Nikshep argues $VVV trades at the cheapest multiple in AI while being the only profitable one, capturing surplus through an automated buy-and-burn mechanism that's already destroyed ~33.8m tokens (42% of remaining supply). Venice's $70m+ ARR grows profitably with subscription burns scaled by tier ($2–$10) firing ~1,250 times daily, and the Dragonfly warrant—denominated in the asset Venice plans to incinerate—signals institutional confidence in the burn thesis rather than betrayal. The next inflection arrives when Venice launches its "minds" marketplace (already flagged in production code), moving from selling inference tokens to outcome-priced agents that crypto enables through identity, payment, and ownership primitives incumbents lack.

Yan argues Grass is a real AI data infrastructure business hiding inside a token, not a typical DePIN project. With 8M+ users sharing internet connections, the network generated $2.75M revenue in Q2 2025, accelerating to ~$50M ARR by Q4 with 197% QoQ growth (verified under NDA by Messari and Grayscale), positioning it to capture share in a $1B web scraping market where competitors like Bright Data disclose >$300M ARR. The token captures all value since there's no equity company above it—a structural advantage the market has overlooked due to thin public information rather than weak fundamentals.

Mike Zajko argues GRASS has achieved a $50M annualized revenue run rate as of Q4 with QoQ growth accelerating from 56% to 197%, verified independently by Messari and Grayscale, by monetizing idle bandwidth from 8M nodes to sell cleaned web data to AI labs at scale. The Foundation structure ensures revenue flows to token holders rather than the operating entity, comparable to Jito's model, while the company has processed 250 petabytes of multimodal data—roughly the entire indexed web—in under 24 months, creating a defensible moat through its filtering and processing infrastructure that frontier labs require for model training.

Jeff Dorman argues Galaxy Digital operates as two distinct businesses—a crypto financial services arm and the Helios data center in West Texas—but investors treat it as one confused story. Galaxy acquired Helios for $65M in early 2023 as a distressed Bitcoin mining site, then signed a 15-year, $4.5B HPC/AI hosting deal with CoreWeave covering 526 MW of its approved 800 MW capacity, positioning it to generate over $1B annual revenue at ~90% lease-level EBITDA margins. A spin-off could unlock value by letting the data center business trade on infrastructure multiples rather than depressed crypto multiples, making Galaxy better positioned to capture growth from stablecoins, DeFi, RWA tokenization, and crypto-AI convergence.

Lord_Wette's thesis is that the AI buildout's scarce asset is not land or GPUs but bankable time-to-power—approved, financed, deliverable capacity. Galaxy's Helios campus has moved furthest along this curve: CoreWeave committed to 800 MW with $1.4B project financing and 80% loan-to-cost, and ERCOT approval for an additional 830 MW creates a catalyst for multi-tenant, hyperscaler-grade validation. Unlike pure optionality plays, GLXY is already contracted and financed execution, with crypto optionality as a secondary engine if BTC/ETH rally.

Nillion's $NIL token has shifted from optional to required: it's the currency for compute credits on the Blind Computer, the stake securing verifiers, and the settlement layer for partnerships like the Vodafone fraud-detection tower project—all scaling with usage rather than linear user growth. The critical risk is that despite rising demand, supply remains net-inflationary at 0.5% yearly minting plus ~2% monthly unlocks through 2029, with no verified on-chain burn mechanism yet; the incoming tokenomics redesign must deliver credible usage-driven burn to flip the math from "usage grows but so does supply" to structural re-rating.

G.W. Jackston argues DIEM, Venice's perpetual $1/day inference credit token, trades at a 410% premium because the market expects the obligation to hold for decades. The real value emerges if Venice builds a monetization stack around DIEM—DIEM lending marketplaces, DeFi collateral integration, premium tiers, and capacity expansion—that generates 95% of revenue without touching the core promise, potentially justifying $50-70 per token versus today's $14.56.

Yan argues Venice's $200M annualized subscription and API revenue is underappreciated, with latest weekly subscriber additions annualizing to $268M run rate. At current trajectory, 12-month forward ARR reaches $260M, compressing multiples to 2.5x revenue versus 15-25x for comparable AI-infrastructure companies. Beyond subscription, VVV's token mechanics—staking, DIEM minting, buy-and-burn, and lock-up demand—capture broader value as revenue growth outpaces the token's market cap expansion.

