GPU & Compute

8 analyses from Devence Lab

September 13, 2026 · 2 min read

The federal pass on data centre pollution does not reach the states that matter

Former EPA officials warned this month that weakened federal rules will let AI data centres pollute more freely. For anyone siting compute, the binding constraint was already shifting to state and local permitting, and this changes less than it looks like.

September 11, 2026 · 2 min read
The bottleneck stopped being GPUs. It is now the grid.
Gartner projects 40% of AI data centres will be power-constrained by 2027. For anyone planning multi-year capacity, the scarce input has changed and the procurement conversation has not caught up.
September 10, 2026 · 2 min read
Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.
Microsoft contracting 10.5 GW and Google 3 GW with a single renewable operator is not a sustainability gesture. It is a hedge against the public grid being unable to supply.
September 9, 2026 · 2 min read
Inference efficiency stopped being a cost line and became a capacity strategy
When power is the binding constraint, every watt saved per token is capacity you did not have to contract for. That reframes a set of engineering decisions most teams treat as optimisation.
September 8, 2026 · 2 min read
Seventy percent of the grid is near end of life. AI arrived at the worst possible moment.
The infrastructure being asked to absorb unprecedented concentrated demand is simultaneously due for replacement. Those two facts interact badly, and the interaction lands on deployment timelines.
September 6, 2026 · 3 min read
Rubin's real claim is a 10x cut in the cost of a token
NVIDIA's next platform is in full production with seven chips and five rack systems. Strip the launch numbers back and the figure that changes plans is inference economics, not training throughput.
September 5, 2026 · 2 min read
Positron raised $875M on a bet that memory bandwidth is the wrong constraint
The Asimov chip drops high-bandwidth memory for up to 2.3TB of LPDDR5X per die. That is not a cost optimisation. It is a claim about which workloads are going to matter.
September 4, 2026 · 2 min read
Confidential computing reached the GPU. Regulated AI workloads just got a new answer.
Hardware-backed isolation is extending from CPUs into GPUs, multi-GPU environments and agent workflows. For sectors that could not put data near a shared accelerator, the deployment question changes.

The federal pass on data centre pollution does not reach the states that matter

The bottleneck stopped being GPUs. It is now the grid.

Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.

Inference efficiency stopped being a cost line and became a capacity strategy

Seventy percent of the grid is near end of life. AI arrived at the worst possible moment.

Rubin's real claim is a 10x cut in the cost of a token

Positron raised $875M on a bet that memory bandwidth is the wrong constraint

Confidential computing reached the GPU. Regulated AI workloads just got a new answer.

We share findings with partners operating in the same constraint space.