Interview with Founder of a Venture Capital Fund Focused on Conversational AI: Forget the False Narrative of 100x Returns—Why I’m Bullish on VVV, GRASS, and NEAR
TechFlow Selected TechFlow Selected
Interview with Founder of a Venture Capital Fund Focused on Conversational AI: Forget the False Narrative of 100x Returns—Why I’m Bullish on VVV, GRASS, and NEAR
AI has increased data volumes by a factor of one hundred, resulting in significant valuation mismatches for Venice, Grass, and NEAR—what truly matters in assessing crypto assets is whether token holders can genuinely capture the business value.
Navigating Web3 tides with focused insightsCompiled & Translated by TechFlow
Guest: Austin Barack, Founder of Relayer Capital (a digital asset fund focused on AI)
Host: Andy
Podcast Source: The Rollup
Original Title: Austin Barack: My AI Bull Thesis (...And What I'm Holding)
Air Date: May 23, 2026
Key Takeaways
This episode of AI Supercycle features Austin Barack, founder of Relayer Capital, discussing Venice, Grass, NEAR, Akash, and the broader Crypto x AI asset framework. Austin’s core thesis is that AI is scaling user data volumes to levels unimaginable for prior internet products—making privacy-preserving AI, data supply, inference compute, decentralized training, and Agent infrastructure critical verticals. He argues Venice and Grass exhibit clear misalignments between revenue growth, user growth, and valuation, while NEAR’s positioning in cross-chain Intents and Agent infrastructure remains underappreciated. For the broader crypto market, Austin emphasizes that investors should assess tokens through the lens of “net token value flow,” rather than mechanically focusing on buyback-and-burn mechanisms—ultimately asking whether token holders meaningfully capture the value generated by the business.
Highlights of Key Insights
Venice and the Real Value of Privacy AI
- "Privacy matters more in AI than in any other context—because you’re sharing health data, financial data, connecting all your files, and revealing your entire life in ways never before possible."
- "This isn’t 10x more data than social media—it’s 100x more."
- "What makes Venice truly compelling isn’t just enabling private AI use—it’s doing so without sacrificing user experience, and even enhancing it."
- "Tokens can become a critically important layer that significantly improves the experience—but most users don’t need to understand tokens to find the product useful."
VVV, DM, and Venice’s Economic Model
- "The role of DM is simple: each DM token entitles you to $1 worth of free inference compute per day on Venice—think of it as a perpetual right, yielding $365 annually."
- "Unused daily quota expires—no rollover. If you use only $0.50 one day, you still start fresh with $1 the next day."
- "If all DM tokens were locked and fully utilized for inference, Venice’s maximum daily cost would be $38,000—roughly $10 million annually—and this cap is hard-bounded."
- "I believe DM should be valued more like corporate debt—not discounted aggressively with excessively high rates."
Grass and AI’s Data Demand
- "Grass collects curated datasets and sells them to cutting-edge AI labs training new models."
- "This isn’t random web crawling—it’s highly specialized, extremely specific, and rigorously high-quality data."
- "Model development budgets are enormous—and Grass benefits directly from this trend: the bigger the investment, the greater the demand for data."
- "Based on recently disclosed figures, the project’s ARR is approximately $50 million. Its current valuation stands at ~$400 million. Valuing such rapid-growth projects at just 5x revenue strikes me as entirely unjustified."
NEAR, Akash, and the AI Stack
- "NEAR Intents are highly practical—and arguably among the best cross-chain swap experiences available today. They also play a pivotal role in the Agent ecosystem."
- "I think NEAR excels on the Intents side. They’re also advancing privacy-focused intents and other AI-adjacent primitives—making it one of the few L1s that has carved out a truly distinctive position."
- "Akash started as a decentralized CPU marketplace, then pivoted to GPUs."
- "My primary areas of focus include: decentralized training, inference and compute markets, Agent infrastructure, data, and consumer-facing model applications."
Token Value Capture and Market Differentiation
- "Hyperliquid first succeeded as a robust business model—so people liked its token, and buybacks were merely one mechanism to distribute value to holders. If the underlying business weren’t sound, buybacks alone wouldn’t lift the token price."
- "The real question isn’t what the mechanism is called—it’s whether token holders capture maximum value from what you’ve built."
