Loading today’s stories
Product1 distinct publisher2 min readUpdated
The deal puts CUDA kernel-level continuous profiling inside an OpenTelemetry-native platform. It also attaches that capability to a database replacement the buyer did not ask for.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The GPU piece of this deal is narrower than the announcement language suggests, and more useful for being narrow. Continuous profiling itself is not scarce; many observability platforms already ship some version of it [9]. What Polar Signals brought is production profiling of Nvidia CUDA workloads, covering both training and inference, resolved down to individual GPU kernels [2].
The automation stops short of that. AutoTune, the Agent0 capability Dash0 is still building, is described as analysing profiling data on a schedule to find CPU and memory improvements, then opening a pull request with the supporting evidence and leaving the merge decision to a developer [5]. Set that against the kernel-level GPU data the acquisition actually adds [2], and the evidence-to-pull-request loop does not yet reach the hardware that dominates the bill [1].
How the deal started matters to the read. By Frederic Branczyk's account, the conversation began with Dash0 CEO Mirko Novakovic describing the cost, performance and operational trouble he was having with ClickHouse, and Branczyk noting that Polar Signals had already built several iterations of a database for the same shape of workload [10]. Dash0 is taking the Great Lakes engine and the engineering team, and plans to hold metrics, logs, traces and profiles in that single engine once ClickHouse is out from under SignalStore [3].
That is the connection between the storage purchase and the GPU claim. Great Lakes exists because profiling generates high-cardinality data [3], and per-kernel attribution is about as high-cardinality as application telemetry gets. The profiling pitch and the database pitch are one pitch.
Two things bound how durable the advantage is. OpenTelemetry Profiles reached public alpha in March, adding profiles to the data model alongside traces, metrics and logs, and Polar Signals engineers contributed to that work [8]. A buyer can therefore ask for profiles as a standard signal rather than a vendor feature, which runs both ways: Dash0 gains people shaping the spec, and the spec slowly erodes what is proprietary about carrying profiles. Second, the money behind this is recent. Dash0 was founded in 2023, raised $110 million in March at a $1 billion valuation while saying some of it would fund strategic acquisitions, and reports more than 750 customers; terms here were not disclosed [7]. Parca stays open source and maintained [4], which keeps the on-ramp free and the storage engine the paid part.
Agent0 could already correlate telemetry with source code and prepare pull requests before any of this [11]. Profiling adds runtime evidence to those investigations, not autonomy. A human still signs off [5].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Dash0 announced it acquired Berlin-based continuous profiling specialist Polar Signals, adding continuous profiling to SignalStore, Dash0's OpenTelemetry-native data platform.
Polar Signals offers continuous profiling of Nvidia CUDA workloads in production; Dash0 said the technology can profile both AI training and inference workloads with visibility down to individual GPU kernels.
The deal includes Parca, Polar Signals' open source continuous profiling project, which the company said will remain open source and continue to be maintained.
Financial terms of the deal were not disclosed. Dash0, founded in 2023, raised $110 million in March at a $1 billion valuation and said it would use some of the funding to pursue strategic acquisitions; the company now has more than 750 customers.
OpenTelemetry Profiles, the project's profiling specification, entered public alpha in March, extending the OpenTelemetry data model to include profiles alongside traces, metrics and logs; Polar Signals has contributed to that work.
Continuous profiling has been in the developer toolbox for years and many observability platforms now include some form of it; profiling of Nvidia CUDA workloads in production is described as a less common capability.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single vendor-sourced account, no independent verification
All substance comes from one devops.com article built on Dash0's announcement plus the acquired founder's own write-up. The acquisition and the capability descriptions are attributable, but there is no second publisher, no customer testimony, no benchmark for the CUDA kernel-level profiling claim, and no artefact demonstrating the Great Lakes or AutoTune roadmap items.
Buyer scale disclosed, integrated capability unshipped
Adoption evidence exists for the acquirer's platform and for the surrounding standard, not for the thing the story is about. Dash0 reports 750+ customers and OpenTelemetry Profiles reached public alpha in March, but no source shows any user of the merged profiling stack, GPU profiling in a customer environment, AutoTune, or the MCP profiling interface. Parca has an existing open source user base implied by the continuity commitment, but no numbers are given.
Capability framing runs ahead of shipped, verified work
The framing that GPU kernel profiling stops being a specialist purchase sits on an announced acquisition, an unshipped AutoTune feature, an unscheduled ClickHouse-to-Great Lakes migration, and an alpha-stage OpenTelemetry profiling spec, all reported by one vendor-facing outlet. The gap is moderate rather than severe because the acquisition itself, Parca's status and Agent0's existing pull-request behaviour are concrete.
Vendor announcement plus acquired-founder narrative
The article follows an acquisition announcement, attributes capability and roadmap claims to Dash0 and to Polar Signals' founder, and reproduces the deal's origin story as told by the participants. Dash0 has a direct interest in showing that a $110M raise at a $1B valuation is being deployed into strategic acquisitions, and in reframing its own ClickHouse cost and operational pain as a strategic storage upgrade. No adversarial or customer voice is present.
Low: one publisher, mostly forward-looking claims
Confidence is limited by a single-publisher cluster in which the load-bearing forward claims (Great Lakes replacing ClickHouse, AutoTune, MCP profiling access) are unshipped intent. The base facts of the acquisition, funding history, customer count and Parca's status are consistently reported and reasonably firm, which keeps confidence from being lower.
leadership
ClickHouse buys Langfuse, turning a neutral tracing layer into someone's roadmap1 distinct publisher
build
The agent asks, the gateway decides: why read-only is not a security boundary1 distinct publisher
build
SigNoz self-hosting is no longer a compose file: ClickHouse 25 breaks the quickstart1 distinct publisher
build
LoreKit puts agent memory in Markdown files you can grep, not a vendor's database1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 21, 2026