VistaShares

This address is how we stay in touch with the professionals who follow our work.

VistaShares delivers innovative Liquid Alternative investment solutions for today's investors. VistaShares ETFs are actively managed by industry and investment experts, offering a number of distinct strategies. Supercycle® Growth Equity ETFs seek exposure to technology-driven economic Supercycles® that the adviser believes may offer long-term growth potential. Target 15® option-income ETFs seek to generate monthly income while complementing a core equity portfolio. Income is not guaranteed and will vary, and the Target 15® name refers to the strategy's objective rather than a guaranteed or promised return. Investing involves risk, including possible loss of principal. There is no guarantee that any investment objective will be achieved, and past performance does not guarantee future results.

About this domain

vistasharesmarkets.com belongs to VistaShares. We send email from it, and everything we publish sits on our main website at vistashares.com instead.

If a message from vistasharesmarkets.com reached you and you want to confirm it came from us, write to info@vistashares.com or call (844) 875-2288.

What the mail from here is about

Mail from this address is about the physical side of artificial intelligence: the chips, memory, networking, buildings, power and cooling that every model runs on. The software gets the attention. The hardware sets the pace, and it is built by a long chain of companies most people have never heard of, many of them far from the technology sector.

The writing here follows that chain from the accelerator down to the substation, because each link has its own lead times, its own constraints and its own cycle, and they rarely line up.

Where they fail to line up is usually where the interesting questions are.

What the build-out is made of

Training and running large AI models takes specialised chips, and those chips take a great deal else. Accelerators do the computation, but they are only useful when paired with high-bandwidth memory, fast networking between thousands of machines, and storage that can feed them. All of that has to sit in a building with enough power coming in and enough cooling to carry the heat away.

Each layer is its own industry. Chip design and manufacturing sit with a handful of firms and a smaller number of fabrication plants. Memory is a cyclical business with its own capacity decisions. Networking spans switches, optics and cabling. The buildings are put up by construction and engineering firms, and the electrical gear inside them comes from manufacturers whose order books reach years ahead.

Then there is power, which has become the slowest part of the chain. A large data centre can draw as much electricity as a small city, and connecting one to the grid means transmission lines, transformers and substations that take years to plan and build. In many regions the queue to connect is longer than the time it takes to put up the building.

So a chart of chip sales tells only part of the story. The build-out is paced by whichever layer is furthest behind, and that layer changes over time.

Each layer also has its own buyers and its own competitors. The largest technology companies design some of their own chips and buy others, build some data centres themselves and lease others, and sign long contracts for power with utilities and independent generators. A supplier's fortunes depend on which of those decisions go its way, and those decisions are reviewed every year.

That is why the same headline about AI spending can be good news for one part of the chain and indifferent for another. Following the spending through each layer, rather than treating it as a single number, is the only way to see who it actually reaches.

None of this is fixed. Designs change with each generation of hardware, and a layer that is plentiful today can become the constraint after the next product cycle, which is part of why spending forecasts in this area are revised so often.

For anyone trying to understand the exposure, it helps to ask which layer a company sits in, who its customers are, and how much of its revenue depends on this build-out rather than on everything else it sells.

Those questions have plain answers for most companies, and they are usually found in annual reports rather than in commentary about the build-out.

What earlier build-outs suggest

Large infrastructure build-outs have a long history, and many follow a recognisable pattern. Railways, electrification, telephone networks and the fibre laid during the early internet were all real, all transformative, and all went through periods when capacity was built faster than demand could use it. The technology was right; the timing of the spending was not always.

That history does not say the current build-out will follow the same path. The buyers today are among the largest and most profitable companies there are, and much of the spending comes from their own cash flow rather than borrowing. But it does say that the useful questions are about utilisation and timing rather than whether the technology matters.

Two things are worth watching. Whether the capacity being built is being used at a rate that justifies the next round of spending. And whether the suppliers' order books are growing with real orders, or with forecasts that have not yet been committed.

Neither is easy to read from outside, and both matter more to the suppliers than to the buyers, because the suppliers feel a slowdown first and hardest.

Depreciation is another lens. Accelerators and servers lose value quickly as newer generations arrive, so the economics of a data centre depend on how fast its equipment earns back its cost before it is replaced. That shapes how long buyers are willing to keep spending at a given pace.

Buildings, power connections and grid equipment last far longer, which is one reason spending on the physical plant tends to be steadier than spending on the chips inside it, and why suppliers to each part of the build see such different rhythms in their orders.

None of this predicts how the current cycle ends. It does suggest that a fund exposed to the build-out is more fairly judged over a full cycle of spending, including the stretches when orders slow, than over the years when they are rising fastest.

Watching the suppliers' own commentary is often as useful as watching the buyers'. Lead times, backlog and pricing tend to change before capital plans do, and they are reported every quarter by companies across the chain.

Those signals are imperfect, but they are public, and they tend to move before the larger decisions do.