- "Each project and mechanism demands case-by-case analysis. But the central question remains: do token holders benefit from the value the system generates?"
- "Investors now select from a smaller, higher-quality pool of projects. Capital flows are increasingly concentrating around Venice, HYPE, Grass, AERO, NEAR, and Zcash."
- "For investors targeting 3x–10x—or even 5x–10x returns—this moment offers unusually favorable odds. You might still hit 100x, but right now there’s a cohort of projects doing genuinely interesting things—and those are precisely the assets I’m watching and investing in."
Venice Privacy Overview
Host Andy: Not long ago, I tried Venice for the first time. I typed into Venice: “Is this really private?” It replied: “Yes, inference is private,” then explained further. I responded: “That’s so cool.” Instantly, it said: “Yes, it really is— isn’t it? With Venice, you can…”
So the first-time Venice experience delivers a striking realization: everything you’ve ever typed into mainstream AI services—even if not publicly visible—flows to large centralized providers. Your most intimate journal entries, trade secrets, strategic plans—all handed over.
From a top-down perspective—market structure, investment logic, founding team—how do you view privacy AI and Venice?
Austin:
Venice is fascinating because it’s evolved through many distinct phases. I first encountered the project in January last year—when I was closely tracking Virtuals and aixbt. Venice’s early airdrop allocated a significant share to holders of tokens within those ecosystems, which is how I first discovered it.
Even then, it was already an intriguing product. Remarkably, though only about 16 months have passed, AI back then was nowhere near as ubiquitous—or as embedded in daily life—as it is today. Since then, whether Claude, ChatGPT, or others, AI initially replaced Google Search: people said, “I no longer search Google—I ask an LLM directly.” Now it’s progressed to creation, task execution, and even functioning as a full team of Agents working alongside you.
AI Data Volume Is 100x Higher Than Before
Austin:
I think people are gradually realizing privacy matters more in AI than in any other context—because you’re sharing health data, financial data, connecting all your files, and revealing your entire life in unprecedented ways.
Historically, privacy discussions centered on social media: Is my profile public or private? Does Facebook know too much about me? But AI isn’t just 10x more data—it’s 100x more.
What makes Venice truly compelling isn’t just enabling private AI use—it’s doing so without sacrificing user experience, and even enhancing it. You’re not locked into a single model. With ChatGPT, you’re tied to OpenAI’s model upgrades; with Anthropic, to their evolving models; with Gemini or open-source models, you face similar constraints.
In Venice, you can pick the optimal model for each task—or choose which models to use. Customization is high. First and foremost, they’ve built an exceptional consumer product—and most users don’t even know what a token is.
Tokens add an interesting extra layer. I’m bullish on what they’re building. The key insight is that crypto-native consumer products will likely evolve toward this paradigm: tokens can be a vital component that dramatically enhances experience—but most users won’t need to understand tokens to find the product valuable.
Host Andy: This indeed resembles a breakthrough consumer-product model: crypto-powered underneath, yet invisible to users. Yet it introduces an intriguing token structure. Some compare it to Luna: stake VVV to receive DM tokens, forming a quasi-debt structure via inference quotas.
3 Million Users
Host Andy: So how should we interpret the current VVV and DM token flywheel at Venice? And could you elaborate on Venice’s revenue side—they do conduct buybacks, though modestly. How exactly do these two tokens function—and why isn’t this like Luna?
Austin:
They’ve just announced 3 million users—and growth is accelerating rapidly. Roughly 1 million users joined in the past three months, whereas the prior 1 million took about seven months. Growth continues to compound.
VVV and DM Token Flywheel
Austin:
They have two tokens. First is VVV—the protocol’s revenue is used to burn VVV. Users may also stake VVV to access free membership. But the most interesting mechanism is staking and locking VVV to mint DM tokens. You can also buy DM on open markets—but the core path is staking VVV to mint DM.
The role of DM is simple: each DM token entitles you to $1 worth of free inference compute per day on Venice—think of it as a perpetual right, yielding $365 annually.
But unused daily quota expires—no rollover. If you use only $0.50 one day, you still start fresh with $1 the next day. I see this as an elegant, almost loss-leader customer-acquisition tool—distinct from Luna, which spiraled into extreme issuance, ballooning stablecoin supply into the billions or tens of billions of dollars. Venice, by contrast, sets explicit cost boundaries.
