GPU & Compute
29 analyses from Devence Lab
· 2 min read
40% lower LLM costs is a ceiling, not a cache business case
AWS reports up to 40% lower inference cost and 80% faster first tokens from distributed KV caching. Deployers should measure repeat traffic, cache cutover and isolation before renting the cache tier.

· 2 min read
Diamond wafers make package cooling a qualification test

· 2 min read
SageMaker's GPU slices move the cost test to workload shape

· 3 min read
TensorRT's global tuner makes inference speed an accuracy-gated release decision
· 2 min read
SageMaker's training sessions trade GPU orchestration for experiment accountability
· 3 min read
Qwen3 on Trainium tests hardware portability beyond one training recipe
· 3 min read
Cloudflare's 68% cheaper AI reviewer still needs quality-adjusted cost controls
· 3 min read
vLLM's 3x latency improvement comes with an earlier routing commitment
· 2 min read
NVIDIA's UK compute pledge makes agent safety an infrastructure question

· 2 min read
NVIDIA's open Run:ai shifts GPU fairness from scheduling to policy
· 3 min read
Falling token prices do not guarantee cheaper AI infrastructure
· 2 min read
A billion-dollar science pledge is a capacity question, not a deployment plan
· 3 min read
Shared GPUs need enforceable entitlements, not just higher utilisation
· 2 min read
40% more GPUs is a power-scheduling claim. Benchmark watts per token before buying racks.

· 2 min read
Eight times the solar power does not make orbital AI cheap. Test the physics before the fleet.

· 2 min read
A healthy GPU cluster can still fail the workload. Make readiness an active test.

· 2 min read
A robot's GPU is not always where inference belongs. Treat placement as a control loop.

· 2 min read
96% throughput changes the confidential-inference trade-off. Benchmark the protected path.

· 2 min read
A fast model behind a slow benchmark client is not a capacity result. Measure the load generator too.

· 2 min read
Faster grid connections are not a power contract. AI data centres need a flexibility SLO.

· 2 min read
3x fewer CRM errors is not a general model win. It is a case for narrow post-training.
· 2 min read
The federal pass on data centre pollution does not reach the states that matter
· 2 min read
The bottleneck stopped being GPUs. It is now the grid.
· 2 min read
Hyperscalers are buying gigawatts directly. That tells you what they expect the grid to do.
· 2 min read
Inference efficiency stopped being a cost line and became a capacity strategy
· 2 min read
Seventy percent of the grid is near end of life. AI arrived at the worst possible moment.
· 3 min read
Rubin's real claim is a 10x cut in the cost of a token
· 2 min read
Positron raised $875M on a bet that memory bandwidth is the wrong constraint
· 2 min read
Confidential computing reached the GPU. Regulated AI workloads just got a new answer.