guys pivoted from, had a crack team from the beginning in terms of what they were able to deliver on incredibly low funding, and they delivered a quality bridge. But we all know that bridges aren’t really great businesses, and so they’ve since pivoted to building an options protocol on top of Hyperliquid. So, rather than trying to build liquidity up yourself, you can piggyback off existing Hyperliquid liquidity, which means you can offer options on everything that’s on Hyperliquid.
If you look at Robinhood, their largest revenue-generating product is options. There’s definitely retail demand for them. You could say some of that is satisfied by perps, but there are benefits to options. I think one of them is that you don’t necessarily churn through your user base in the same way, right?
Everyone eventually blows up on perps because you just keep adding collateral when it’s going down and you get liquidated. If you look at the amount of traders that get washed out on the perp side, it’s pretty high. Whereas with options, unless you’re selling stuff, it’s pretty hard to wash out, and your downside is pretty limited to your premium paid. And you have kind of a higher—
Find a way to wash out.
Yeah, but it’s harder. The percentage of your user base that you churn through is lower.
Is that true? I just feel like with Robinhood, the volume they have on zero-day options makes it feel like people are probably full-porting.
Yeah, but you’re not—I mean, if you’re full-porting, that’s a different story, but you’re not having these washouts where your user base is just gone. Right?
Yeah, you don’t have a 10/10 event.
One-time fee. And the zero-day stuff is very high-fee-generating, right? Because you’re paying a fee on notional exposure, not on premium. So, the higher volume you do, the more revenue you’re generating. That’s why they’re pushing them. But it’s path-independent leverage, which is nice.
I’m not going to sit here and shill options, but I do think, in general, there are a lot of benefits to them. Even when SpaceX was launched on Hyperliquid, they were about 5% of SpaceX volume. So, I think it’ll also drive a lot of volume to Hyperliquid themselves, and I think it’s a pretty compelling options product.
You’ll get options on weekends, which you don’t really have. You have zero-day stuff on weekends. You’ll have options on stuff that hasn’t even gone live, like Anthropic and stuff. So, I think there are a lot of compelling, novel ways to express bets via options.
This, in particular, is part of the problem solved by piggybacking off Hyperliquid liquidity. Naturally, you can’t assume that just because you build it, they will come. I think that’s kind of the risk you’re underwriting when investing in Syn: Do you think they can draw in enough taker-side liquidity and bring enough users to the platform? And they’re all competing—
I think that everything around Syn just kind of smells pretty bad, though. I feel like everything over the past month has been—I don’t know. It’s hard for me to buy SNX, given the history and everything.
What’s the issue, specifically?
All the shilling of it recently. I don’t know. It just smells wrong to me. People actually use it, and it’s trading like not that much more than—
I think it’s a couple of times higher. Drift isn’t really doing much in terms of volume either. Neither of them are doing much. Obviously, Drift is doing quite a bit more because Syn just started, but it’s more a bet on which model you think has a higher probability of success.
I don’t think you’re buying either of them for their fundamental value at the moment, but rather for the probability of them being able to deliver on something much bigger in the future. I wasn’t expecting you guys to shill SNX, to be completely—
Well, what’s the value accrual for Syn? How does it operate?
They’re going to do a bunch of buybacks. They’ll enable buybacks. I think what’s been mentioned is 80% of fees. But right now, it’s nothing because fees are nothing. They’ll basically do some kind of governance thing to enable buybacks.
Got you. And how would you know this is working? How do you know—
Volume.
Because, yeah, I bought some—you know, you shilled me into it. I bought some, but I actually sold a chunk of it. What I’m thinking is that the hype was very ahead of the fundamentals. I think it was trading higher than Derive at some point during the pump, and then Derive did pretty well.
Well, Derive had the open listing, and then—
At what point do you think it proves out the thesis that this cross-margin thing is working? As of right now, volumes are pretty trash, right?
Yeah.
On that, liquidity doesn’t seem that great either.
