Where scientific infrastructure decisions get made.

If you run a scientific computing centre, this is the one week of the year when vendors, peers and agencies are in the same room at once. A week here settles conversations that would take six months by email.

What is announced here shows up in the proposals you will receive the following quarter. Following it means entering the negotiation knowing what already exists and what is still a promise.

Europe is settling AI infrastructure sovereignty right now, and it is the same debate that reaches Latin America next. Understanding the European path means discussing the national programme with a repertoire the rest of the room does not yet have.

If you will be specifying interconnect in the next two years, what is demonstrated here defines your options. The copper reach that shrinks with every generation is decided in this room, and lands in your budget eighteen months later.

It is the science underpinning the component you will buy three years from now. If your planning horizon is longer than your purchase cycle, this is where it gets informed.

If OFC is the market for optics, this is the science feeding it. For anyone who needs to tell product from laboratory promise, the difference becomes visible in this room.

It opens the year with the twelve-month view in optics. If your annual plan closes in the first quarter, this is the information that arrives in time to be part of it.

It is where public supercomputing policy meets engineering. If your institution depends on public funding to grow, the argument that works in the call for proposals usually starts here first.

It is the rare event that shows the aftermath: how a public supercomputing consortium allocates access and measures results. Anyone arguing for a governance model for shared infrastructure finds the documented precedent here.

The rack of 2030 is decided here. Whoever follows it specifies cooling a generation ahead and rebuilds the room once, instead of learning too late that the floor cannot carry what the science now demands.

It is where open designs leave paper and enter production. If your schedule depends on lead time, the signal of real availability shows up here before it shows up in a quote.

The accelerator shows its insides before the datasheet exists. For anyone who has to defend a generational upgrade to a board, that is the difference between arriving with engineering and arriving with the wrapper around it.

Not every bottleneck is compute. If your cluster delivers less than the arithmetic promised, the explanation usually sits in memory bandwidth and the speed of feeding data, which is what this conference is about all year.

It is the European state of the art in the physical medium. Anyone deciding between copper, active optical cable and transceiver with fibre finds the technical data here before it becomes a sales argument.

An entire track on the crossing of photonics and AI, which is exactly where the cost per port of your next cluster will be decided.

It is the technical source of the promise to cut interconnect power and cost. Anyone judging maturity timelines finds the real schedule here, not the commercial one.

Coverage complementary to the January edition, weighted towards sensing. Relevant to anyone instrumenting an experiment, not only to anyone wiring a datacentre.

The crossing of quantum and photonics is where the promise meets the component that actually exists. Useful for telling what is already engineering from what is still physics.

Research at a stage before product, in the city that is the historical cradle of the optics industry. For long-horizon planning, it is an early and faint signal.

Here the number is measured, not announced. When you distrust a datasheet figure and need someone with no stake in the sale to confirm it, the independent measurement is usually in these proceedings.

Virtual and low cost. If your problem is computing next to the instrument rather than in the datacentre, this community discusses constraints the large conferences ignore.

A window on Asian optics research, which today sets component availability and price with more weight than any other region.

Asian investment in photonic networking for AI is what moves your transceiver lead time the most. Following it here anticipates scarcity; it is not geographic curiosity.

It is the deepest step of the chain: what is fabricated here determines what will exist to buy in five years. Relevant to anyone planning infrastructure on a decade horizon, not a budget cycle.

The largest drug-discovery market decides here what it will demand of compute over the next twenty-four months. Whoever sizes infrastructure for pharma hears the requirement before it becomes a request.

It is where medicinal chemistry meets computational chemistry. If your cluster serves both groups, this is where you see where their needs converge and where they fight over the same resource.

The computational chemistry division runs inside this meeting. One trip covers the community that consumes double-precision compute intensively, a profile opposite to AI and poorly served by whoever only talks about AI.

If there is one place where the buyer of drug-discovery infrastructure is gathered, this is it. It is neither a science congress nor a technology fair: it is precisely the border between the two, which is usually where the decision stalls.

Each sequencer generation redefines the data volume of the pipeline. What is announced here becomes a storage and compute requirement eighteen months later, and whoever heard it early reaches the budget without a shock.

Population genomics is the workload growing fastest in storage demand. If your institution has taken that path, what is discussed here defines how much disk you will need to buy before you need it.

Whoever actually specifies the genomics cluster is the bioinformatician, not the buyer. This is where they say, among peers, what is really missing from the machine they already have.

It is the methodological reference for reproducible pipelines. Anyone who must prove data provenance to an ethics committee or a funding agency finds here the standard already accepted abroad.

This community consumes double-precision compute in volume, and looks nothing like an AI workload. Whoever sizes a cluster listening only to the AI crowd delivers a machine chemistry cannot use.

It is here so you do not search in vain. The cycle is triennial and the next edition is in 2028, which also says something about this community's rhythm: it plans in years, not quarters.

Small format, deeply technical. It is where methods are born, years before they become compute demand, and therefore the earliest signal available in this vertical.

Computational physics is the silent vertical. It consumes double precision in volume, almost never shows up in AI conversations, and is often the single largest user of a university cluster.

