Ventura Labs
July 28, 2025

James Woodman & Joshua Brown: Targon, Manifold Labs, Bittensor Subnet 4, Compute Markets | Ep. 53

Manifold Labs’ CTO Joshua Brown and COO James Woodman break down the architecture and vision for Targon, Bittensor’s Subnet 4. This episode dives into their strategy for building a transparent, liquid, and ultimately dominant marketplace for digital compute.

Architecting a Verifiable Compute Market

  • “We're actually getting guarantees from NVIDIA themselves that whatever reports we're getting back from these CVMs are valid... there's a cryptographic key that only NVIDIA has and they can verify whether this attestation report was generated with this hardware.”
  • At its core, Targon’s architecture is designed for verification and efficiency. The system operates in 72-minute intervals where validators ping miners to confirm their hardware is legitimate using NVIDIA’s own cryptographic attestation. Miners submit bids for their compute, and an auction system sorts these bids into price buckets. The lowest bids are paid out first from the subnet's emission pool, creating a powerful market dynamic that drives prices toward a true equilibrium. This entire process is built in Go, a deliberate choice for speed and stability over the more common Python in the Bittensor ecosystem.

The CME for Compute

  • “The compute market is really messed up… There's a lack of transparency in the market today. It's like trading OTC Discord back in 2021 basically.”
  • “Why is the CME been so successful? Because it's effectively financialized something as simple as corn… Targon and Manifold are developing this CME, this exchange for compute that everyone goes to.”
  • Targon's grand vision is to fix the opaque and inefficient traditional compute market. By creating a public, liquid exchange, they aim to eliminate the need for brokers and opaque pricing, where financing costs are high and true costs are hidden. The initial strategy is to bootstrap the supply side of this market using Bittensor’s token incentives, guaranteeing 100% utilization for compute providers—their single most important metric. This creates a foundation to eventually build a two-sided marketplace with sophisticated financial products like forward contracts for compute.

From Emissions to Enterprise Revenue

  • “If you don't get organic demand, it's done. It's just going to death spiral… We're returning over the last eight days… $52,000… That's about 1/7th of the incentive.”
  • The clock is ticking. The team recognizes that the window to use token emissions for bootstrapping is finite, perhaps only 6-12 months. The primary focus is converting incentives into sustainable, organic revenue from inference clients. The flywheel is already spinning: Targon is generating thousands of dollars in revenue, which is used to buy back its own subnet token. The next major unlock is introducing collateralized, non-interruptible compute tiers to attract enterprise clients who require uptime guarantees, moving beyond the current market for cheaper, interruptible workloads.

Key Takeaways:

  • The conversation outlines a clear, pragmatic strategy to build a defensible moat in the decentralized compute space. It’s a race to convert the initial momentum from token incentives into a self-sustaining business with real-world demand.
  • Price Discovery is the Product: Targon's auction mechanism isn't just a feature; it's the core product. By forcing compute providers to bid for their payout, the system creates a hyper-competitive environment that reveals the true, market-driven price of compute, incentivizing efficiency and driving costs down.
  • The Race for Organic Revenue: The entire model hinges on achieving "escape velocity" where organic revenue from inference clients outpaces the reliance on network emissions. With $52,000 returned to the subnet in just eight days, they are proving the model works, but scaling this revenue is the central challenge.
  • The Future is Financialized Compute: The end goal extends far beyond simply renting out GPUs. By establishing a liquid spot market, Targon is laying the groundwork to introduce financial derivatives like forward contracts and options, allowing enterprises to hedge against compute price volatility just as they do with other commodities.

For further insights and detailed discussions, watch the full podcast: Link

This episode reveals how Manifold Labs is building a transparent, real-time marketplace for GPU compute on Bittensor, moving from crypto-native incentives to a model designed to attract enterprise-level demand.

Introduction to Targon's Architecture

  • Validators and Miners: The system uses a centralized set of validators that communicate with miners. Each miner runs a single "minor proxy," which is their sole point of contact with the Bittensor chain and the validators.
  • CVMs (Confidential Virtual Machines): Behind each proxy, miners operate multiple CVMs. A CVM is a secure, isolated environment running on the miner's hardware, such as an 8x H200 node. The proxy reports the IP addresses and price bids for each of its CVMs to the validators.
  • Bidding Process: Miners submit a simple list of IP addresses and their corresponding bids in cents per GPU per hour (e.g., $2.00/hr for each H200 in a cluster). This bid represents the "ask" price from the supply side.

