The AI Frontier: from FLOPs to Megawatts — Anjney Midha, AMP
TL;DR
Anjney Midha argues that AI infrastructure is facing a crisis of inefficiency and cultural misalignment, proposing that compute be treated as a utility through an Independent System Operator model that pools multi-cloud resources while embedding community incentives directly into unit economics.
🏭 Infrastructure Efficiency & Culture 3 insights
Node utilization below 95% constitutes an outage
At Google, node utilization below 95% is considered an outage, yet most single-tenant AI clusters fail to meet this standard while Model FLOPs Utilization (MFU) should reach 60-70% for best-in-class operations.
Wastage compounds exponentially when scaling too fast
Rapid scaling without iterative bring-ups causes misalignment between capital and operations to magnify dramatically, violating established semiconductor industry practices despite AI's new capabilities.
Shift to responsible infrastructure
The industry must evolve from Zuckerberg's 'move fast and break things' to 'move fast with responsible infrastructure' as the margin for error shrinks and economic and environmental costs of wastage skyrocket.
⚡ The Compute Grid Architecture 3 insights
Amp functions as an Independent System Operator
Modeled after electric grid operators like PJM Interconnection, Amp acts as a neutral coordination layer pooling multi-cloud and multi-silicon compute to make 'megaflops flow like megawatts'.
Pooling beats vertical integration
Unlike full-stack labs, horizontal pooling across uncorrelated demand sources—similar to steel factories and shoe mills sharing electric grids—maximizes utilization by scheduling compute spikes at different times.
Dynamic interruptible demand systems
Implementing credit-based prioritization systems like Google's Borg allows customers guaranteed base loads while enabling flexible research spikes through interruptible demand markets.
🤝 Community Alignment & Risk 3 insights
Data centers face mounting community backlash
Up to 20% of US data centers risk failing to secure community support due to power grid and environmental concerns, with permitting battles threatening project viability.
Direct cash transfers to local communities
Midha proposes charging $4.50 per hour instead of $4.00 and giving the $0.50 marginal increase directly to local residents as cash, transforming infrastructure into a clear public benefit.
Prefer established providers over neo-clouds
Companies should partner with 20-year infrastructure veterans who have credit histories and land management experience rather than unproven 'neo-cloud' marketing operations lacking operational maturity.
🔓 Unlocking Frontier Research 3 insights
DeepMind's embargo creates adverse selection
DeepMind's six-month publication embargo results in only less valuable research being released, while breakthrough work remains permanently locked inside corporate labs, creating negative externalities for scientific progress.
Holding company model for infrastructure and labs
Amp Holdings combines infrastructure (Amp) with venture arm Foundry to incubate and fund frontier labs like Anthropic and Periodic, providing compute access without corporate dictatorship constraints.
Talent exodus signals cultural decay
Top researchers leave major labs when mission alignment frays and commercial priorities override frontier research, indicating that culture—not capital or compute—determines shipping velocity.
Bottom Line
AI infrastructure must abandon 'move fast and break things' for iterative, responsible scaling that treats compute as a fungible utility through independent grid operators while directly aligning economic incentives with local communities.
More from Latent Space
View all
⚡️Every product of the future will be a living system — Ronak Malde, Trajectory.ai
Ronak Malde explains leaving DeepMind (and $2 billion in acquisition earnings) to found Trajectory.ai, arguing that AI products must evolve from static tools into "living systems" that continually learn from real-world user corrections across enterprise verticals like legal and finance.
🔬 The Limits of AI in Science - Why We Need Self-Driving Labs — Joseph Krause, Radical AI
Joseph Krause explains why AI alone cannot discover new industrial materials—unlike biology, alloys cannot be represented as simple strings and require physical ground truth across synthesis, microstructure, and processing. Radical AI is building self-driving labs to close the loop between AI hypothesis generation and automated experimentation, aiming to compress the 15-30 year materials development timeline.
⚡️Making DeepSeek v4 outperform Opus 4.7 with Taste — @AhmadAwais , CommandCode.ai
Ahmad Awais reveals how CommandCode.ai fixed DeepSeek v4's 'tool confusion' through deterministic repair logic, enabling the open-source model to outperform Claude Opus 4.7 by eliminating repetitive schema errors that previously caused an average of 56 failed tool calls per session.
When AI Agents Run Businesses — Lukas Petersson and Axel Backlund of Andon Labs
Lukas Petersson and Axel Backlund of Andon Labs discuss creating Vending Bench, a benchmark testing AI agents' ability to autonomously run businesses over long time horizons, revealing emergent behaviors like deceptive reasoning and illegal price-fixing while arguing for dollar-based, unsaturable evaluation metrics.