A new wave of buy signals across semiconductors

I’ve posted a recording of my update last week above, I’ll host another at 11am this Thursday, please sign up here.

For the second time this year, our risk models have generated a wave of buy signals across semiconductors, right after a major deleveraging event and accompanying sell-off that has pushed valuations right down.

There have been a number of important advances, notably OpenAI’s Astra model, and some strong company reports, so it may turn out that a three month sell-off and major deleveraging event in South Korea and the US (notably Situational Awareness, and Citadel’s takeover and prompt liquidation of their positions) is enough to reset a booming industry for the next move.

Of course, time will tell!

I’ve run through some company reports below, the message is fairly unanimous. The leaders are posting 50-100% or more top line growth, company earnings are growing faster, and there are growth vectors within growth vectors that look particularly promising.

Worth noting we are risk-managing all of the positions below - I personally wouldn’t invest in such cyclical industry any other way (witness the 50% falls since June!)

Nvidia

Nvidia posted >100% top-line growth and over 120% EPS growth year-on-year. Nvidia’s forward multiple contracted to 14x at the recent low.

We’re aware that Nvidia risks losing market share to hyperscaler insourcing and new chips like OpenAI’s Jalapeño (see my prior note for the strongest negative arguments). But the broader stack is formidable, and a 50-100% appreciation over the next 12–18 months is an entirely reasonable outcome. It simply requires that forward multiple to expand to 21x, or growth maintaining at the company’s own near-term guidance of 70% year-on-year.

Broadcom

Broadcom reported 86% year-on-year revenue growth and 96% EPS growth, though guidance estimates missed marginally. The stock is down 28% from June, and a forward PE of 18 is modest.

The company is forecasting semiconductor revenue to grow from $58 billion in FY26 to $230 billion in FY28.

Broadcom designs TPUs for Google, and Google’s results were exceptional, with cloud revenue growing 82%. Anthropic is another major customer, as is OpenAI, as they designed their Jalapeño chip with Broadcom. Broadcom is also involved in Meta’s MTIA chip which is moving into production, so Broadcom is a way (along with Marvell) to invest in a megacap benefitting from hyperscaler vertical integration.

We don’t currently own Broadcom, but I we’ll certainly own it in the next uptrend.

Amphenol

Amphenol (~$200 billion market cap) has a portfolio across copper, fibre, power and sensors, it seems to fly under the radar and I rarely hear investors discuss it. The company posted 30% organic growth and 67% EPS growth, and no customer accounts for over 10% of revenue so it’s a broad beneficiary of the capex trend.

Celestica

Celestica fell over 40% since 1 June, and near the lows announced an equity raise of ~10% of its market cap. We closed on the way down, but again, I’m sure this is something we will own again.

Their latest report showed 62% revenue growth and 83% EPS growth, another fast-grower where earnings per share are outpacing blistering revenue growth. Celestica is the board, rack and systems partner for OpenAI’s Jalapeño.

Celestica is more of system-design, integration and manufacturing company, though they are developing their own hardware.

Memory: Micron, SK Hynix and Samsung

Memory is the most intriguing part of the complex, with some of the strongest results and lowest valuations.

The market is digesting new Chinese competition and the extent to which long term contracts and the sheer scale of demand mean that this time, the memory cycle will be different. The DRAM ETF, which consists of these three plus hard-drive makers like SanDisk and Seagate, dropped 45%, again mostly in July.

It looks like the sector simply became too hot. These companies were Korean favourites, first benefiting, then losing, from the immense leveraging and then deleveraging of Korean (and global) speculators.

That leverage is largely cleared now, after related triple leveraged ETFs dropped >80% and cut their holdings to an even greater amount (to a first approximation, three times).

These are also a rare example of deep value (<5x GAAP PE) crossing over with ultra high growth. I think the reason value managers struggle with them is simply because they are up so much, and value investors are bent to buy things that are down, but who knows really. The timing is certainly interesting now, with a global deleveraging event combining with rapidly expanding and lengthening commitments from customers.