Currently, the number of DM tokens mintable per Venice token declines as total circulating DM increases—effectively imposing a hard cap of ~38,000 DM. At present, if all DM were locked and fully used for inference, Venice’s maximum daily cost would be $38,000—~$10 million annually—and this cap is immutable.
Currently, roughly 10,000 DM tokens are used daily for inference—implying an annualized cost of ~$3.5 million. This cost is offset by their business revenue: Pro and Premium subscriptions priced from $18–$68/month (and higher), plus users purchasing tokens or additional credits to access models.
Notably, their daily token usage has surged from initial billions to ~70 billion recently—a ~15x increase over recent months. So unlike Luna, Venice operates with a defined maximum potential cost—and DM users simultaneously engage subscription services. If users require >$1/day per token, they purchase additional credits. Costs are easily covered—and already substantially exceeded by revenue.
DM Should Be Priced Like Corporate Debt
Austin:
Another compelling aspect of DM is its guarantee of future compute access rights. Markets currently discount it at ~20%, pricing it around $1,800.
I believe this asset should be priced more like corporate debt—using 8%–12% discount rates. At 10%, its fair value approximates $3,650. For context, when I first tracked it, the price hovered near $200.
Host Andy: I remember thinking—how can a $365/year asset trade at just $200 unless the market doubts Venice’s ability to sustain the mechanism?
Austin:
Exactly. At that price point, it felt like a no-brainer investment opportunity. Even now, I still see meaningful upside.
Yet zooming beyond DM to Venice’s overall economics reveals staggering numbers—and its growth pattern diverges sharply from most crypto-native projects. It mirrors AI-specific growth trajectories—precisely why it’s so compelling.
Is $20 Venice Still Undervalued?
Host Andy: So you firmly believe VVV trades near $20 today. Do you still consider its $1.5B–$2B valuation range markedly undervalued?
Austin:
Yes. I first bought at ~$2.50 in January—when daily token volume was just billions. Today it’s ~15x higher.
Back then, daily token volume was only in the billions—now it’s ~15x greater. User count grew from 1.5M to 3M. By my estimate, revenue is at least triple what it was then.
Venice currently trades at ~20–30x annual revenue—and it’s still growing monthly at ~20%. Viewed this way, its valuation remains extremely low. You could even compare it to OpenRouter: similar valuations, yet Venice likely boasts higher revenue and faster growth.
The key distinction is Venice owns direct customers. It’s not pure backend infrastructure—it’s a platform users actively engage with daily. Personally, Venice is now my sole interface for using AI.
So I believe its potential remains vast. Of course, this is strictly my personal view—not investment advice.
How Grass Makes Money
Host Andy: I’m less familiar with Grass. You’ve referenced it multiple times—it appears poised for rapid growth. Though its price may have pulled back today, I hear its ARR exceeds $50M, accelerating into triple-digit growth. Could you briefly outline Grass’s core revenue model—how it monetizes, and why it’s so compelling?
Austin:
Grass collects curated datasets and sells them to cutting-edge AI labs training new models. These labs generate models at breakneck speed—but require massive amounts of data to do so. And this isn’t random web crawling—it’s highly specialized, extremely specific, and rigorously high-quality data.
That’s Grass’s role: as model-development budgets balloon, Grass rides this wave. More investment → greater data demand.
Grass’s Triple-Digit Growth
Austin:
The Grass team has been building for years. I recall one quarter last year where they did ~$3M in revenue; by year-end, they hit $12M–$13M quarterly. My estimate is they’re growing even faster now. They’ll host a token-holder call in the next 4–6 weeks—we’ll get updated details then.
But this is unequivocally a triple-digit-growth project. Based on recent disclosures, ARR sits at ~$50M—though I suspect it’s nearing $80M now. Its current valuation is ~$400M. So pricing a hyper-growth project at just 5x revenue feels completely unjustified—it’s a prime candidate for repricing.
Host Andy: Any operational ties between Grass and Venice?
Austin:
None currently. Venice typically doesn’t build its own models—so no relationship exists today. Who knows about the future? But I view them as complementary sides of the same equation: one addresses how you use AI—and privately use it—the other addresses how models are built in the first place. Grass and Venice handle these respective dimensions.
Is Grass’s $400M Valuation Too Cheap?