Yeah. There’s a bit of a chicken-or-the-egg situation, right? I think they’re onboarding market makers—
Mm-hmm.
—to help provide maker-side liquidity, and ultimately, you do need to bring users to provide taker-side liquidity. Otherwise, it’s not really interesting for market makers. But I think over time, you basically have to wait and see. If adoption doesn’t really take place, then either people don’t want options or the product doesn’t work.
I think you could probably do some sort of spread analysis. I don’t think they have much in terms of overlapping assets right now, but once you see Bitcoin options and some other things, it’ll be a much easier apples-to-apples analysis of which platform can provide better liquidity.
They’re also doing something different. Derive, and even Deribit back in the day, had a very professional-feeling front end, where you have a bunch of the Greeks, which most people don’t know how to think about. These guys are going with the Robinhood UX/UI, where it’s just this slider, a kink at the strike, and a P&L-type thing. So, it’s definitely geared more toward retail in that sense.
Yeah, I don’t really have a strong opinion on this space, to be honest. I haven’t really seen anybody succeed. I don’t know enough about what it really requires on the back end to do it either. I’m kind of sitting this one out, but if I had to, I’d probably lean Derive, to be honest.
Yeah, and broadly, the market is definitely looking better. It seems like Saylor just being a non-factor is good. It’s tough to see a world where he’s going to sell much BTC. He’ll sell some, maybe, but not an insane amount. He’s also unlikely to buy, based on his cash pile and STRC not coming back to give him new, fresh powder. So, he’s a non-factor.
You’re seeing ETH rip, which I think just scares most people, but is also a bit of a precursor for an alt rally. The thing is, I think the most recent ETH move has been heavily perp-driven. If you look at ETH UI and the jump from the high 1700s to the current price, which is 1930, it’s pretty perp-driven. Not to say that it needs to retrace immediately, but it does indicate that people are positioning for a trade more likely.
3. Grass, Revenue & Token Economics
You always wonder who’s buying ETH, but Muneeb has publicly shown that he’s got a big position. Ideally, you kind of see some alts follow. I think you have some event-driven stuff. AERO is one that I’m watching, with the migration to mainnet supposed to happen this month. But we’ll see if these things end up being a sell-the-news situation, which I’d say was the case with Grass—a combination of sell-the-news and—
Yeah, why don’t we talk about Grass a little bit? I know you had a thread here.
Yeah, I think there was a misunderstanding. The crux of it is that Grass’s revenue is seasonal and very back-half-heavy. What happened was people were comparing the back half of 2025, at 14 million, to the front half of 2026, at 17 million, and suggesting that growth is pretty weak, which, if you’re looking at those 2 numbers, makes sense.
But they’ve mentioned their seasonality before, and they mentioned it on the call. So, sure, first-half 2026 revenue is expected at 17 million, but the full year is 70 million.
And operating costs are roughly $30 million a year, so $70 million in revenue and $30 million in costs gives you $40 million in operating profit. Obviously, that gets reinvested and goes into everything else, but I wouldn't really anchor a multiple on operating profit; I'd anchor it more on revenue.
So I kind of adjust FDV. I reduce it by about 30% because a bunch is sitting in the treasury, and the foundation kind of controls it. They've shown an appetite to be more surgical and careful with how they distribute the tokens.
People were expecting an airdrop in season 2, but ultimately they just paid in cash because they don't want to give away these tokens at this price. That drew a lot of backlash, and we can chat about that as well, but I think the takeaway is that they're being measured with how they distribute the token.
My personal view is that an adjusted FDV is more appropriate, similar to what you do with HYPE. The other reason you can do that is because there is no equity component. If you're looking at revenue of $70 million and FDV of, call it, $250 million to $260 million, it's a sub-4× revenue multiple on a protocol that's growing 4× year over year.
So it seems like good value, and they have new product lines—or a new one in particular that people are excited about, myself included—launching at the end of summer. It seems like it's one of the more misunderstood tokens because you see a lot of takes about there being an equity component and all of this stuff, along with fear that they're not returning capital to the token, so that means the token is getting rugged, which is just not the case.
