Yan Liberman

Yan Liberman@YanLiberman · 21 天前赛道影响

RT @momilio: Spent some time building an Onchain Options Dashboard to track all major analytics Will drop my thesis sometime in the futur…

090184

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Yan Liberman@YanLiberman · 2026/07/30观点输出

RT @Apptronik: Huge congrats to the @GoogleDeepMind team on launching Gemini Robotics 2! Watching Apollo 2 use advanced reasoning to naviga…

0300209

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Yan Liberman@YanLiberman · 2026/07/29赛道影响

RT @ZeMariaMacedo: 1/ The most important story in tech no one's paying attention to isn't happening in SF but in Ukraine: Ukraine has achi…

030135

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Yan Liberman@YanLiberman · 2026/07/21赛道影响

RT @Shaughnessy119: Renting vs Owning Your AI Intelligence The American dream was owning a home. I think the future version is owning your…

0240169

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Yan Liberman@YanLiberman · 2026/07/16观点输出

RT @Delphi_Digital: A new episode of Hivemind is out now! This week we discuss the outlook for crypto and equities, token and equity value…

080118

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Yan Liberman@YanLiberman · 2026/07/16市场影响

Generally the difference in buyback multiples is based on durability of the revenue and growth expectations. People are much more confident in HL revenues persisting and growing (esp with the expansion of non-crypto assets as a revenue driver) With crypto you also end up having some project specific reasons for discounted multiples

2021577

AI：偏平台经营与生态扩张，强调交易平台和应用入口的长期位置。

Yan Liberman@YanLiberman · 2026/07/14观点输出

RT @Shaughnessy119: Extremely clear analysis on solana:Grass7B4RdKfBCjTKgSqnXkqjwiGvQyFbuSCUJr3XXjs by my partner @YanLiberman "The seaso…

020200

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Yan Liberman@YanLiberman · 2026/07/14赛道影响

RT @ZeMariaMacedo: Over the last year building our fund-of-funds, I've spoken to 500+ emerging managers across AI and deep tech What I fou…

090186

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Yan Liberman@YanLiberman · 2026/07/14观点输出

Supply and revenue aren't the same thing. Revenue comes from labs buying datasets, not from node count. It grows by selling more to existing labs and onboarding new ones. And not all nodes are equally valuable, or even necessary. The 150k figure isn't the network shrinking, it's the network routing traffic to the nodes that are actually useful: stable connection, right geography, right device. In oversupplied regions the marginal node is pure redundancy. You only need a strategic fraction of supply to keep scraping what's in demand, and they already hold hundreds of petabytes of collected data that doesn't vanish if marginal nodes unplug. On the differentiation question, I think it's the wrong comparison. Bright Data sells bandwidth. Grass sells finished datasets. Grass scrapes at petabyte scale, runs its own GPUs to filter and classify, and delivers the ~5% that's actually high signal, for roughly what the raw data would have cost plus compute. Labs buy that because the alternative is burning their own GPUs and engineers on it. That's a different product, not a bigger proxy pool. (Also, I think you mean Bright Data, not Brightcove. Worth noting their 100M+ IPs aren't opt-in users. Per their own docs those come mostly from an SDK embedded in partner apps, where the developer consents and the end user largely doesn't know.)

005320

AI：偏平台经营与生态扩张，强调交易平台和应用入口的长期位置。

Yan Liberman@YanLiberman · 2026/07/14政策影响

$Grass's July token holder call has been largely misunderstood because few realize that revenue is seasonal. With that in mind, the growth has been very strong. TLDR -2026 expected revenue of ~$70m -Adjusted FDV at $0.36 is ~$250m, excluding Foundation-controlled treasury supply -That's ~3.6x 2026 expected revenue -H1 '26 revenue grew ~6x over H1 '25 -FY26 expected revenue is ~4x FY25 actuals -Ongoing opex is ~$2.5m/month, implying $40m of operating profit in 2026 The seasonality is the whole misunderstanding. People are comparing H2 2025 ($14.3m) to H1 2026 ($17m), seeing a small step up, and concluding growth stalled. That's the wrong comparison. Frontier labs make their largest data purchase decisions around training cycles, and most of those cycles get contracted in the second half of the year. Compare like for like, H1 '25 to H1 '26, and revenue grew ~6x. Ongoing cash expenses are ~$2.5m a month, mostly infrastructure. On $70m of 2026 revenue, that's ~$30m of annualized cash opex and roughly ~$40m of implied operating profit. Margins should expand as revenue grows because the cost base is largely fixed. They own the infrastructure rather than renting it. "Why don't they return that money to token holders?" For the same reason that no start-up returns capital to early investors. It should be reinvested for growth and profitability. In 2025 they made compute/storage investments that reduced opex by over $1m per month. In addition to the $70m they expect to earn via training data in 2026, they'll also be launching products in the LCR space this summer. I expanded on why LCR is interesting in a previous write up, but the current leading players in the category, Exa and Parallel, just raised at $2.2b and $2b respectively. My summary above is partially sourced from the recap of the call they shared: https://t.co/JylhTnB7oK Disclosure: Delphi Ventures has held a position in GRASS for approximately two years and holds it as of publication. This is not investment advice and the opinions expressed are my own.

7125115.4K

AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。