The 72-Minute Interval and Price Discovery

  • The Targon Tower: To ensure consistency, all validators reference the "Targon Tower," a centralized oracle providing a single source of truth for the TAO price, the maximum allowable bid for miners, and details on currently active auctions (e.g., the H200 auction).
  • The Emission Pool: Validators calculate the "emission pool" at the start of each interval. This is the total USD value of TAO rewards flowing into Subnet 4 for that 72-minute period, which determines the total budget available to pay miners. Josh explains, "It is the US dollars of the 41 cut," referring to the 41% of emissions that go to miners and 41% to validators.

CVM Attestation and NVIDIA-Backed Security

  • Attestation Reports: At the start of an interval, and then randomly throughout, validators request an "attestation report" from each CVM. This report is a cryptographically signed data blob containing GPU information, generated using keys baked directly into the NVIDIA hardware.
  • NVIDIA Verification: Validators send this report to NVIDIA's servers to confirm its authenticity. This process provides a strong guarantee that the CVM is running on a legitimate H200 GPU, as only NVIDIA holds the keys to verify the hardware signature. Josh notes, "We're actually getting guarantees from NVIDIA themselves that whatever reports we're getting back from these CVMs are valid."
  • Strategic Implication: This direct hardware verification is a key differentiator, providing a level of trust and quality control that is crucial for attracting serious users and preventing fraudulent miners from gaming the system.

The H200 Auction Mechanism Explained

  • Bucket-Based Payouts: Bids are grouped into "rings" or buckets by price (e.g., all bids at $2.19, all bids at $2.20). The system starts by paying out the lowest-priced ring in full.
  • Finding the Market Price: It continues paying out rings until it reaches a price point where the emission pool can no longer cover all the bids in that ring. The remaining funds are then distributed pro-rata among the CVMs in that final, partially-funded ring.
  • Actionable Insight: This mechanism forces miners to bid their true operational cost. Bidding too high risks not getting paid at all, while bidding lower secures a payout. This creates a competitive environment that continuously discovers the real-time, market-clearing price for interruptible H200 compute.

The Vision for a CME of Compute

  • Market Inefficiency: James highlights the lack of transparency, comparing the current state to "trading OTC Discord back in 2021." Targon aims to solve this by creating a public order book where the true cost of compute is visible to everyone.
  • Decentralization as a Scalability Solution: Centralized superclusters like those built by OpenAI and xAI face immense power and logistical constraints. James proposes a more scalable model: "What you probably want if you could pull it off is, you know, a couple million dollars of hardware here, 30 million there, a hundred million there, and it's all connected."

Bootstrapping Supply and Ensuring Quality

  • Incentives as a Bridge: The TAO emissions provide "decentralized demand," paying miners to bring their hardware online before organic, paying customers arrive at scale. This ensures there is a ready supply of compute for Targon to sell.
  • Avoiding the "Public Dumping Ground": James emphasizes that quality is paramount. To avoid becoming a marketplace for low-quality, "garbage compute," Targon uses the attestation process and plans to introduce more sophisticated quality checks. He warns, "You don't want to become a public dumping ground. That would be the end of Targon, I think, if that happened to us."

The Path to Self-Sustainability and Enterprise Adoption

  • Compute Efficiency: A key focus is on "compute efficiency"—using less hardware to deliver the same quality of service. This allows Targon to serve more clients with its existing supply, a crucial step for scaling its business.
  • Collateral and Non-Interruptible Compute: The next major step is introducing a collateral system. This will allow miners to post a bond to guarantee uptime and performance, creating a premium, "non-interruptible" tier of compute. This is essential for attracting enterprise clients who cannot tolerate downtime.
  • Strategic Implication: For investors, the introduction of collateralized tiers is a major catalyst to watch. It signals a move from serving the interruptible spot market to competing for high-value, long-term enterprise contracts, unlocking a much larger addressable market.

Defining the "Institutional Miner"

  • From Renting to Owning: This marks a shift from early miners who rented hardware from brokers to sophisticated operators who own their assets. This is crucial because Targon's thesis is to disintermediate the very brokers that renters rely on.
  • Actionable Insight: The emergence of these institutional miners is a sign of market maturation. Researchers should track the composition of miners on the subnet, as a shift towards owner-operators indicates a more stable and professional supply base.

Final Advice for Bittensor Builders

  • Josh's Technical Advice: He urges developers to build their core product first and not get overly constrained by Bittensor's initial tooling. "Figure out what your actual needs are and build the system around what the actual idea is, not necessarily around Bittensor itself."
  • James's Business Advice: He stresses the importance of identifying the core commodity being produced and then figuring out how to sell it effectively. He advises builders to focus on creating a scalable, liquid market for their commodity, as that in itself is a powerful moat.

Conclusion

Targon's strategy demonstrates a clear path from a crypto-incentivized supply base to a revenue-generating, enterprise-focused compute marketplace. Investors and researchers should monitor the launch of collateralized compute tiers and the growth of organic revenue as key indicators of its potential to disrupt the traditional cloud market.

Others You May Like