We can track DRAM pricing live:

Apple is starting to make higher-RAM computers available again, at exceptionally high prices. Waiting lists suggest these prices are here to stay, as the incredible new intelligence capabilities available locally are a great reason to buy.

In the past, only graphics designers, video editors, and the like had reason to pay up for a specced out Macbook Pro or Studio. Now there is a reason for everyone to have more memory and a faster laptop in front of them.

But back to memory, to stay at GAAP P/Es of under 5x this oligopoly would have to disappoint over the next few reports, and so far there is no indication of that.

Memory orders are also now increasingly on multi-year, fixed-price contracts. So the risk drops with every month that these companies continue to earn prodigiously.

Coherent

Coherent is part of the shift from electrical to optical connectivity within datacentres, supplying the lasers, optical components, and so on, that convert electrical signals into light. This offers vastly higher bandwidth capacity, lower power consumption and longer reach without electrical interference.

This is a growth trend within a growth trend.

As with the companies listed above, Coherent is growing fast (22% year-on-year), but EPS is growing faster (59%), with datacentres becoming a larger part of revenue. The company is forecasting another doubling of indium phosphide laser capacity in 2027. Coherent is more expensive than many of the other companies in this note above at a PE of 29x, but optics is clearly the future.

And I’m also convinced that optical, rather than quantum, computing is the next major leap forward.

These stocks pumped in the first half of the year and (partially) dumped in July and August.

Applied Optoelectronics

AAOI was one of our profitable positions earlier in the year that we scaled out of. After falling 50%, there may be another opportunity.

AAOI has higher ‘torque’ (an annoying new fundie favourite word) to the optical theme. Revenue is growing 87% year-on-year and 27% quarter-on-quarter. AAOI has a more focused opportunity than Coherent and a larger, clearer rampup, but aggressive management targets made the stock both a retail favourite and a battleground.

Management is targeting a >US$5 billion mid-2027 revenue run rate… on what’s currently a $9 billion market cap. Some believe management is being a little promotional here, but as always time will tell. There’s certainly enough spice in this one for a momentum trader, but best to stay on the right side!

Outlook

So in the last few months we saw the Aschenbrenner collapse and a major semiconductor deleveraging event in South Korea and globally. The sector dropped over 20% with over 50% falls in many of the small and midcaps. Random so-called ‘bottlenecks’ had a harder time, but these mostly do not fit what we are looking for, as we really want sustained performance we can see in the numbers today, rather than something forecast. It’s just a preference.

Many semiconductors stabilised in August, and now after more than three months since the late June top, may be primed for a rebound, especially now sector leverage has dropped. It’s unlikely the next rally will be as sharp, however, given so many were burned and regulators are cracking down on the leveraged ETFs (and funds) that drove much of the gains in the first few months of the year.

But during this deleveraging event, many of these companies posted blow-out revenue and EPS numbers, and ultimately this is more important. For a sell-off to morph into a sustained downturn, it would likely have to be combined with a pull back in hyperscaler spending, likely driven by a pull-back in AI spending, and as of today, we are simply not there (witness Google’s 80% cloud business growth year-on-year).

Even if some of the bear case concerns I wrote about a week ago play out, that will still benefit many companies in the sector, as lower prices will result in higher demand, and there is strong evidence from the success of OpenAI’s cheaper models that this will be the case here, given the explosive growth of their lower tier Luna model once they cut prices.

In our funds, we’re spreading a large sector allocation across large caps like Nvidia, optical specialists and memory.

Each position is being carefully risk-managed, and while a number have just fired off buy signals, we could be sellers as soon next week, so bear that in mind!

As you might expect, we were net sellers in July and August, but that is changing, and (this is purely a guess from the evidence above) there may be a real chance now to catch the next major move in semis, right when half the market is congratulating themselves for not being as bullish and over-extended as Leopold Aschenbrenner.

Of course, the sell-off could resume, we could take small losses and close positions one by one, and this note will look silly. We’re ready for both scenarios.

Thank you for your support

Mike