Tech Fund

Leading Edge Semi Shortages - a deep dive techinvestments.io/p/leading-edge…

The data center CPU market is getting crowded, with large players looking to gain share: $AMD $NVDA $INTC $ARM $QCOM $AMZN $MSFT $GOOGL The most obvious ways to play the growth and increasing competition in this market: $TSM $SNPS $CDNS $ASML and $MRVL (ASIC partner)

Bernstein stock picks in HBM: "We believe Samsung is best positioned to meet the higher speed required by NVIDIA. Micron is disadvantaged particularly with the choice of base die node. SK hynix also suffered a small delay with an issue in its base die no long ago, according to our checks. All these and having more ample clean room space than rivals should allow Samsung to gradually gain share in HBM ( Exhibit 42 ), despite that Samsung’s capacity allocation may favor conventional DRAM slightly over HBM to optimize profitability."

$INTC's new culture – chip has to work on the second tapeout, or you're fired

Memory pricing has fully exploded over the last quarters, driving a surge in memory revenues. Sustainable? We discuss our detailed outlook on the DRAM/HBM space: techinvestments.io/p/ai-and-memor…

$AVGO vs $MRVL - Evercore channel checks Both good companies, still holding both

A deep dive on power bottlenecks for the AI data center, and key stocks to overcome the power bottleneck techinvestments.io/p/power-bottle…

Software is well positioned to see a sustained rally Take $INTU for example, the leader in accounting software: 1. It has never been easier to start a business and those businesses will need accounting software 2. Businesses aren't going to vibe code their own accounting software. It's a small cost item and you need to make sure it's fully tax compliant. Links with govs are also needed so that they can automatically file your taxes. 3. I'm definitely not going to buy any software from vibe coded competitors. Vibe coded software is prone to security breaches for devs that don't fully know what they're now doing. Stick with the leaders that have secure, battle tested software. 4. Businesses are now subscribing to AI workloads inside their accounting software. For example, I subscribed to an AI which automatically classifies my costs, the result is that my P&L and accounting are now fully automatically generated. So, EPS will likely accelerate from here, same with the PE multiple which has lots of potential for recovery. Attractive risk-reward

"Shortages are spreading from Memory to wafers, CPUs, optics" - $ANET Gross margins will be coming down due to silicon price inflation. Expect all companies reliant on silicon and without sufficient pricing power to get margin headwinds. Who benefits? The strong players in the semi supply chain with pricing power are still trading on attractive valuations: $TSM, Semicap like $ASML, the DRAM triopoly (SK Hynix, $MU and Samsung), are some of the most obvious winners in silicon. $NVDA has locked in tons of capacity so will benefit. A big winner can also be $INTC. For years, there has been no demand for their Foundry, but this is changing now where even TSMC has not sufficient capacity.

@malaysiangovm I already bought power semis last year - still riding the bull market here

@techfund1 x.com/malaysiangovm/…

"Power semi content will 10x with the move to 800V racks. Lots of power semi content outside the rack with solid state transformers" - ON CEO We did the last review of this space in Feb, link below. Will update in the coming weeks as the outlook continues to get better $ON $IFX $WOLF $MPWR

UBS - $AMD is much less capacity constrained than $INTC, AMD server CPU revenue growing at 80% this year "The market obviously realizes that INTC’s guidance reads very favorably for AMD – particularly server CPU given commentary that implies INTC is undershipping the market by ~20%. The main question here is AMD’s supply, but our field work through the Q was indicating that supply for INTC’s parts was far more constrained than for AMD parts – so we think this sets up very well for AMD and we see revenue being guided up at least $1B Q/Q to the low $11B range (Street ~$10.4B). Server CPU: Intel’s CY2026 outlook implies its Data Center & AI segment growing ~40% Y/Y and we now see AMD server CPU growing as much as 80% this year with units up ~40-45% and pricing now up in the 20% range – helped in part by the ramp of the new Venice parts. From a competitive standpoint, our checks remain constructive. AMD’s CPU portfolio continues to compare favorably with Intel’s offerings, and the lack of meaningful timeline updates for Diamond Rapids and Coral Rapids reinforces our view that AMD should maintain a competitive advantage across the x86 ecosystem through C2026. Data center GPU: given our work on the rack timing, we reduce our C2026 DC GPU revenue slightly from $14.5B to $14B with some revenue pulled out of both CQ3 and CQ4, but overall we are raising our C2026/2027/2028 revs/EPS from $47B/$71B/$91B to $49B/$75B/$95B and from $6.65/$11.57/$15.94 to $7.65/$13.09/$17.62 driven almost entirely by data center CPU. We increase our PT to $455 from $310, based on a 35x avg peer multiple applied to C2027 EPS. Because AMD is in the very early stages of ramping into a huge XPU TAM, we would argue that the stock should command a higher multiple than peers and given our ~$18 EPS in C2028, think a ~$455 PT makes sense at ~25x PE multiple compared to peers’ average of 24x for C2027E."

Deep dive on Cerebras $CBRS techinvestments.io/p/cerebras-dee…

Market panicking about AI demand while GCP Gemini enterprise token consumption surged 60% over the last three months

For those looking at AI capex consensus forecasts and are concluding that we're near the end of the growth story for AI semis, well, consensus has been massively wrong for the last three years (chart below). Long term, one has to think about the % of OpEx that will be spent on AI. Even making reasonable assumptions here still shows massive upside in AI capex. There's $60 trillion in global OpEx, a lot more capex will be needed even if only 5% of that will be allocated towards AI workloads.

@fabknowledge He's probably using ASML guidance, although he's the Lasertec analyst, so might also have some insights from them

@techfund1 How many layers total? I do think it’s growing but seems a bit high

$ASML EUV layers per node (Goldman)

@DrFrederickChen Maybe vs originally planned at A14, but transistor scaling will require more litho steps or more advanced litho

@techfund1 Doubtful. Companies have reduced layers.

@ChipsSaas here you go sec.gov/Archives/edgar…

@techfund1 what is this from? do you have link?

$AMZN + Cerebras are working on a similar disaggregated inference solution as $NVDA + Groq Prefill => runs on Trainium 3 Decode => runs on Cerebras