Mog Debates Will on Revenue vs Research at Exploit Summit

The question is simple: “Research is more important than revenue.”

Will Squires of Macrocosmos argues for it. Mog (Marcus Graichen of Taostats) argues against. They’ve argued online. In Montreal this September, they do it in person for the first time.

Will: Push the Ceiling

Squires runs Macrocosmos with Steffen Cruz as a co-founder. Twenty-four people building in Macrocosmos, multiple PhDs, and a stubborn refusal to take shortcuts.

The receipts:

• Some of the earliest decentralized 7B and 14B models, pretrained on Subnet 9 (IOTA)
• IOTA: arguably the fastest pipeline parallelism implementation in decentralized settings, with an orchestration layer built from scratch while most teams lean on Hivemind
• A compression-based bottleneck architecture that slashes communication overhead and makes distributed training hardware-agnostic
• Orion: a 100B parameter model trained on distributed compute

His stance: decentralized intelligence is hard enough that only relentless research unlocks what centralized labs can’t replicate. Revenue is a lagging indicator. Build something fundamentally better, and the money follows.

Macrocosmos’ Orion-100B Raises the Ceiling for Decentralized AI Training

Mog: Raise the Floor

Mog started Taostats as a Google spreadsheet in late 2022, tracking incentives while running nodes through the bear market. It’s now the ecosystem’s primary data and staking interface.

The receipts:

• Blockmachine (SN19): launched on mainnet with real customers and revenue from day one. Decentralized RPC, no API key needed, billions of monthly requests, and it’s expanding past Bittensor. Taostats runs on it and reports major cost cuts.
• GM (SN28): private gateway routing prompts to Claude, GPT, and Gemini through Intel TDX enclaves. No provider ever sees raw user data.
• Hippius (SN75) and Vidaio (SN85): storage/compute and video compression
• Corcel: the first major API gateway into Bittensor’s AI capabilities

His argument: ecosystems grow when people ship things others pay for. Cheaper, faster, more private infrastructure creates the flywheel that funds the research. Beautiful tech nobody uses is a hobby.

Is Hippius (SN75) Better Than Dropbox?

Why It Lands

This isn’t philosophy. It’s a live tension inside a protocol still finetuning its incentive and governance layers.

Will is asking whether we can train frontier models decentralized and win. Mog is asking whether anyone can use the network and pay for services on it today. Both have shipped. Both have skin in the game. Neither is wrong, which is exactly why it’s worth watching them debate it.

Whoever wins the room, the answer shapes how subnets prioritize, how teams get funded, and what “success” even means here.

Where Do You Stand?

Research first, or revenue first? Or some ratio nobody’s found yet? Drop it in the comments. The community’s opinion matters as much as the stage.

Event Details

Exploit Summit, Bittensor’s flagship conference

• Location: Montreal, Canada
• Date: September 28–29, 2026
• Buy your ticket with a 30% discount: exploitsummit.com (use TAODAILY30 as access code)