Tech Thoughts Newsletter – 3 March 2023.

Market: European inflation reads were unexpectedly hot this week, but markets were generally a little calmer, and arguably more fundamentally driven. A slightly quieter week for results, too. 

Portfolio: We added to some of our long-term conviction positions this week: TSMC, Microsoft and LAM Research. 

Another week and more AI news. Despite the initial noise around Bing and Google, it seems increasingly clear to us that search interfaces might not be the obvious place for chatbots to exist. 

Our conclusion on the week of the “Google vs Bing” sparring match was that the biggest problem of implementing chatbots into search was the business model: Each search becomes much more costly to run through inference, but search monetisation is based on the link the user chooses to click – not on the searches a user runs – it’s a fundamental mismatch and it isn’t clear how either Bing or Google could create a business model that works. 

However, as more has emerged around large language models – including their “hallucinations” – where AI really does just make stuff up – it feels like the more important barrier to using chatbots for search is more simply that it isn’t that compelling from a user perspective. If we want to know about some specific new technology, we’d much rather not risk ChatGPT hallucinating and giving us the wrong information, and instead continue to use Google (which is really still better than Bing) to get to industry product pages/scientific articles.

But, just because we might be pouring cold water on the search use case, doesn’t mean that we don’t believe AI will have compelling applications elsewhere…

This week, Snapchat announced it was introducing a chatbot powered by OpenAI’s ChatGPT, called “My AI”, which will be pinned alongside conversations with friends. Initially it’s only available to Snapchat Plus subscribers ($3.99 a month) but the idea is that it will be rolled out more broadly: “The big idea is that in addition to talking to our friends and family every day, we’re going to talk to AI every day.”

What Snap is doing feels more of a natural home for ChatGPT than search – a chatbot alongside a messaging platform, effectively as an entertainment use case. And importantly from a business model perspective, subscription feels much more of an obvious way to monetise it vs ads and search. (There’s a question as to whether AI will more broadly move business models to subscription vs advertising given the cost structure – Meta last week introduced subscriptions, albeit on a much smaller scale – and there are also reports this week that it will bring in an AI powered chat into WhatsApp/Messenger products). 

When we look at the AI value chain, our exposure is predominantly through the chip/hardware/infrastructure side, given the competitive moat players like TSMC, ASML, and Nvidia have built through scale, tech, relationships with customers, and where we see business model and returns certainty vs new end-user/entertainment applications.

More broadly in the results and newsflow we covered around AI this week, there were several discussion points we had as a team:

• The Microsoft AI enterprise opportunity (which strikes us as a much bigger opportunity than the Bing resurrection..) – it is already charging for AI services like meeting summaries in Teams
• Can Meta use AI to mitigate ATT? (which we talk to below)
• Can software businesses like Workday, Salesforce (both reporting this week, both spoke on AI as a potential product enhancement) integrate AI into their products and get users to pay for it?
• Can it make businesses more productive? Can workloads be better optimised via AI? 

Onto results: 

Platform software plays showing proper leverage and a move towards sustainably profitable growth – and benefitting from spend consolidation 

• Salesforce (owned) reported a BIG beat on the margin, and solid top line growth
• Clearly Salesforce has been under a lot of activist scrutiny around its high cost-structure and it’s a business where on our metrics, sales productivity is much lower than peers. So it’s no real surprise they’re able to move margins up, just the magnitude and speed is much quicker than we might have expected. 
• Q4 revs +17% yr/yr cc (5% ahead of expectations) and 29.2% margin. Guiding to 10% yr/yr for FY24 (while consensus was expecting mid to high single digit). 
• But the standout number was really the guided op margin of 27% FY24 (CY23) and 30%+ in Q125 (Calendar Q1 24).
• Other details – they’re “disbanding” the board’s M&A committee – which removes the risk of big deals (I think we’d argue that Salesforce’s track record of M&A success is not stellar – although Tableau and Mulesoft execution looks to be improving..). 
• They’re also upping the buyback to $20bn to fully offset stock based compensation – that’s another company (after Palo Alto last week) to call out a conscious effort to bring SBC lower – for so many years that has just been an ignored (but very real) cost component with all the investor focus on Non-GAAP EPS – how times have changed!
• Workday (owned) also reported a solid set of results – beating and raising guidance. Q4 came in a bit better than the guide and expectations – subs revenue was guided to come down to 19% growth in Q4 from 22% in Q3 – and actually it stayed at that 22% yr/yr clip. And its margin was ahead too. 
• The FY24 guide has been narrowed a bit, around the same mid-point (18% yr/yr subs growth) but margin is guided to 23% (so firmly on track for the 25% mid term), which – adjusted for a depreciation change – is about 50bps ahead of expectations. 
• On the call they were very clear that they should get back to above 20% subs revenue growth when we get back to a “more normal” macro environment, alongside margin expansion. 
• There is some acknowledgement (the same for Salesforce) that some deal cycles are longer with more scrutiny but the numbers speak to the idea that spend is consolidating around them. “Companies are doubling down on becoming more efficient through the downturn and we’re a beneficiary of that”.
• In the details there was also a new go-to-market partnership announced with AWS – we think these alternative sales channel routes (like Palo Alto with a renewed emphasis on systems integrators last week) will be a key efficiency driver for many of these businesses. 