Yeah, I think there was also some backlash that we can touch on briefly around just the size of the rewards that are allocated, right? And I know, Yan, you've got a take on this. If you have, call it, 3 to 4 million residential IPs that are basically running their own Grass nodes and allowing Grass to use their unused bandwidth, what winds up happening is that bandwidth isn't used or distributed equally.
Certain geographic areas wind up getting more throughput, and that drives a larger share of the rewards. Maybe you can talk about that, because I think that was one of the key points of backlash, but also probably another thing that a lot of people misunderstand about how the network actually operates.
Yeah. I think we're at about 8.5 million nodes, but, like you mentioned, they're not all made equal. There's some conflation of this being DePIN: with Helium, you had to buy some equipment and set it up, and there was a decent lift, so a higher reward was required.
With this, I think most people forgot they were even running it. It's a plug-in; your computer doesn't operate any differently. It's completely unchanged—it just runs in the background and uses a little extra bandwidth when there's capacity for it. So I think people were conflating things there.
Ultimately, there's redundancy. You don't necessarily need 1 million residential IPs if you have 1 million in the same area; it's just overkill, right? There's value to be had in being one of the few available in that area and being in an area that's sought after.
What that ends up playing out to is that you don't necessarily need an insane number of nodes to make this work, particularly because they've already downloaded quite a bit. They've verticalized a bit and have their own data center, so they're storing quite a bit of data already. They don't necessarily need to rescrape for all of these.
I think what they're actually able to do that's compelling is that they're not just a data pipe. They do a lot of curation and filtering, delivering a filtered, specific product to these labs because the labs are buying data from a bunch of different sources. They don't have curation in-house most of the time, or if they do, it's a pain in the ass. So there's a premium that they pay for cleaned-up data.
All that to say, the size of their network doesn't need to be as massive now. I think they showed that with the amount that they paid. They dropped about $3 million, if I remember correctly, and they're set to make about $70 million. They'll have ongoing rewards, and those will be in USDC, in the interest of preserving the token supply.
With the amount of data they have and the distribution of nodes, it just doesn't need to be that high. Andre even mentioned that nodes dropping off is going to save them money because there were so many people botting for the airdrop who aren't really providing useful bandwidth or operating in a useful area.
As a result, they have to spend less on anti-botting on the node side. So it's kind of a win-win for them: the toxic nodes are also not incentivized to participate.
Yeah, and I think it leads to this: people just don't want to hold tokens.
Yeah, I think you're going to go to the same place I was going, Kevin, so go ahead.
Yeah, no, I was just going to say I think it leads to this bigger conversation I wanted to have with you guys, because it feels like this debate gets stirred up every 6 to 12 months at this point. More recently, with Venice's equity raise, there's this whole tokens-versus-equity value-accrual debate going on.
One question I had for you guys was: Is there a world where that token-equity model split can make sense, or does it almost always lead to misaligned incentives? Jose, maybe we'll kick off with you, because I know you're about to chime in with something.
Yeah, no, I just think people don't want to hold tokens in things that have an equity entity that captures revenue and makes some kind of effectively trust-me promise to buy back the token. I think that's the long and short of it, honestly, in my opinion.
I think we need more structures like ACE and other ones where you have at least some kind of contractual relationship between the equity and the token, or something that's very clear and delineated. Venice was a really good example, right? They came out with really bullish news that they raised a round from Dragonfly and others at a $1 billion valuation.
Then Rob Hadick made that sort of appearance on the podcast, which wasn't very bullish for the token, and I think ultimately it was bearish overall because people were like, “There's an equity entity, and we're basically relying on them honoring their commitment to buy back this token.” But they could stop at any point, right? They could, and I think people just don't want that anymore when it comes to tokens.
They can coexist, but they can coexist where the ambiguity that gives the token a 10× higher valuation than it deserves doesn't exist. The thing with Venice is that people have treated the token like Erik is a good guy and is going to make the token valuable, and that was always baked into the price of VVV.