David argues Cerebras' IPO signals AI's structural shift from training to inference, where demand scales with compute rather than users. Venice's dual-token ecosystem—VVV for staking and platform access, DIEM for direct API credit exposure—is positioned to capture this market, which JP Morgan sizes at 10 to 50 times training's scale. POD, the token behind Dolphin's distributed inference network, offers complementary exposure through its buyback mechanism and staker allocations.

G.W. argues Venice.ai represents crypto's next primitive by offering private, uncensored AI inference requiring no crypto literacy—a grandmother can use it for basic queries. VVV functions as an access key; staking it earns pro-rata API capacity and mints DIEM, a $1/day inference credit token enabling a rental marketplace where idle capacity becomes productive capital, eventually attracting external DeFi protocols and positioning AI agents as blockchain's next billion users.

Nikshep argues Venice's economic structure inverts the AI incumbents' cap-table trap. While OpenAI projects $85B in losses by 2028 and needs $200–280B annual revenue by 2030, open-source models (GLM-5.1, Kimi K2.6, DeepSeek V4-Pro) now match frontier performance at 5-15x cheaper pricing for the 80% of workloads already saturated to "good enough." Venice's zero training costs, token burn mechanics (42% of genesis supply destroyed), and agent-native architecture position it to capture inference demand that agents structurally cannot route through surveillance-based labs requiring KYC, with the agent economy projected at $3-5 trillion by 2030.

Kevin compares VeniceAI's $VVV token valuation against similar infrastructure plays, finding it expensive on most metrics. OpenRouter handles 60x VVV's inference volume at 2x the valuation, while TogetherAI does 45x the volume at 11x the valuation. Inferring Venice's ~$8.4m ARR from recent token burns yields an 80x revenue multiple versus 26x for OpenRouter and 7.5x for TogetherAI, though this depends on unconfirmed burn rate assumptions.

Nikshep argues NEAR is positioned as the coordination layer for the emerging agent economy, citing Anthropic's advances with Claude Opus 4.6 enabling autonomous agents to sustain long-horizon tasks and execute 100+ tool calls across teams of subagents—capabilities that require a blockchain to securely coordinate AI agents at scale.

Bull pitch on NEAR at $1.28 / $1.67B mcap, ~94% off ATH. The setup nobody is pricing in — vesting fully completed Oct 12 2025 (no more cliff unlocks; the 4-year supply overhang is gone), inflation halved 5%→2.5% Oct 30 2025 via protocol upgrade v81, 70% of fees burn permanently (with sufficient activity NEAR is structurally net deflationary), House of Stake/veNEAR governance went live. **Founder asymmetry**: Illia Polosukhin is one of the eight co-authors of *Attention Is All You Need* — the Transformer paper that powers GPT-4/Claude/Gemini/Llama. Co-founder Alex Skidanov was Engineer #1 at MemSQL, a two-time ICPC World Finals medalist, designed the only sharded distributed DB that worked at scale. The market is currently valuing their company at less than the seed-round valuation of half the AI agent startups in San Francisco. **Real thesis: agents can't use Visa.** When autonomous agents replace humans as users, the entire payment stack breaks — weekend bank hours, KYC for every counterparty, days-to-settle, not programmable. NEAR has shipped more agent-native infrastructure than any L1 competitor: - **Nightshade 2.0 sharding** — 600ms blocks, 1.2s finality, $0.0019 avg fee, benchmarked at 1M+ TPS across 70 shards. - **Chain Signatures** — one NEAR account derives addresses on Bitcoin/Ethereum/Solana/Cosmos/XRP/Aptos/Sui via MPC threshold-signing. Native multichain control from a single account. No wrapped tokens, no bridge honeypots. - **OmniBridge** — settlement minutes vs hours. - **NEAR Intents** — $3M→$13B cumulative cross-chain volume in 2025 (a 200,000%+ jump). Fee switch now active. Ledger, Sui, Starknet integrated. - **Confidential Intents** (Feb 2026) — TEE-isolated private shard parallel to mainnet. No client-side ZK (UX killer for every privacy chain). MEV protection. Selective compliance disclosure. - **IronClaw** — open-source verifiable agent runtime in encrypted TEE. WASM sandbox per tool, AES-256-GCM credential vault, multi-LLM backend, MCP plugin support. **Catalysts**: Bitwise + Grayscale spot ETF filings (Grayscale to convert GTAO Trust on NYSE Arca with Coinbase Custody), NVIDIA Inception membership, Brave private-inference partnership, fee switch revenue. **Honest bear case**: $117M TVL is small (RHEA Finance is concentration risk). Governance controversy — Chorus One opposed the inflation halving as forced through despite a failed initial governance vote. Memecoin overhang on AI/crypto narrative. Execution risk vs Solana's deeper liquidity and consumer DeFi. ETF filings ≠ approvals. **Asymmetry**: at $1.67B with vesting done, halved inflation, fee burn, ETF filings in flight, $13B+ routed cross-chain volume, transformer co-author at the helm — downside bounded by L1 floor, upside multi-X if the agent thesis lands.