Host Andy: So Grass trades at ~5x revenue. In crypto, some assets trade at 20x, 30x, 40x, or even 50x revenue. Does the ~$400M range feel like a no-brainer?
Austin:
Yes. Crucially, crypto does host other low-multiple assets—but they lack growth. People enter crypto to invest in growth.
So many low-multiple cases hold little water—because there’s no cash flow. Grass, however, is arguably the best example of explosive growth in the space. That alone warrants attention—not to mention, in my view, it remains quite cheap.
NEAR’s Cross-Chain Swap
Host Andy: Do you have a thesis on NEAR? Are you tracking it?
Austin:
I’ve consistently tracked NEAR. Even ignoring AI components, NEAR is fascinating—it underpins substantial cross-chain swaps. Last October–November, NEAR gained traction as infrastructure for Zcash inflows/outflows.
NEAR Intents are highly practical—and arguably among the best cross-chain swap experiences available today. They also play a pivotal role in the Agent ecosystem. To me, NEAR ranks among the most suitable infrastructures for cross-chain swaps—avoiding dependencies plaguing other projects.
They’re growing rapidly. If you’re an L1 today, you likely need to excel in one of three directions: deliver vertically integrated app experiences, be 10x better at one thing, or dominate a specific application category.
I think NEAR excels on the Intents side. They’re also advancing privacy intents and other AI-adjacent primitives—making it one of the few L1s with a truly unique positioning.
This reminds me of NBA player archetypes. Today’s market hosts many new L1/L2 projects—like promising rookies. Over time, some become superstars; others fade. Then there are “role players”—excelling brilliantly in their niche. Think Lu Dort or Alex Caruso of OKC.
NEAR feels like that kind of role player. It’s not LeBron James—but it’s vital because it dominates its domain.
Akash GPU Market Update
Host Andy: Another persistently undervalued project—Robbie constantly highlights—is Akash. Unfortunately, he’s absent today. Akash entered distributed inference, distributed models, and decentralized training very early—right?
This sounds like Crypto AI’s first narrative wave. Then came meme-driven, faux-Agent projects. Now we appear to be entering the next phase—decentralized inference and model training—with far more mature products. Have you reviewed Akash’s latest developments? Any investment views?
Austin:
I’ve followed Akash—they began in decentralized CPU markets, then pivoted to GPUs. You can actually track how much data flows through OpenRouter. A notable portion routes through Akash ML—which is impressive. And this data is public—anyone can verify it.
That said, Akash isn’t among my closest-tracked projects. Still, it’s exciting to see a long-standing, iterative team finally achieve genuine product-market fit—and watch that fit accelerate.
AI Stack Breakdown
Host Andy: There’s a project called Gitlab on Base—tiny market cap, yet strong daily token production. Base is now seeing a wave of highly speculative AI tokens, with many micro-verticals needing unpacking.
Broadly speaking: within the AI stack, which layers are best positioned to scale massively once integrated with blockchain? We’ve seen Venice offering private inference and censorship-resistant ChatGPT; NEAR as Agent-market infrastructure; Akash with Akash ML; Grass focused on datasets.
Which key AI-stack segments are most likely to be displaced—or optimally deployed—onchain?
Austin:
First, privacy contexts—including private and censorship-resistant LLM usage. Second, data collection for model training—that’s Grass’s domain.
Third, inference compute and compute markets—you mentioned Akash. Other inference markets are emerging too. There’s also a project built around DM offering additional services—letting users sell idle compute—called AnC. It’s an intriguing project I’ve watched closely, especially its integrations with Venice and DM—though it hasn’t launched a token yet.
Another crucial frontier is decentralized model training. The challenge lies in building open-source models while retaining ownership and monetization via private weights. Several teams are exploring this: Pluralis is among the most interesting; Nous Research does compelling work around Hermes; Prime Intellect and others are also active here.
So my primary focus areas are: decentralized training, inference and compute markets, Agent infrastructure, data, and consumer-facing model applications.
Net Token Value Flow Framework
Host Andy: Recently you’ve emphasized another idea: we need new frameworks to understand token models and economics. You’ve strongly backed projects like Aerodrome and Hyperliquid.