Then Rob went on the podcast and talked about how, if you read the docs, this is what Venice is. It's like, okay, but people weren't buying it for just that mechanism, right? So I think tokens and equity can coexist if whatever the token is used for is clearly spelled out and there's no ambiguity.
The problem is that the ambiguity leads to a much higher token valuation, which people love, right?
People aren't bidding this to $1.25 billion in order to mint—
Right. Venice's VVV makes sense as a token; it just doesn't make sense as a token at the price it was at, unless you think that the equity can ever be realized someday and then all the tokens are going to come back at some point.
Tokens and equity can coexist. It's just that when they coexist, the token is a much smaller market cap, the expectations are very tempered, and it really is just this pure utility thing you're getting out of it. The equity is always going to get the bulk of the real cash profits. That's the thing: you just have to get rid of the equity.
When times are good, the token can get rewarded. But when things go south, or if you're trying to get an exit, then it goes to [__].
For the record, on Grass, there is no equity, right? There's a foundation-structured model. The foundation owns everything. They said 20% of expenses—so, call it $6 million a year—is paid to Wynd Labs out of the $70 million in revenue, but the foundation owns all the IP, the hardware, and everything in between.
So, there is no token-equity bifurcation. I think it's more a matter of the market not believing it, not understanding it, and having unrealistic expectations. Venice is slightly different, right? There's definitely a clear equity component. I'm in the camp that the token ends up being the long-term value-accrual approach, because I do think—I don't know how a business like this will play out. I don't know if they'll go public or be acquired or whatever, but that's a long time between now and then.
He's not even going public, though, right? I trust Erik—he's one of the founders I trust most in the space to actually understand this—but you shouldn't have to trust him, right? And there's still the case that, in general, founders and investors are going to own way more of the equity than they do of the token, right?
So, if you have cash flows running through this and the thing's making its own money, there's a conflict between paying out your cash flows to buy the token or just paying them out to equity holders, even assuming they don't go public.
Yeah, but for me, that's distributions, right? It's got to be an equity exit, or it's got to be an exit if you're doing venture.
Yeah, I mean, even an exit, then, right? That's what I like about the ACE thing. It just gives you this—and for those who don't know, ACE is this thing that MetaLeX built, which we're investors in, for full disclosure.
The idea is you can have a token, but you can effectively KYC and become an equity holder, right? So, it's similar to the stablecoin model in the sense that with stablecoins, anyone can hold the stablecoins on-chain, but in order to redeem them for fiat, you have to be KYC'd.
The idea is that you get the upside and liquidity of a token, but the downside protection of being able to convert to equity. This isn't the only model. I think there are a few others that are interesting.
When I got into crypto, I wrote a lot of posts about this going back 10 years ago, or 8 years ago at this point, and I thought that all value should go to the token. You shouldn't have equity entities at all. This made sense for the purely on-chain stuff, like AMMs and MakerDAO CDPs.
At that point, I think we all still thought that a lot of this stuff would be on-chain primitives. What we've found is that the vast majority of valuable crypto projects need an equity company. Their cash flows aren't generated on-chain, right? In the Venice case, they need to buy hardware, run servers, serve API requests, and collect fiat revenue.
In that case, your cash flow isn't on-chain. You need an equity entity. I just think that if that's the case, we need to figure out a structure where the token can have some kind of programmatic or contractual value capture. I think the market's asking for it at this point.
Although Venice is still incredibly valuable, so it's still a win. But I do think it would go up a lot if they had something more solid—a solid link to the equity.
Yeah, I think in Venice's defense, they airdropped half the supply. They never said this was an equity component, and they never did a raise into the token.
And then I guess
For sure.
If we had to poke holes, it's that they let the narrative run. But what were they really supposed to do on that front? People were coming up with their own narratives, and they've stated that their goal is to burn as much of the entire supply as possible.
