Jansen Tang

Tang argues that full AI autonomy is years away, but "remote teleoperation" is viable today. He cites a "wage arbitrage" opportunity where labor in low-cost countries (e.g., Malaysia, Philippines) controls robots in high-cost countries (e.g., US, Australia) for tasks like security, cleaning, and retail. This creates a "Business Process Outsourcing" (BPO) model for physical labor. Companies that facilitate this teleoperation infrastructure or supply the hardware (Unitree G1s mentioned) stand to capture the 40-60% cost savings margin between developed and developing market wages. LONG. This is a practical, immediate use case for humanoid robots that bypasses the technical bottleneck of full AGI, allowing for immediate revenue generation. Latency issues in teleoperation, hardware maintenance costs, and potential political backlash against "outsourcing physical jobs."

Virtuals Protocol is built on Base (Coinbase's L2). Tang notes they are one of the earliest and largest projects on the chain, driving significant transaction volume through agent-to-agent commerce. He mentions speaking directly with Jesse Pollak (Base lead) about scaling. Virtuals views itself as a "nation-state" of agents. As AI agents begin transacting autonomously (buying services, paying for compute), transaction velocity on Base increases. Coinbase (COIN) monetizes Base through sequencer fees. High-frequency agent commerce is a new, non-human revenue stream for the exchange's L2. LONG. Betting on the infrastructure layer that powers the "Agentic Economy." Ethereum gas spikes (which Tang noted as a problem) could force migration to alternative high-throughput chains (like Solana).

Tang confirms that humanoid hardware costs are plummeting (Unitree G1s dropping from $60k to $20k) and dexterity is high enough for commercial use (retail, hotels). He explicitly states the "teleoperation" phase will generate the massive datasets needed to train fully autonomous models over the next 18 months. While Tang uses Unitree (private/Chinese), the validation of the "teleoperation-to-autonomy" pipeline is a massive bullish signal for the sector. Tesla (Optimus) is the primary US-listed beneficiary of this hardware curve, and NVIDIA provides the compute necessary to process the "egocentric data" (video/joint movement) required to train these foundation models. LONG. The "teleoperation" bridge makes the robotics timeline immediate rather than distant. Hardware failure rates in the wild; regulatory pushback against remote-controlled physical labor.

Tang confirms that humanoid hardware costs are plummeting (Unitree G1s dropping from $60k to $20k) and dexterity is high enough for commercial use (retail, hotels). He explicitly states the "teleoperation" phase will generate the massive datasets needed to train fully autonomous models over the next 18 months. While Tang uses Unitree (private/Chinese), the validation of the "teleoperation-to-autonomy" pipeline is a massive bullish signal for the sector. Tesla (Optimus) is the primary US-listed beneficiary of this hardware curve, and NVIDIA provides the compute necessary to process the "egocentric data" (video/joint movement) required to train these foundation models. LONG. The "teleoperation" bridge makes the robotics timeline immediate rather than distant. Hardware failure rates in the wild; regulatory pushback against remote-controlled physical labor.

Tang details the tokenomics: The protocol charges a tax on agent commerce and token swaps. 70% goes to devs, 30% to the Virtuals treasury. This treasury revenue is used to incentivize the ecosystem and effectively reduce supply (buybacks/burns implied via liquidity structuring). The launch of Eastworld Labs expands the TAM from digital chatbots to physical robotics. If the "teleoperation" thesis holds, the VIRTUAL token becomes the currency for cross-border physical labor arbitrage, significantly increasing velocity and fee capture. LONG. Direct exposure to the specific protocol aggregating AI agent liquidity. Protocol execution risk; competition from other AI agent launchpads (e.g., ai16z).