Translation + commentary on Bittensor founder Jacob Steeves's Tsinghua University talk. Cameron walks through Jacob's framing of "incentive computing" as the universal pattern behind both Bitcoin and AI. Five-step argument: **(1) One pattern underlies every powerful adaptive system**: state · objective · feedback · adaptation · loop. AlexNet 2012 broke MNIST not by hand-coding what digits look like, but by letting the network self-adapt to a target. The same loop describes RL, genetic algorithms, slime molds finding shortest paths through mazes, river deltas, the structure of leaf veins. **(2) Bitcoin is the first production-scale implementation of this pattern** — not as money, but as a self-adaptive computer that produces hashes. The numbers are absurd: 1000x the compute of America's six largest cloud providers combined, 10²¹ hashes/sec, 23GW continuous power (Thailand-scale). 700-9000x more efficient at producing hashes than centralized cloud — because it's borderless, always-on, autonomous, and permissionless. Bitcoin is the world's largest supercomputer, optimized purely for hash production. **(3) Incentive computing** generalizes the pattern by replacing "reward = a number in a computer" with real money. ML's reward signal can't pay 200 countries' worth of contributors; Bitcoin's can — that's why the entire planet became a mining network. But hashes are useless outside Bitcoin. The question is whether the same mechanism can mint *anything*. **(4) Bittensor is the generic version** — replace "miners produce hashes" with "miners produce any useful work": storage, compute, ML models, gradients, data, robotics. Validators score, network mints. PyTorch for incentive computing. **(5) Five proven examples already running on Bittensor**: - **SN62 Ridges (SWE-Bench coding agents)** — top miner makes $60K/day. The agent that beat Claude/OpenAI on SWE-Bench was 7,000 lines written by an unknown person. "An AI lab with no engineers — it doesn't define how to solve the problem, it only defines the incentive." - **SN3 τemplar (cross-internet collaborative pre-training)** — successfully trained a 70B-parameter model across the open internet. Has never been done before. Cameron notes the founder later "ran away" — full piece coming. - **GPU markets (SN51 Lium, SN4 Targon)** — borderless permissionless GPU rental → world's lowest GPU prices. - **SN64 Chutes (open-source inference)** — #1 open-source provider on OpenRouter, 9.1T tokens. Briefly served more DeepSeek queries than DeepSeek itself. - **Robotics + long tail** — drone simulation, US stock signals, sports betting, drug discovery, weather forecasting, quantum compute, commodity trading. **dTAO** (live since Feb 2025) makes the network self-referential — subnets compete in capital markets for emission allocation. The market itself decides which incentive mechanisms get the next round of TAO. **The deeper point**: AI is being captured by a tiny number of closed labs (OpenAI, ~3K employees, you'll never own any of it, your data goes who knows where). Incentive computing distributes ownership and makes the rules visible. Anyone can enter, contribute, and own a piece — even if Bittensor isn't the project that wins, the *shape* of the AI economy will change because of this idea.