It's just a matter of timing. If they were taking all the revenue, they would burn all of the supply a lot faster. But I do think they'll continuously increase the burn in different ways. I think they recently increased the discretionary component to $20,000 a day or something like that a few weeks ago.
Hypothetically, using ACE as a great example, what could this have looked like in a parallel universe? I do think one thing the equity raise obviously gave them, to your point, is that they're running a real business that's getting real traction, has real expenses, and whose expenses are scaling.
They're buying hardware, actually running the network itself, and hiring engineers. All of those are expenses that a fast-growing startup has, and you need capital to continue to build and scale that business. So, raising $65 million through some type of ACE-like offering—I'm curious, hypothetically, how could that have looked?
Yeah, I mean, there's the ACE structure, and there's the MetaDAO structure, too. With ACE, you just have to basically set a conversion price from the token to equity.
The idea is that you could issue tokens but raise money into the equity. The token trades freely, but at any point—and they could also commit to using most of the value to buy back tokens in the same sort of loose way they're doing now—token holders don't need to believe them.
In the case that they don't, or that the company gets acquired or something, they can convert to equity. At least someone can convert to equity and arbitrage the token-to-equity spread, if that makes sense.
The other way is just to have some kind of actual contractual agreement, which, to be fair, is what it sounds like Grass did. I'm not super up to date with their structure. Some kind of contractual agreement between a foundation or some kind of equity structure and something that represents token holders could say that all the value, or some fixed portion of the value, will accrue to token holders.
I think what people are worried about is what we've seen many times: the token gets abandoned altogether and the equity gets bought out, like with Vertex or Axelar and these kinds of examples. ACE solves for that and achieves, in my opinion, a good trade-off.
I would love for all these things to be liquid tokens that you could buy to speculate on these businesses. But right now, you're speculating both on the business and on the honesty or trustworthiness of the business's promises to actually honor the token, which really affects the expected value of these investments.
Yeah, I think Grass, to their credit, has a structure where the team members are contractors to the foundation. They don't own anything. The foundation just pays them, and everything is owned by the foundation.
I think, to Grass's credit, they started this in 2022. This was a pretty novel design for that period, when this was much less of a consideration. I think they're 90% of the way there. Obviously, there's room for improvement, but I think the improvements are more recent developments.
Yeah, and I think both, obviously, longer term, are benefiting from some pretty substantial secular tailwinds.
Yeah.
Yeah, go ahead.
Because you're right, Kevin. It's a good point. There's this narrative that companies and tokens should just buy back—they should be doing buybacks, like HYPE and stuff. Most startups should not be using all their revenue to buy their own token. They should be reinvesting in growth, and no startups really do that.
In crypto markets, that doesn't really get priced or valued, right? People want to see the buyback because HYPE has created this meme that the only way for tokens to be valuable is to put 100% into buybacks.
But I do think that if you have the premise that, once there's excess value, it will accrue to the token, people will be more patient. Because you don't know that that's the case, and it might be 10 years in the future, you have to believe this team will honor that promise 10 years from now. You want to see the cash flows now.
If we have structures that people believe in and understand better—whether it's the Grass structure, an ACE structure, or something like that—then people can start paying for and token projects can start being rewarded for reinvesting in growth and showing really awesome growth, which is, to be honest, what Grass has shown.
Yeah, they reinvest in data centers. They cut over $1 million in monthly OpEx last year. So, they're printing cash right now.
Yeah, and I'd also say that raising $65 million from 10 to 20 institutional-type funds, right—raising institutional capital versus whatever it is, 1,000 or 10,000 high-net-worth investors, but also retail—
There is a bit of a tax—and you could argue a pretty large tax—on going the token route. I think Grass went through this recently with a lot of backlash from token holders and node runners, or node operators. There is this balance as a founder: you’re almost paying a tax to have a liquid token, especially at an early stage, because you have to answer to a much wider group of people who all have their own opinions and takes on what you should be doing. Whereas, just in terms of ease and speed, I can understand why you’d try to raise an equity round instead.
Ease, speed, and you get a premium nowadays.
