Owning the AI trade before the market named it 

Pzena Investment Management’s Rakesh Bordia says the portfolio bought its best artificial intelligence holdings years before anyone attached the label to them.

Most portfolios express a view on artificial intelligence through a similar list of names and under similar frameworks. Rakesh Bordia is portfolio manager at Pzena Investment Management. He argues that a value process arrives at the same exposure by a different route, and leaves it a good deal earlier.

Cheap before it was AI

Elite Material is a Taiwanese producer of copper clad laminates, which form the base of printed circuit boards. It was in Pzena’s emerging markets portfolio well before anyone thought of it as an AI holding. Bordia says less than 1 per cent of the business was AI related when the firm bought it. The company held something close to a monopoly in its niche.

Lite-On Technology, which makes power systems for data centre servers, arrived the same way. Pzena didn’t buy either for a theme. Both were good businesses trading in the cheapest quintile of their universe.

“As value investors, we have to identify these opportunities well before the market is willing to give them credit for,” Bordia says.

Handing over the winners

Once the market did give them credit, the shares ran and both left the cheapest quintile. Under Pzena’s process, that means they stop being candidates. Pzena sets position sizes bottom up on valuation, risk and reward and downside risk analysis. It makes no reference to what a stock weighs in the index.

That produces a portfolio that looks nothing like the benchmark against which managers measure most emerging markets money. At the moment, it also produces a substantial underweight to the technology supply chain.

The practical effect is that the portfolio tends to own the enablers of a theme early and cheaply. It then hands them over at the point most investors have become comfortable buying them.

The memory question

Memory is a case study where Bordia has a different frame to many other investors. He sees DRAM as a commodity business. In his view, a chip Hynix makes is not meaningfully different from one that Samsung or Micron makes.

The memory that AI applications use is a different product. But on his numbers, it accounts for only about a quarter of demand.

Supply is the other half of the case. Each of the incumbents is adding capacity over the next two to three years. New Chinese competition has also arrived with the listing of ChangXin Memory Technologies in Shanghai.

Demand is responding as well. Bordia points to Apple raising prices on its non-iPhone devices. He also points to notebooks on US retail shelves selling at the same price and specification as a year ago, while carrying significantly less memory.

“I would have never guessed that hardware memory names have margins higher than software companies”

The conclusion he draws is that the current level of earnings is a pricing artefact of a near-term shortage. It is not a reflection of what these businesses earn through a cycle. He also concludes that the market has priced the shares for perpetual growth, in an industry with an unusually wide range of outcomes.

The perceived losers

The work has moved to the other side of the trade. Bordia says opportunities are opening up in what the market has decided are the AI losers. They sit in information technology services, in advertising and in parts of software. In those areas, the assumption is that the technology takes the earnings away rather than adding to them.

His summary of the past five years is that AI has created plenty of opportunity for the portfolio, rarely at the moment the market agreed it was one.

That is the same question Pzena asks of any cheap stock, whether the problem is temporary or permanent. This is the harder version of it.

The market’s record on disruption is mixed rather than poor. It was wrong about how quickly online retail would finish physical retailers. It was right about classified advertising and newspapers, where the multiple never recovered because the earnings never did.

Nothing in the current evidence settles which of those the advertising and services names resemble.

What keeps the approach coherent is that none of it starts from a view on AI at all. “We don’t invest in themes,” Bordia says. “We invest in underlying fundamental businesses.”