Nistar Intelligence

Signal, not noise.

Perspectives from the intersection of power, artificial intelligence, data centers and capital—built for the people developing, financing and operating the next generation of compute.

68Insights

6Signals

2024–26Archive

NISTARSignal map

Select a signal to filter the research library

Latest signals

AI + ComputeGPUaaS vs. the Token Factory: When AI Infrastructure Stops Renting GPUs and Starts Manufacturing Intelligence AI + ComputeAI Is Learning to Notice What Humans Miss: How Claude Found a Hidden Pattern in Viral DNA AI + ComputeThe Cure May Already Exist: How AI Is Turning Medicine Into a Search Problem AI + ComputeAI Is Becoming Research Infrastructure: What OpenAI's Navier–Stokes Claim Means for the Next Era of Compute DevelopmentThe Next Data Center Constraint Is Social License PowerA Utility Letter Is Not Power: What Actually Makes a Megawatt Real AI + ComputeTraining, Inference, and the Geography of AI Compute: Why Workload Type Will Determine Where the Next Data Centers Get Built

Featured analysis

What we’re watching.

Recent analysis on the infrastructure forces changing how AI capacity gets powered, built and financed.

Lead analysis / 01Sep 25, 2026

AI + ComputeTraining & InferenceMarkets

GPUaaS vs. the Token Factory: When AI Infrastructure Stops Renting GPUs and Starts Manufacturing Intelligence

The first generation of AI clouds monetized scarce accelerators by the GPU-hour. The next generation is moving higher in the stack, selling the output of those accelerators as tokens, requests and agentic workflows. That transition changes who carries utilization risk, how infrastructure is financed, how software creates margin—and eventually why the cost of the megawatt underneath the GPU may become one of the industry's most durable competitive advantages.

By Jay SivamRead analysis

Brief / 02Sep 25, 2026

AI + ComputeAI Research & Scientific Discovery

AI Is Learning to Notice What Humans Miss: How Claude Found a Hidden Pattern in Viral DNA

Anthropic says roughly 950 Claude agents searched an enormous genomic dataset and surfaced a previously uncharacterized architecture hidden around a known viral reverse transcriptase. Its function remains unknown, and repeated AI searches failed to rediscover it. That may be exactly why the result matters: scientific AI is beginning to move beyond answering questions toward noticing anomalies humans never thought to ask about.

Jay Sivam

Brief / 03Sep 15, 2026

AI + ComputeAI Research & Scientific Discovery

The Cure May Already Exist: How AI Is Turning Medicine Into a Search Problem

A dying rare-disease patient improved after an AI-assisted system helped surface a drug regimen physicians had not tried. But the more important story began years earlier, when physician-scientist David Fajgenbaum repurposed an existing medicine to save his own life. Today, Every Cure is attempting to industrialize that idea: search the world's existing medicines against the world's diseases and find treatments hiding in plain sight.

Jay Sivam

Research library

Explore all insights.

Search by subject, author or keyword. Switch between editorial cards and a compact research index.

Discipline

68 results

AI + Compute

Training & Inference

GPUaaS vs. the Token Factory: When AI Infrastructure Stops Renting GPUs and Starts Manufacturing Intelligence

Markets+ Capital+ Power

Jay SivamSep 25, 2026

AI + Compute

AI Research & Scientific Discovery

AI Is Learning to Notice What Humans Miss: How Claude Found a Hidden Pattern in Viral DNA

Technology+ Infrastructure

Jay SivamSep 25, 2026

AI + Compute

AI Research & Scientific Discovery

The Cure May Already Exist: How AI Is Turning Medicine Into a Search Problem

Technology+ Infrastructure

Jay SivamSep 15, 2026

AI + Compute

AI Research & Scientific Discovery

AI Is Becoming Research Infrastructure: What OpenAI's Navier–Stokes Claim Means for the Next Era of Compute

OpenAI says an internal AI system produced a proposed solution to one of mathematics' Millennium Prize problems using roughly 10,000 agents, 130 billion output tokens and formal machine verification. Whether the proof earns universal acceptance remains a scientific question. The infrastructure signal is clearer: frontier compute is beginning to produce new knowledge.

Technology+ Infrastructure

Jay SivamSep 8, 2026

Development

Community & Siting

The Next Data Center Constraint Is Social License

Data center development has traditionally been framed around land, power, fiber and permits. Another constraint is becoming harder to ignore: whether the surrounding community believes the project should be built. Electricity costs, water, noise, transparency, jobs and local economic benefits are increasingly influencing project schedules, permitting and even financing.

Strategy+ Policy

Jay SivamAug 27, 2026

Power

Grid & Interconnection

A Utility Letter Is Not Power: What Actually Makes a Megawatt Real

Utility letters can be extremely valuable evidence in data center development. So can load studies, service plans and executed agreements. But the title on a document does not determine how real the megawatts are. What matters is what the utility has actually studied, committed to, conditioned, scheduled and ultimately constructed.

Engineering+ Development

Jake BeckerAug 26, 2026

AI + Compute

Training & Inference

Training, Inference, and the Geography of AI Compute: Why Workload Type Will Determine Where the Next Data Centers Get Built

AI infrastructure is often discussed as one enormous category of GPU demand. But training and inference place different demands on power, networks and geography. Frontier training rewards enormous, tightly connected compute systems. Inference adds another constraint: the user. As AI shifts from building models to serving them at global scale, workload type will increasingly influence where infrastructure belongs.

Strategy+ Development

Sean KurzAug 24, 2026

Policy

Cost Allocation

Who Pays for the Grid? The New Cost-Allocation Fight Around AI Data Centers

The AI power debate is moving beyond whether enough electricity can be built. The next question is who pays for the infrastructure required to deliver it. As hundreds of megawatts of new demand drive generation, transmission and distribution investment, utilities and regulators are building new rules around cost causation, minimum payments, collateral, stranded investment and ratepayer protection.

Markets+ Power+ Capital

Robert DizonAug 19, 2026

Infrastructure

Electrical Systems

A Data Center Load Is No Longer a Megawatt. It Is a Waveform.

AI infrastructure is changing the electrical character of the data center. Total megawatts still determine scale, but increasingly they do not tell the whole story. Large synchronized GPU clusters can change power demand rapidly, creating a new engineering question: not simply how much electricity the campus uses, but how the load behaves over time.

Engineering+ Power+ AI + Compute

Jake BeckerAug 18, 2026

Nistar

Building at the intersection of power and compute?

Nistar works across energy infrastructure, data center development and AI compute to turn complex opportunities into executable projects.

Discuss a project