Why we are long the build-out.

Gross & Cidecian Capital is run by two partners, Lars Gross and Christian Cidecian, both based in Zürich. We are a private house. We commit our own capital, privately. Our positions stay private, and so does the reasoning behind them. What we occasionally share, under News, is our thinking, never the book.

Most investment writing exists to raise capital. This does not. It exists to force us to be precise, and to give anyone who reads along a fair, undecorated view of how we actually think. There is no marketing layer between the thesis and the reader, because there is no product being sold here.

The framework that follows is long on purpose. The world we invest in rewards people who do the reading and punishes people who mistake a headline for an understanding.

Gross is an artificial-intelligence engineer. He builds production AI systems and develops the analytical models the house uses for pattern recognition and anomaly detection: inference stacks, retrieval pipelines, evaluation harnesses, the plumbing that turns research into signal. His edge is proximity. He works every day on the same infrastructure the house backs, and he can read the engineering reality of a chip roadmap or an inference benchmark the way most analysts read a balance sheet.

Cidecian covers the other half of the mandate. His background is in informatics and technology, with a specialisation in blockchain infrastructure and commodity analysis. He has spent years building expertise in digital assets and the commodity cycles that underpin the technology build-out, where on-chain data, supply-chain dynamics and macro flows create informational asymmetries that most equity-trained analysts do not see.

The division of labour matters. One of us can judge whether a piece of silicon will do what its marketing claims; the other can read the on-chain plumbing and the physical supply chain that decide whether a thesis has a foundation underneath it. A position has to convince both of us before it earns capital, and either of us can veto.

Beyond domain knowledge, we invest on the basis of analytical models we have built ourselves. They are designed to do what a human cannot do at scale: process large volumes of heterogeneous data, including technical research, chip roadmaps, patent filings, inference benchmarks, on-chain transaction flows, commodity supply data, options positioning and macro indicators, and then surface the anomalies and the structural divergences between what the market has priced and what the underlying data implies.

These are not dashboards with canned signals. They are purpose-built systems, trained on the specific domains we invest in, updated continuously and interrogated critically. The output is not a trade recommendation. It is a structured set of observations that either confirms or challenges our prior thesis, and every position must survive the disagreement between what the models surface and what we already believe.

That layer, combining domain expertise with machine pattern recognition, is what gives us the confidence to hold concentrated positions when the consensus is uncomfortable, and to step back when the data stops confirming the thesis. The models do not replace judgment. They sharpen it, and they make it harder to fall in love with a position the evidence no longer supports.

We are in the early years of the largest capital-deployment cycle of our lifetimes: an industrial build-out of compute, energy and data infrastructure to support models that will become economically indispensable across every white-collar function. Alongside it, digital assets are maturing from a speculative class into long-duration settlement infrastructure, and the commodities beneath both, from energy to metals to rare earths, are structurally undersupplied relative to the demand that artificial intelligence and electrification will place on them. Markets understand these trends directionally. They consistently mis-estimate their magnitude, their durability, and who actually captures the rent.

The house is concentrated where the engineering reality, the on-chain data and the commodity supply picture are furthest from the market narrative. That gap, between consensus and evidence, is where the mispricing sits, and it is the only place we are interested in committing capital.

It is worth being precise about the build-out, because the word has become a slogan, and slogans get mispriced in both directions. It is not a single product cycle. It is a multi-decade re-platforming of the global economy onto machine intelligence, and it has a physical footprint that most equity narratives still treat as an afterthought.

Start with the data centre. A modern AI campus is not a building full of servers. It is a power station with compute attached. The binding constraints are increasingly electrical and thermal rather than purely silicon: interconnect bandwidth, substation capacity, transformer lead times, cooling, and the multi-year queues to connect new load to the grid. Each constraint creates a layer of winners that the headline names do not capture, and each is governed by supply chains that cannot be willed into existence on a quarterly cadence.

Then consider the shape of demand. The market spent two years transfixed by training, the dramatic and capital-intensive act of building a frontier model. We think the durable story is inference: the recurring cost of actually running those models, which scales with the number of users, agents and workflows that depend on them. Training spend is lumpy and finite per generation. Inference is annuity-like, and it compounds with adoption.

Finally, follow the physical inputs back to the ground. Every accelerator requires copper, cobalt and rare earths. Every campus requires steel, aluminium and transmission equipment. After a decade of under-investment, the gap between what the build-out demands and what the physical economy can deliver in the next five years is, in our view, one of the most durable and least appreciated sources of asymmetry available.

For most of the last forty years, computing got cheaper and more abundant on a predictable schedule, and software ate the world by riding that curve. The current moment is different in kind, not just in degree. For the first time in a generation, the constraint on progress is not the cleverness of the software but the availability of physical capacity: chips, power and the buildings to house them.

The capital response has been enormous and is still early. The largest technology companies have shifted from asset-light models to some of the most capital-intensive enterprises on earth. History suggests that when an industry re-rates from asset-light to asset-heavy, the value does not stay where the attention is. It migrates down the stack, toward whoever owns the scarce physical input.

That is the lesson we carry from earlier build-outs, whether railways, electrification or fibre. The companies whose names defined each era were rarely the ones that captured the most durable economics. The picks, the rails and the scarce inputs often outlasted the celebrated platforms built on top of them. We position the house with that pattern firmly in mind.

Semiconductors are where engineering literacy pays the most direct dividend, because the marketing and the physics frequently disagree. The metric the market fixates on is arithmetic throughput, the raw number of operations a chip can perform. In practice, the binding constraint on modern AI workloads is very often memory: getting data to and from the processor fast enough to keep it busy.