And the headache of dealing with the disgruntled fan base on Twitter.
It’s kind of a problem. I think it’s one of the first times in the history of the last year where there’s definitely a discount to being a token project in the same space as an equity project. You see this across the board. I think Grass is an interesting example, where I think this thing would be raising at a much higher valuation as an equity company than the $400 million valuation it has as a token project.
I view that as an opportunity, because I think it’s an awareness thing. I agree with what you’re saying in general, but I think that applies to the ones that are flirting with both structures.
You kind of need people to believe in tokens—basically, to believe that they won’t get rugged on tokens—or have some standard that people adhere to. Otherwise, these things will keep getting underpriced, especially in bear markets, and then overpriced in bull markets. You get this negative selection where, if you’re a really high-quality project, why would you go through the hassle of doing a token, dealing with angry token holders, and all this crap just to be underpriced versus equity? It has to be at least the same price and then have the bonus of liquidity, I think, for this to make sense for people.
You could argue that being slightly discounted with liquidity is okay. But trading at a third of what you should be, or whatever you want to call it, is too much. It’s a flows thing, too, right? There’s just less capital sloshing around, and the capital that is in there is less long-term-oriented.
The counter is that, in a world where there are fewer quality investments to chase, you’re okay sitting in some that are not slow cooks, but slower cooks than a black bull-type pump.
Equities are getting more liquid, too, though. Sorry, private shares and SAFEs are getting more and more liquid, too. A lot of companies now are having these secondary offerings, even for employees and stuff. It’s kind of a weird one. Tokens are occupying a weird place.
I do think that the liquidity of secondaries in the equity world is overstated. There are constant blackout periods when they’re doing a raise or an employee tender and all this stuff. Then, when you want to trade, you have to pay 5% to the broker.
Oh, you want to sell with fees? Good luck.
Hopefully that’ll get fixed. There are a few people trying to fix that.
Oh, yeah?
More to come on that soon. I don’t think they’ve really spoken about it.
4. AI Infrastructure & The Next Trade
I think, as I was saying before, Vana and Grass are both benefiting from big secular tailwinds. I want to end this with the idea that there’s always a bull market somewhere, even if it’s not necessarily in crypto. I’d love to get an update—maybe start with you, but you guys chime in, too—on the broader AI value-accrual trades and what you’re looking at right now.
The memory trade has pulled back a little bit. It’s had a hell of a run so far, so I think a lot of this is just blow-off-top-type consolidation. There’s a lot of leverage in the system, especially for some of these Korean names. Inference providers and neoclouds are starting to look interesting—or have been interesting—but maybe there are some interesting levels here. I’m curious if there’s anything else you guys are looking at.
I’m long those. I’m long basically the Mag 7, memory, and some of the neoclouds. I think it’s sort of a balanced portfolio, in a sense. They do seem to move together. Right now, the Mag 7—obviously, my Mag 7 has underperformed a lot—are going up while the riskier end of the spectrum is going down.
A lot of these people are very leveraged, especially Koreans. The Korean index market just keeps closing because people get margin-called, and it gaps down to the limit. Speaker 3
Wasn’t there a stat floating around? There was a stat: 1 in 30 adults—
got margin-called.
That’s so dope. Respect.
Retail margin is at an all-time high. They still post about this, too. Maybe I can find it or put it in the show notes.
I think these things will be volatile, but I’m still bullish. When you read the estimates by smart people, especially the stuff SemiAnalysis publishes, there’s some amazing research. Their memory update from April still says they think prices are going to go up 2–3× from here. Supply is not going to catch up to the needs of agentic workflows, so I’m still bullish. It’s a long-term position for me. I had decent entries on them, so I’m not thinking of selling.
The market reacts very strongly to things like Meta announcing that they’re selling compute, and then all the neoclouds go down. I don’t think it’s very rational. They still have tens of billions of dollars of contracts with CoreWeave and others. I think owning land and having power is still going to be incredibly valuable going forward, like with IREN and similar plays in that world.
