Insights

Research and analysis on AI systems and infrastructure.

Structured breakdowns of emerging technologies, search systems, and applied intelligence.

Memory Is the Real Bottleneck — And Everyone Is Still Optimizing the Wrong Thing
Insights · Jun 12, 2026

Memory Is the Real Bottleneck — And Everyone Is Still Optimizing the Wrong Thing

The assumption: more GPUs equals more AI capability. The reality: the bottleneck has shifted from compute to memory, leaving expensive hardware sitting partially idle while teams track the wrong infrastructure metrics.

Read analysis →
Every AI Data Center Built Since 2022 May Be a Stranded Asset. The Industry Has No Tool to Measure This.
Insights · May 14, 2026

Every AI Data Center Built Since 2022 May Be a Stranded Asset. The Industry Has No Tool to Measure This.

Data center infrastructure is long-lived. The 20 to 40 year depreciation schedules applied to data center facilities reflect a real historical truth: the buildings last for decades. The assumption underpinning $7.6 trillion in projected AI infrastructure investment through 2031 is that the facilities being built now will be productive through the depreciation period. The assumption is not obviously wrong for the building itself. It is almost certainly wrong for a significant portion of the infrastructure inside.

Read analysis →
The $650 Billion AI Buildout Has a 10% Problem That Is Stopping the Other 90%
Insights · May 13, 2026

The $650 Billion AI Buildout Has a 10% Problem That Is Stopping the Other 90%

AI infrastructure investment is constrained by capital and by GPU supply. More capital deployed = more AI capacity built = more compute available. The hyperscalers are spending hundreds of billions — therefore, compute is expanding rapidly.

Read analysis →
The Real AI Infrastructure Crisis Is Power, Not Compute
Insights · May 5, 2026

The Real AI Infrastructure Crisis Is Power, Not Compute

GPU availability dominated the AI narrative, from H100 allocation wars to cloud pricing drops. But the real constraint has shifted: not compute, but power.

Read analysis →
Token Prices Are Falling. Your AI Bill Is Rising. Both Are True.
Insights · May 4, 2026

Token Prices Are Falling. Your AI Bill Is Rising. Both Are True.

Per-token costs are collapsing, but enterprise AI bills keep rising. This piece explores why token deflation does not translate to cost deflation—and the hidden mechanics driving AI spending upward.

Read analysis →
Benchmarks Are Quietly Breaking AI
Insights · May 4, 2026

Benchmarks Are Quietly Breaking AI

AI systems are no longer optimizing for capability → they are optimizing for benchmark environments.

Read analysis →
The Industry Is Spending on Training and Calling It AI Infrastructure. The Bill for What Actually Runs AI Has Not Arrived Yet.
Insights · May 4, 2026

The Industry Is Spending on Training and Calling It AI Infrastructure. The Bill for What Actually Runs AI Has Not Arrived Yet.

AI infrastructure investment is dominated by GPU clusters for training frontier models. The assumption embedded in every infrastructure spending analysis from 2022 through 2024: training is the primary compute cost. Build the training clusters; inference will be manageable.

Read analysis →
The "Agentic AI" Transition Is Happening Without the Infrastructure to Make It Safe
Insights · Mar 14, 2026

The "Agentic AI" Transition Is Happening Without the Infrastructure to Make It Safe

Agentic AI deployment metrics. "44% of companies deploying or assessing AI agents." "Telecommunications: 48% adoption." The narrative: agentic AI is being widely deployed. The implication: it's working at scale.

Read analysis →
The AI Talent Market Has Inverted — and Nobody Has Measured the Real Supply
Insights · Jan 3, 2026

The AI Talent Market Has Inverted — and Nobody Has Measured the Real Supply

AI talent as a technical scarcity problem. "We can't find ML engineers." "There aren't enough AI researchers." The narrative: the bottleneck is specialized technical talent. Salaries spike. Visa programs expand. Universities launch AI programs. The assumption: build more AI practitioners and the talent gap closes.

Read analysis →