Useful quantum depends on classical infrastructure. This is the event where that boundary is discussed in engineering terms, without the commercial promise that usually accompanies the subject.

It is the quantum series that speaks the language of whoever signs the cheque. Useful for calibrating what is already a business argument and what is still laboratory talk.

Europe funds quantum with public money and demands results. Seeing how that argument is built there helps anyone who must defend the same investment here.

It is the only one in the quantum series with genuinely executive language. If you must explain quantum to a board without promising what does not exist, the vocabulary is here.

Climate modelling is one of the oldest and heaviest high-performance workloads. Anyone who inherited a climate cluster knows it looks nothing like what the AI industry describes.

The climate vertical has a strong Brazilian presence and is underserved by vendors. If your institution does forecasting or environmental modelling, your peers are here and the infrastructure comparison happens in the corridor.

The main conference is pure machine learning. It is in the workshops that the scientist with a real compute problem shows up, and it is with them that the infrastructure conversation makes sense.

If autonomous agents really do start running experiments, your cluster's workload profile shifts from scheduled batch to continuous demand. This is where you can see how close that is.

Context, not decision. What appears here only becomes an infrastructure requirement years later, but whoever follows it recognises the requirement when it arrives instead of being surprised by it.

The submission volume is the data point itself: this community grew faster than the infrastructure that supports it. Whoever operates an academic cluster feels that mismatch in the job queue.

Computer vision is AI's entry point into microscopy, medical imaging and remote sensing. If your institute generates images in volume, the compute demand starts here.

It is the conference with the field's long memory. Anyone wanting to tell what is new from what is returning under a new name finds the whole timeline here.

Every Brazilian scientific society in one place, alongside the ministry. If you need to understand where national science policy is heading before it becomes a funding call, it is the densest week of the year.

It is the entry point for the drug-discovery and computational chemistry verticals in Brazil. The group running simulations on a workstation today is in this room, and it is the one that asks for a cluster tomorrow.

Biochemistry and molecular biology are migrating to computational pipelines. Anyone serving this group needs to understand that their bottleneck is rarely raw compute, and almost always data.

Agricultural and animal genetics is a Brazilian vertical that almost nobody serves with dedicated infrastructure. The sequencing data volume in these groups is already comparable to human genomics.

Clinical genetics works with sensitive patient data. Here the infrastructure conversation necessarily runs through data sovereignty and legal compliance, not only performance.

It is the highest density of decision-makers in the entire calendar. The directors of the region's national centres are all in the same room, and the partnerships formed here typically precede formal institutional agreements by years.

If your institution depends on the federal academic network, this is where the next two years of investment priorities are debated, before they become a funding call.

The Brazilian computing society is opening its agenda to high performance and research cloud. That is a market signal, not just a calendar entry: the community that trains the buyer is changing subject.

This is the community that specifies the Brazilian academic network. Understanding what is discussed here means understanding, in advance, what the federal network will demand of your connection.

The future demands of the national academic network are debated here before they become a specification. For anyone operating a point of presence, it is the difference between planning and reacting.

It is the core of the Brazilian academic AI ecosystem. Anyone looking for the group that will need a cluster in the next two years will find it in this conference's author list.

Four conferences in the same week and the same city. One trip covers practically the entire Brazilian computational intelligence community.

A Portuguese language model requires local training, and local training requires a local cluster. It is the emerging vertical with the most direct sovereignty argument available today.

Mining at scale is an input-output workload, not a compute one. Whoever confuses the two buys surplus processing and finds out too late that the bottleneck was storage.

Health is where sensitive data and performance demands meet. Anyone serving a teaching hospital must solve both at once, and this is where you see how others solved it.

It is the Brazilian architecture and high-performance community, with international recognition. In 2027 it returns to Brazilian soil, which completely changes the cost of taking part.

Training is where the future buyer learns the vocabulary. Whoever only shows up at purchase time negotiates with someone who learned the subject from somebody else.

The Northeast concentrates expanding research centres and little local infrastructure supply. It is the region where the gap between demand and installed capacity is widest.

A regional school is a small, direct-access format. If you want to know what the groups in your region actually run, rather than what they say they run, this is where.

São Paulo concentrates the country's highest density of research institutions. It is the region where an architecture decision is most likely to become a reference for the others.

It is the gateway to Spanish-speaking America. Fifty years of history mean the region's collaboration networks run through here, and whoever arrives from outside needs those networks before needing anything else.

The Spanish-speaking academic network has its own topology and constraints, different from Brazil's. Anyone wanting to operate on both sides needs to understand both.

Regional networking context. Useful for understanding why the last mile to the institute often costs more than the entire cluster across much of the region.

Regional academic AI context outside Brazil. The attached summer school means the region's next generation of researchers is in the same week and the same place.

An early signal of hardware availability. What is announced here usually indicates, months in advance, what will be missing from your quote.

A cluster does not fail for lack of processors, it fails for lack of a room. Here they discuss the part of the project that almost never appears in the proposal and that decides whether the machine turns on.

The Japanese vision of an all-photonic network is the horizon of interconnect. It is not a decision for today, but it defines where the physical medium is heading over the next decade.