To close, stepping beyond AI: how do you view “net token value flow”—analyzing crypto assets via credit (revenue) and debit (expenses), using plus-minus accounting? How is the industry’s approach to token economics evolving? What’s your current mental model? Do you agree investors should analyze net token value flow like a balance sheet?
Austin:
I see multiple valid perspectives—and this isn’t a one-size-fits-all matter.
Let’s start with high-level mechanisms like buyback-and-burn. Hyperliquid popularized this—people say, “Look how well Hyperliquid does buybacks.” But for every Hyperliquid, nine other tokens adopt identical buyback-and-burn schemes—and perform poorly.
What’s the lesson? Hyperliquid first succeeded as a robust business model—so people liked its token, and buybacks were merely one channel to distribute value. If the underlying business weren’t sound, buybacks alone wouldn’t lift the token price.
This is the first common misconception.
The second question is whether you’re truly creating value for token holders. Whether via buyback-and-burn, buyback-and-distribute, reinvestment, or treasury accumulation for balance-sheet flexibility—the core issue remains: do token holders capture maximum value from what you’ve built?
Hyperliquid does this. Aerodrome does this. With Grass, many wish for more buybacks—but clearly, all contracts are with the Foundation, and all revenue flows into the Foundation’s bank account—assets ultimately controlled by token holders.
So many analytical paths exist.
Buyback-and-Burn Works Only in Specific Cases
Austin:
Then there’s token liquidity. Hyperliquid theoretically has a monthly unlock cap—but actual unlocks may total only 200K–300K tokens. Meanwhile, buying pressure from ETFs, DAT, and the assistance fund vastly exceeds that. Naturally, buyers outnumber sellers.
Take Aerodrome: lock AERO as veAERO, and post-July Ethereum mainnet expansion, veAERO becomes sAERO. Holders earn all platform revenue—and can direct token emissions toward liquidity pools most in need of liquidity and highest revenue generation.
Some argue that if emission value exceeds revenue in a given cycle, it’s “net negative.” I consider this view entirely misguided.
The correct analysis asks: how much revenue did the system generate this cycle? How much token supply increased—but wasn’t sold? For instance, Aerodrome recently rebranded one mechanism as the Momentum Fund—functionally resembling continuous Foundation-led buybacks. Many who earn AERO choose to lock it as veAERO to earn more. Others simply hold, confident in the token’s future.
From this lens, the amount of tokens flowing to open markets each cycle—i.e., weekly—is far lower than the platform’s concurrent revenue generation.
Combine this with recent launches—Atlas, Aura, and others—and Aerodrome’s revenue has surged meaningfully. Here, “revenue” means earnings accruing to token holders—now clearly exceeding emission value.
So each project and mechanism demands individual analysis. But the core question remains: do token holders benefit from the value the system generates? That’s the analytical anchor—building from there.
Two Emerging Groups in Digital Assets
Host Andy: I think the entire industry is converging on similar mental models—albeit highly nuanced ones. Two groups now seem to be crystallizing: one comprises revenue-generating, fundamentally sound companies; the other includes narrative-driven, highly specialized—but technically potent—projects like Zcash, Venice, and NEAR, focused on AI privacy. Plus there are purely onchain-business projects—while the middle ground seems relatively quiet.
Austin:
I agree. What makes this market fascinating is that the set of truly compelling tokens has shrunk. With clearer distinctions between authentic projects and hype-driven ones, perhaps only 10–20 tokens now boast exceptionally strong fundamentals.
Hence, these tokens clearly outperform the market. For the first time in a long while, investors can select from a smaller, higher-quality pool. Capital flows are concentrating around Venice, HYPE, Grass, AERO, NEAR, and Zcash.
Zcash is another privacy-native project. Some worry Bitcoin may increasingly reflect Michael Saylor’s influence (a separate topic), while Zcash embodies Bitcoin’s original ethos—and shares structural similarities.
Zcash lacks revenue today, yet remains an intriguing asset: higher prices amplify its utility—increasing likelihood of consolidation, stronger consensus, and deeper community value.
So we’re in a uniquely interesting phase: selecting the right tokens has become easier—requiring sharper diligence to distinguish authentic projects from hollow hype.
For investors targeting 3x–10x—or even 5x–10x returns—this moment offers unusually favorable odds. You might still hit 100x, but right now there’s a cohort of projects doing genuinely interesting things—and those are precisely the assets I’m watching and investing in.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
