The 17th ACM Symposium on Cloud Computing convenes in Singapore on November 18–20, 2026. Anyone hoping for a preview of the papers will have to wait: SoCC runs a two-round submission process, and the second round’s notifications do not go out until September 26, 2026. The second-round deadline itself closed on July 14, with the response period running September 10–12. Round one, whose abstracts were due February 6 and whose notifications went out April 29, has produced acceptances — but the assembled program is not yet public.
What is public, and is arguably more informative than any individual paper, is the call for papers. A conference’s topic list is a negotiated document. It states what a community has agreed to treat as its own problem, and reading this year’s list against the field’s history shows a discipline that has substantially redefined its boundaries.
The Venue’s Unusual Mandate
One structural detail explains a lot about SoCC’s scope. It is the only conference co-sponsored by both ACM SIGMOD and SIGOPS — the special interest groups for data management and for operating systems respectively.
That dual sponsorship is not a bureaucratic accident. Cloud computing is precisely the domain where the two traditions became inseparable: you cannot reason about a distributed query optimizer without reasoning about the storage and scheduling substrate beneath it, and you cannot design that substrate well without knowing what the data systems above it are trying to do. Most venues force a choice of primary audience. SoCC’s charter is the intersection, which is why its topic list spans “distributed/parallel query processing and optimization” and “operating systems and system support” as peers rather than as neighbors.
What the Topic List Now Contains
Reading the twenty-seven topics of interest, three shifts stand out.
Machine learning appears twice, in opposite directions. The list carries both “systems for machine learning training and/or serving” and “machine learning for clouds and cloud-based systems.” These are genuinely different research programs. The first treats ML as a workload with awkward requirements that infrastructure must accommodate — the concern driving most of the scheduling and accelerator work of the past several years. The second treats ML as a tool for building infrastructure: learned schedulers, learned query optimizers, learned cache admission policies, failure prediction. Listing both as first-class signals that neither direction is a novelty item anymore.
Sky computing is now a named topic. This is the multi-cloud research agenda — the argument that the useful abstraction is not any single provider’s cloud but a layer above all of them, where workloads move to whichever provider offers the best price and capacity for a given job, mediated by a broker rather than by a procurement contract. It is a direct response to lock-in, and its appearance as a standing topic rather than a workshop theme marks it as a durable line of inquiry. It is also the research counterpart to the regulatory pressure documented in the EU Data Act’s cloud switching provisions: the law is trying to make switching a right, and sky computing is trying to make it a mechanism.
Economics has entered the systems agenda. “Data markets and data economy” sits alongside “energy efficiency, sustainability, and management” and “administration, service level agreements, and manageability.” A decade ago these would have read as adjacent business concerns rather than systems research. Their presence reflects something real: at current scale, the binding constraints on cloud infrastructure are frequently power availability, carbon accounting, and contractual structure rather than any technical limit. Research that ignores those constraints describes systems nobody can deploy — a point the field has absorbed as data center power constraints became a planning reality rather than a forecast.
The list also retains topics that have quietly become core rather than speculative: confidential computing, cross-datacenter data management, edge and fog computing, multi-tenancy, serverless and microservice programming models, and virtualization. Blockchain systems and Byzantine fault tolerance remain, though notably framed alongside “decentralized ledgers” rather than as a standalone enthusiasm.
Why Singapore Is a Reasonable Place to Hold This
The location deserves a comment beyond logistics. A substantial share of the questions on SoCC’s topic list — cross-datacenter data management, sovereignty-adjacent concerns embedded in confidential computing and multi-tenancy, edge infrastructure, energy constraints — are questions whose answers differ by jurisdiction and by region. The Asia-Pacific region concentrates several of them at once: dense urban edge deployment, acute power and land constraints on datacenter siting, and a patchwork of data-residency regimes that do not resemble either the American or European models.
Research communities tend to generalize from the infrastructure they can see. A cloud research agenda formed entirely at venues in North America and Europe will encode assumptions about power availability, regulatory structure, and network topology that hold in those places. Holding the conversation elsewhere is a modest corrective, and for a field whose central abstraction is supposed to be location-independent, a useful one.
What to Watch on September 26
For anyone tracking this venue rather than attending it, the second-round notifications are the date that matters. Two signals will be worth reading in the accepted set.
The first is the ratio of ML-systems papers to everything else. OSDI ‘26 in July skewed heavily toward LLM serving and inference infrastructure. If SoCC’s program shows the same concentration, that indicates the field’s attention is genuinely consolidated rather than venue-specific — which has real consequences for anyone whose infrastructure problems are not AI-shaped, because the tuning advice and platform defaults that follow research attention will not be aimed at them.
The second is whether sky computing produces deployed-system papers or remains largely architectural. Multi-cloud portability has a long history of compelling designs that founder on the practical reality that providers differ in ways abstraction layers cannot fully hide. Papers reporting real cross-provider deployments, with honest accounting of what leaked through the abstraction, would move the topic from aspiration to engineering. That is the distinction worth watching for, and it is the one the program will actually settle.