Bull thesis on Bittensor / TAO at ~$3B mcap. Frame: "TAO 2026 = ETH 2016 = BTC 2013." **Core mechanic**: Bitcoin paid miners to produce hashes that secure the network but are otherwise worthless. Bittensor pays miners — data scientists, ML engineers, AI researchers — to produce *useful* AI work. Validators score outputs via Yuma Consensus; TAO flows to whoever produces the most valuable work. Network is organized into 128+ subnets, each focused on a specific task (trading signals, LLM training, computer vision, code generation, financial forecasting). Some subnets generating millions in revenue, with Intel and PwC partnerships. **Tokenomics mirror Bitcoin**: 21M fixed supply, no pre-mine, no VC allocation. First halving Dec 14 2025. BTC price went 83x in the year after its first halving in 2012. **Smart-money signals**: Barry Silbert / DCG launched Yuma Group dedicated to accelerating Bittensor. Grayscale filed Form S-1 to convert GTAO Trust into a spot ETF. Stillcore Capital (Mark Jeffrey, Jason Calacanis, Rob Greer) targeting $1T mcap by 2030, aiming to own 1% of all TAO. Unsupervised Capital projects $4,800 by Dec 2027 (19x), bull case $10,800 — and that's *before* Covenant-72B, Jensen mentioning Bittensor, and PwC's formal alliance. **Subnet-level conviction picks**: - **Targon (SN4)** — decentralized AWS for AI; Targon VM gives encryption + hardware-backed protection so hardware operators can't access data, weights, or workloads. Co-authored a paper with Intel in March 2026. Built by ex-OpenTensor founders (Robert Myers — among first 3 people ever in the Bittensor Discord; James Woodman ex-GSR). - **Vanta (SN8)** — disrupts the $20B prop firm industry. Single eval, 100% profit split, fully on-chain verification. Already net profitable on revenue vs miner emissions. Hyperscaled is the Hyperliquid version. - **Chutes (SN64)** — #1 open-source provider on OpenRouter, 9.1T tokens processed. Decentralized AWS with no CEO. - **RESI (SN46)** — institutional-grade real estate intelligence. 98% accuracy remote appraisals on a $600T asset class running on broken legacy MLS systems. 1000+ appraisals + nationwide lender partnership in week one. Strategic investment from Stillcore. - **Affine (SN120)** — built by Const himself (Bittensor co-founder, wrote the Yuma Consensus + subnet architecture). Continuous evaluations on open-source reasoning models, leverages Chutes for hosting. - **Score (SN44)** — first subnet ever to partner with a Big Four firm. Manako product distributed by PwC France to retail, manufacturing, logistics, energy enterprise clients. - **Oro (SN15)** — autonomous AI shopping agents. 45 Oro agents have outperformed GPT-5.4 on hard online shopping evals. **Frame**: Bitcoin = money. Ethereum = apps. TAO = intelligence. The gap between what TAO has built and how it's currently priced is one of the most asymmetric opportunities in crypto.

Modular Capital argues the U.S. power deficit of 50 GW through 2028 makes Bitcoin miners' converted data centers the lowest-cost path to AI infrastructure, with energized capacity priced at $30–50/MWh versus $80–150/MWh for alternatives like nuclear and fuel cells. Bitcoin miners holding 10+ GW of approved grid interconnection now command structural economics worth $5–10M per gross megawatt in HPC colocation deals, with recent transactions clearing at $1.24–$2.17 per critical IT watt annually and 80–97% EBITDA margins, while the sector at $4M equity value per approved MW implies only 50% conversion pricing, leaving asymmetric upside for operators with large approved portfolios and credible execution track records.

In a world saturated with AI agents, Altman's Worldcoin identity project becomes essential infrastructure — you need a provably-human layer. Fernando frames identity-for-AI as a category hiding in plain sight: when 'more things look like people than people do', the iris-scan primitive becomes the on-ramp for every other consumer product that needs to distinguish humans from bots.

CoreWeave's $6B Jane Street contract is the largest-ever AI cloud deal with a non-AI-lab customer — validates Rittenhouse's February thesis that CRWV's long-term success depends on diversifying beyond hyperscalers and AI labs toward enterprises. Analog: AWS's early cloud-native-startup focus before enterprise proliferation. Signal for $NBIS too, which has been emphasizing the same enterprise pivot.

Market is overlooking $GLXY's Helios datacenter business because it doesn't trust $CRWV will fulfill its 15-year, $1B/yr contract for the first 800MW. Gab argues Helios is priced at zero in the stock today — so if CRWV delivers, there's a meaningful mispricing inside a crypto-native equity. Asymmetric setup on the equity side of the AI-compute trade.