Physics doesn’t negotiate.
Every watt that enters a data center leaves as heat. No exceptions. The playbooks written for the last era don’t transfer. The dashboards lie.
We’re an R&D lab. We build at the seam between physical infrastructure and automated computing, where power and thermal constraints meet the software that has to survive them.
The old playbook was never designed for this density. We’re writing the new one.
Not consulting. Not advising. Building.
Abstract After measuring a 292x cost gap between a rented B200 and frontier API providers on a batch inference workload, the logical next question was whether the same pattern held for operational intelligence: could a smaller, dedicated model handle the continuous judgment calls required to run a GPU fleet? The batch inference test had exposed KV cache contention as the dominant bottleneck on shared API infrastructure. Processing similar structured data at scale, but this time continuously rather than in batch, seemed like a valid test of whether that contention would degrade operational quality the same way it degraded throughput. ...
Two in the morning, eighteen hours into the run. Seven LLM backends processing the same stream of GPU cluster anomalies. Same thermal cascades, same NVLink errors, same KV cache evictions. I’m watching the scoring dashboard update in real time and the numbers are breaking my assumptions faster than I can take notes. The $32-per-day model is getting the diagnosis wrong more often than a free one running on my workstation. ...
292x. That’s not a rounding error. That’s the cost multiplier between running a batch inference job on a rented B200 GPU and sending the same workload through Claude Opus 4.6’s API. The job was straightforward: generate one or two contextual sentences for each of a million documents, extracted JSON from the corporate PDF archive I’ve been building a RAG pipeline around. Those sentences get prepended to each chunk before embedding into Qdrant’s 768-dimensional vectors with BM25 sparse indexing. It’s the contextual layer that makes retrieval actually work, the step I described in the previous article about why a million PDFs won’t organize themselves. ...
Three months ago, I plugged a Blackwell GPU into my lab bench and pointed it at a million PDFs. Corporate documents: contracts, engineering reports, compliance filings, vendor proposals, maintenance logs, insurance certificates. The kind of archive that accumulates over two decades of running critical infrastructure. A million files. Not sampled, not curated, not cleaned. Raw. The plan was straightforward: parse the documents, chunk them, embed them into a vector store, wire up a retrieval layer, and start asking questions no keyword search could answer. Retrieval-Augmented Generation. The acronym that launched a thousand vendor decks. ...
A few weeks ago we hit a production issue on a cloud environment — one XCP-ng host was showing IOPS contention caused by a single guest VM. The classic noisy-neighbor race condition on shared storage. The diagnostic path was obvious: cross the dom0 guest list with iostat on the host, find the VM hammering the disk, and work the problem from there. Straightforward correlation — the kind of thing an experienced operator resolves in fifteen minutes with two terminal windows. ...
Part 2 ended with a promise: find the cliff. Run the MoE model from four concurrent agents upward until the physics says stop. We scaled to eight. The cliff never came. This is Part 3 of the Local LLM Bench series. Part 1 covers the single-request baseline. Part 2 established the MoE advantage under concurrent load. The model: Qwen3-Coder-30B-A3B — a Mixture-of-Experts architecture that activates only 3.3B of its 30B parameters per token. On consumer GPUs, that sparse activation leaves ~90% of memory bandwidth idle at batch size 1, creating headroom that concurrent agents fill. Dense models activate all 32B parameters on every token — already at the bandwidth ceiling before the second agent connects. Part 1 explains why these specific models were chosen (best in class for each architecture); Part 2 conclusively eliminated Dense under concurrent load. This benchmark tests MoE only. ...
In Part 1, we established the baseline: MoE delivers 168 tok/s on a single RTX 3090, 4.1x faster than Dense. Clean single-request numbers. One prompt in, one response out. That’s not how swarms work. An orchestrator like Claude Code dispatches four coding tasks simultaneously. The local model serves all four. Under concurrency, memory bandwidth saturates, per-task throughput drops, and the architecture of the model — not the GPU, the model — determines whether you get useful parallelism or just contention. ...
Meta’s Prometheus data center in New Albany, Ohio is scaling to 1.2 GW. To get there, they’re building behind-the-meter natural gas turbines — two 200 MW Socrates generation facilities, supplied by dedicated gas pipelines, isolated from the grid. In Virginia, the same story plays out with diesel generators, enough of them that it became the top legislative concern entering the 2026 session. The industry talks about PUE as if it were a verdict on environmental efficiency. It isn’t. PUE measures one envelope — the data center facility. Total facility power divided by IT equipment power. A PUE of 1.3 means 30% overhead for cooling, lighting, and support systems. That’s the metric everyone optimizes, the number that shows up in sustainability reports, the figure that earns applause at conferences. ...
I went looking for sustained-load benchmarks comparing MoE and Dense coding models on consumer GPUs. Not demo bursts on a Mac Mini — sustained autoregressive generation on real coding tasks, where architecture and interconnect are the only variables. I found plenty of one-shot numbers. Nobody had published the comparison that matters: same hardware, same quantization, same inference engine, MoE versus Dense, across GPU configurations. Methodology visible. Numbers verifiable. So I ran the tests. Dual RTX 3090s with NVLink, custom liquid cooling, a 6 kW isolation transformer feeding a double-conversion UPS. Not elegant, but thermally and electrically honest — sustained inference loads without throttling, no measurement fiction. The hardware details are below. ...
Last Friday, the White House ordered every federal agency to stop using Anthropic products within six months. The Defense Secretary designated the company a “supply chain risk to national security” — a label normally reserved for foreign adversaries like Huawei or Kaspersky. Anthropic’s crime: they refused to remove two safety guardrails from Claude before deploying it on classified Pentagon networks. No AI for mass domestic surveillance of American citizens. No fully autonomous weapons without human oversight. ...
In Brazil, when advising a customer on endpoint security, there was a mental model we never said out loud. The technical discussion would cover detection rates, false positives, memory footprint — the usual. But underneath it ran a question that never made it into the RFP: who do you want knowing what you’re doing? Russians or Americans? Kaspersky was the default for most of the market — and not because of ideology. Norton and Symantec had spent years earning their reputation for turning Windows machines into molasses, and McAfee was McAfee. Kaspersky worked. It was lighter, faster, cheaper. The fact that its telemetry flowed to Moscow rather than Langley was a feature, not a bug, depending on which side of the table you sat on. ...
The Concorde burned one ton of fuel per passenger to cross the Atlantic. One hundred seats. Three and a half hours. Mach 2. The most advanced commercial aircraft ever built — and every engineer who saw it wanted to believe it was the future. The 747 did the same crossing in seven hours. Four hundred seats. A quarter of the fuel per passenger. No afterburners. No sonic boom. No government subsidies keeping it alive. ...