The open-source thing is another thing that scares people a lot. If the Chinese models get so good, are all these computing investments going to go to waste? I don’t really understand the logic of the bear case there. I guess it’s that OpenAI and Anthropic don’t have pricing power anymore, and therefore they go bust on all these power and data-center arrangements that they’ve made.
If open source gets really good, it’s definitely bad for OpenAI and Anthropic, but it’s really good for the neoclouds and inference providers. All the margin moves to the inference providers because, with open-source models, you serve them yourself. There’s no margin from a model company there. Compute consumers get cheaper prices, but the neoclouds and inference providers can charge a higher margin.
It’s just better for the world in general. More people get access to cheap intelligence, and more stuff gets done. Intelligence is the ultimate Jevons paradox. I don’t think you can ever have enough intelligence.
I’m long for the next few years as this stuff percolates through society. We could be in a local top. Enterprise spending got really ahead of its skis, and there are a lot of people talking about setting limits—$200 a month or whatever it was that Grok and SpaceX were setting. But the proportion of people using AI is still incredibly small. The percentage of people using agentic and multi-agent workflows is incredibly small in the scheme of things.
There’s still a lot of work to do to percolate this stuff through society. I think this enterprise-ROI thing is a bit overstated. If your enterprise isn’t getting ROI from AI, it’s a skill issue in my mind. There’s no workflow, in my opinion, and no industry that cannot be made more efficient with this stuff. Realistically, if you don’t get ROI from AI, you’ll be replaced by someone who does, because there’s no workflow or industry that can’t be made more efficient with it.
That’ll just take time, and, yeah, I’m staying long through it.
Okay, okay.
Go ahead. No, I’d love to hear it.
I think those arguments against the ROI are also just apples to oranges. They’re saying, “We’ve improved efficiency by 5% or 10%, and our token spend has doubled.” It’s like, okay, sure, but 5% or 10% on a very large number and token spend on a very small number actually net out in the end.
I hate when they start applying percentages to 2 different-size numbers that net out in the end. It makes no sense. Brad made some good points about frontier models, and there’s an argument to be made that they increase their edge because of the revenue they’re able to generate, with that capital allowing them to continue being on the cutting edge of training.
Which I think holds true. And then there are going to be enough tasks, or enough decisions, that you have to make that are so valuable that this model can do 99% or 98% of the quality of another model.
1 minus 0.98 to whatever power for the amount of decisions, and that compounds. That's the spread. And so I think there's a big argument to be made that the cost is still cheap. It's not about the cost savings or the incremental cost difference, but about the compounding magnitude of the output and the impact that it makes.
Obviously, some decisions are binary: you can get to a yes answer with a variety of models. There's a routing component where you can route things to shittier models, but I think there are an insane amount of tasks, and everyone's kind of blown away when they switched over to Fable, right?
Granted, you're starting to see some catch-up in distillation from GLM-5.2, but each incremental one that gets released is insane. I think the demand for the frontier will still be strong.
Fable's just so good as well. I don't know if you guys have been using it, but I don't know how I could sell any AI stocks while using Kimi. It's like—and to think that in 6 months that will likely be an order of magnitude cheaper.
If we stay on the same sort of trajectory we're on, we'll have something way better at the frontier. It's just [expletive] crazy to me. Yeah.
Yeah, 100% agree. And I think it dovetails really nicely with that conversation we had with Tommy. What was it, last month, around open source versus closed source? The idea is that the pie is going to get much bigger on both ends of the spectrum, which is bullish—just mega-bullish.
Yeah. The open-source thing is also weird to me. We have AGI, supposedly, and jobs are going to be fewer, but we can't stop distillation. We can't stop the Chinese models from distilling.
Yeah.
No, I agree.
You're saying if we had it, they should be able to stop it?
Yeah, I think so.
Why?
It seems like they should. I just feel like it doesn't seem like—I don't know. I'm sure it's really complicated. But at a high level, the activity doesn't feel like it should be that hard to identify, right? These millions of pings on these questions are obviously coming from very clustered sources. They're using IP masking and everything like that, but it just seems like a problem that could be solved by an AGI, right?