This memory wall reshapes the investable landscape. It elevates the suppliers of high-bandwidth memory, advanced packaging and the interconnect that lets thousands of accelerators behave as one machine. It rewards power efficiency, because at scale the cost of an inference is dominated by energy, not by the sticker price of the silicon. And it punishes the naive assumption that the most famous accelerator automatically captures the most value at every layer.

We read this terrain through the engineering, not the press release. A roadmap that promises a generational leap is only as good as the packaging yield, the memory supply and the software maturity behind it. When those realities diverge from the narrative, the market is usually slow to notice, and that lag is where we want to be positioned.

If you trace the build-out far enough, you arrive at the power grid. Compute is, at bottom, a way of converting electricity into useful work, and the appetite of large-scale AI for electricity is on a trajectory the existing grid was never designed to meet. The scarce resource of the coming decade may not be the chip at all. It may be a firm, around-the-clock supply of power and the equipment needed to deliver it.

This is why we treat energy and electrical equipment as part of the AI thesis rather than as a separate macro bet. Transformers, switchgear, grid interconnection and reliable generation are the physical gating items on how fast intelligence can actually be deployed. The lead times are measured in years, the supply chains are concentrated, and the demand is no longer hypothetical.

The consequence is that some of the cleanest exposure to artificial intelligence is found in companies that have nothing to do with software and everything to do with moving electrons. We hold that exposure where the demand catalyst is tied directly to the build-out and the supply constraint is verifiable.

The case for holding digital assets has nothing to do with price predictions and everything to do with information. A public blockchain is, among other things, the most transparent ledger ever built. It lets a disciplined analyst observe accumulation, distribution and liquidity directly, rather than inferring them from lagging and easily massaged disclosures. That is a genuine edge, and it is why this part of the mandate sits with Cidecian.

We treat the major assets as monetary and settlement infrastructure with credible, inelastic supply, and we treat most of the rest with deep suspicion. A token with no on-chain substance behind it is a story, and stories are not collateral. When we hold crypto, it is because the on-chain reality supports the position, not because a narrative is fashionable.

The digital economy likes to imagine itself as weightless. It is not. Every model run, every robot and every data centre rests on a base of mined and refined material that takes years and serious capital to bring out of the ground. For more than a decade, that base has been starved of investment, partly through capital flight and partly through a regulatory environment that made new supply slow and expensive to build.

We are not generalist commodity traders, and we do not take a view on every cycle. We hold exposure only where two things are true at once: the supply constraint is real and verifiable, and the demand catalyst connects directly to the same build-out that drives the rest of the house. The thesis is patient by nature, because physical supply cannot respond quickly, and that slowness is precisely the source of the opportunity.

We are not value investors in the traditional sense, and we are not momentum traders. We treat valuation as a question of what the current price implies about the future, and then we ask whether the engineering and supply evidence makes that implied future too pessimistic or too optimistic. A high multiple is not a reason to avoid a position if the market is still underestimating the durability of the demand behind it. A low multiple is not a reason to own one if the business sits on the wrong side of the build-out.

Patience is part of the method, not a personality trait. Physical and engineering theses take time to resolve, and the market often pays for them in a single, sudden re-rating rather than a smooth ascent. We would rather be early and correct, and sit through the doubt in between, than be perfectly timed and wrong about the underlying reality.

A concentrated book is a deliberate trade. We accept sharper drawdowns in exchange for the chance to compound at a higher rate. Pretending otherwise would be dishonest. There will be periods in which the house falls further and faster than a broad index, and anyone reading along should size their own expectations accordingly.

What discipline buys us is not the absence of losses but the containment of them. Every position has a clear thesis and an explicit statement of what would prove it wrong. When that condition is met, the position is cut, not debated, not defended, not quietly re-justified. The hardest skill in investing is not finding good ideas. It is admitting, on time, that a good idea has stopped working.

We also separate conviction from leverage. High conviction earns a larger weight in the underlying asset. It does not earn reckless use of options or leverage, which are reserved for situations where the payoff is genuinely asymmetric and the cost of being wrong is bounded and known in advance. The objective is to survive the bad years intact enough to compound through the good ones.

We run the house from Zürich, and the choice is not incidental. Switzerland offers a stable legal framework, a serious culture of financial discretion and a long institutional memory for the difference between stewardship and salesmanship. It is a good place from which to think in decades rather than quarters.

We keep the house private and personal on purpose. Managing public money introduces a second master, the redemption cycle, that quietly distorts every decision and pushes managers toward consensus positioning and away from the uncomfortable, concentrated bets where the real asymmetry lives. By keeping our scale and our scope deliberate, we answer to the thesis and to the evidence, and to nothing else.

Honesty about reasoning. We are clear with ourselves about why a position is taken and what would prove it wrong, and just as clear when we are wrong. Our holdings and the positions behind them stay private; where it is useful, we share some of the thinking, never the book, under News.

We do not promise returns. We do not promise a smooth ride. Concentrated portfolios can and will draw down sharply. We hold ourselves to one thing: that what we say in public matches what we actually believe.

This site is not investment advice. It is not a solicitation. It is not an offer. The firm is not authorised or supervised by FINMA, and neither of us is a registered adviser in any jurisdiction. We run a private house and write about our thinking publicly because the discipline of writing is what keeps the reasoning honest.

If at some point in the future we work with outside capital, it will be privately, by invitation, on clearly disclosed terms and subject to eligibility. Until then, the most useful thing you can do is read, and, if it resonates, introduce yourself.