I don't want these models to get distilled anymore. And when you talk to GLM, it thinks it's Claude a bunch of times. You can see all the fingerprints of Claude, and it's extremely obvious. There's an argument as to how much of it is distillation versus genuine innovation, because I think there are a lot of people—and I don't know the answer to that—who I really respect who are very complimentary of the actual innovation from the Chinese models.
But still, there's clearly a lot of distillation going on. I wonder if they'll be able to stop it. They're probably investing a lot in stopping that at this point, given how existential it is for them, at least. But open source getting really good is bullish for everyone, right? Everyone who isn't an Anthropic or OpenAI shareholder, it would seem.
Yeah, I think you can make the argument that the better open source gets and the cheaper it gets, the more demand there is for using AI more broadly across more industries. In which case, I think the demand for frontier AI to be the leader, or the orchestrator of sorts, of the new things you want to build or create also increases.
Yeah, that's why there are arguments for why China should ban its open-source models—keep its open-source models internal or domestic. It's too beneficial to allow everyone to have access to them.
I think if open source ends up mattering, though, Anthropic and OpenAI will also release some open-source models, right? OpenAI did a while ago, and it was pretty good at the time. But if you're at the frontier, you can do whatever, right? You have the best models. You can choose to open-source a slightly older version if it ends up being really important.
Being at the frontier is all that matters, I think. I don't think anyone uses Chinese closed models, honestly. If they close the models, even if they are at the frontier, they would struggle to get usage because people just really don't like—even with the open-source Chinese models, people are like, "Oh, I don't want to use Chinese models." You can just run them on your own computer.
It's kind of weird that there's still this argument going around, but I think if they were closed, there'd be very little chance that anyone uses them. So I don't think they could do that. But they could probably go open source with a more restrictive license that forces inference providers and so forth to pay them some share—some revenue share—if they're serving it at scale.
I think that could make sense, but going closed is—you basically get no advantage to being closed unless you're at the frontier, I think. It's interesting, too, because it's in everyone's interest, other than the 2 frontier labs, that open source is as close as possible to the frontier, right? Which is why you see Nvidia plow a bunch of money into open source and their Nemotron models and stuff like that. You're seeing the Mag 7 plow a bunch of money into them. If that happens, it's good for everyone.
But it's harder than—if you asked me 3 years ago, I would never have said that Meta would have failed this badly, or that Microsoft would have failed this badly at AI. They just have Copilot and Llama, or whatever the new Meta model is. They just aren't that good, and it's kind of surprising to me that it would have been that hard for them.
5. Why AI Is Still A Long-Term Bull Market
Even with frontier model providers like OpenAI and Anthropic, if they were to verticalize more quickly and get a larger, let's say, share or exposure to inference providers, or actually own their own compute, maybe that's some of these deals they've struck to get that type of exposure. There's a certain point where there's probably a tipping point where open-sourcing some of their models can actually make sense because they have a way to capture the value still, and potentially capture more value. But it's a little game-theory-esque.
I was going to say it's scary to think what markets would look like if the AI thing didn't happen.
Crypto would probably be a lot higher.
Crypto would be higher.
Because GDP would be in the [expletive], and we'd be printing. Oh, yeah.
Crypto would be the only game in town. Crypto would be a lot higher.
Yeah.
Meme coins would have FIFA ads. [Laughter]
I was thinking about this the other day. If this was 2021 or something, wouldn't FIFA just have NFTs everywhere?
It's in America. NFTs were so hot. I'm so glad NFTs died so that they didn't get stamped all over the World Cup.
Anyway, I think—yeah.
Anything else you want to cover?
No, that's a lot of great stuff. I think we got plenty. We can dive into more of this next time, because obviously these are all themes that we're going to continue to track. So I think that's a good place to end it. I really appreciate you guys jumping on and sharing your insights. Until next time.
Yep, see everyone.
See you later.