Portfolio view: 2023 has been termed the year of efficiency for tech but especially for software, where high sales spend has been a feature of the last 5-10 years’ free money environment. Workday and Salesforce are two businesses with very strong product portfolios and positions in their industries which should enable them to drive margins higher without a material impact on the top line growth. These SAAS businesses that can finally start to show leverage and compound returns is exactly where we want to be positioned in the software space. They should also benefit to the extent that there will likely be lower “free money” competition (removing the M&A mandate from Salesforce speaks to that). 

In a tighter spend environment we also see both Workday and Salesforce as being very well positioned to benefit from software spend consolidation 

Data/consumption plays – bill shock vulnerability and a much less proven business model

• Snowflake guided to a disappointing 2024 product revenue growth rate of +40%, marking a significant slowdown from the 70% 2023 number. 
• They noted some “bookings reticence” and caution around larger, long term contract expansions. It looks to us like some customers may have had some bill shock which is now being felt through substantially lower bookings (and smaller deals) to bridge customers to the end of their contract length (so we assume customers that have used their consumption allowance, instead of signing up to a new large contract, “buy as they consume” until the end date of their contract..). 
• New customers are also ramping their consumption much lower
• Staying on the theme of cost efficiency, the CFO stated that it is not a “growth at all costs” company – we’d note that sales and marketing costs as % of revenue were an eye-watering 54% last year… and we’re in wait and see mode as to whether this is a business which like (like Workday and Salesforce are doing) can see real leverage in their business model.  
• Worth noting that they, like Workday, have entered into a joint go-to-market strategy with AWS – again it seems to us increasingly clear that some of the high internal sales cost is being viewed by businesses as unproductive.
• On AI: “One of the challenges with these new technologies that people come up with a lot of interesting questions, but without a solid business model that’s not going to take off…”

Portfolio View: Consumption business models still haven’t really been through a cycle. It now looks like the boom 2020/2021 period was really overspend and we’re now in period of consolidation that’s even more pronounced than the more “traditional” SaaS-type models. We continue to be cautious. 

Semicap still solid demand, innovation continuing to drive node on node manufacturing efficiency

• In semicap equipment, ASMI (which had positively profit warned earlier in the year) came out with its full year results – ahead of the guide and reporting solid orders. That’s good news for our holdings in the sector because it means forward looking order metrics should look good across the board (while there are niches in semicap and memory players are likely to be a bit more vulnerable, it’s rare to see a huge divergence between players)
• Q1 is guided to EUR660-700m which would be a 6% qtr/qtr decline – in line with AMAT seasonality (and about what we should expect given exposures). Q2 guidance is for slightly up qtr/qtr which is good news,  and then on WFE spend the mid to high teens forecast is broadly better than the 20% decline most of consensus is baking in.
• All in all good news for our semi cap exposure – overall capex looks to be holding up better than the market expects and visibility/order books still strong (ASMI commented on the call that orders so far this year have still been healthy).
• Like many of the semicap businesses, we think they were slightly premature in their assessment of the China restrictions – a good deal of orders that they cancelled they then rebooked, and even outside of this they stated that their Chinese customer orders were strong, all related to mature nodes, power/analog and wafer manufacturers – again that speaks to our view that China is still spending and will still have a strategic position in semis. 
• It’s been the SPIE lithography conference this week which we’ve been listening to for semicap equipment names. It’s a conference which covers all the latest developments in lithography and the emerging technologies around it. 
• The big news that moved ASML and AMAT was AMAT’s new “Cenutra Sculpta” patterning tool. Multiple-patterning is the process of splitting a chip pattern into two (or more) simpler masks, and then combined on patterning films to be etched onto the wafer as a much finer pattern. 
• The initial read from the market was that this was good for AMAT and bad for ASML (they virtually gained and lost the same market cap on the day in share price moves). That’s because the tool could be used to create the layers that are currently being made with EUV double patterning (so effectively two EUV passes). 
• The reality is that this simplification around patterning (and reduction in the number of “complex” layers) has been a feature of the industry forever – it enabled EUV to be utilised at a lower number of layers back when it was first being introduced into production; and the opportunity for further node shrink is ultimately a positive for the whole industry – remember the bear case for the industry is that Moore’s law stops – any incremental tech getting us from node to node in a more cost effective way is a good thing. 
• It’s also worth remembering that it was part of Intel’s reuse strategy that ultimately lost them a lot of learnings in moving smaller architecture over to EUV and has really left them very far behind – that’s a lesson the whole industry remembers.

Portfolio view: we still like our semicap exposure – we increased our position in LAM this week as we increasingly think DRAM should see a stabilisation of spend, and that ASMI comments on tech upgrades in DRAM around high-k/metal gate technology are supportive.

M&A reinforcing Infineon’s competitive advantages in the power semis space 

• Infineon (owned) has made an all cash bid for Canada based GaN for $830m. GaN is a provider of Gallium nitride-based products and solutions for power conversion and has more than 200 employees, and 2000 customers including some large names like Siemens, BMW, Dell and Samsung.  
• Gallium Nitride is widely used instead of silicon in manufacturing of semiconductors and chips and offers cost benefit advantages vs silicon/silicon carbide for various use cases. The key benefit of gallium nitride lies in a superior switching performance, resulting in higher efficiency and lower system cost. Some applications like chargers and adapters will over time migrate directly from silicon to gallium nitride.
• At its Q4 results, Infineon stated that its GaN business is small but by far the highest growth rate. 
• Staying on EV, at its Investor Day, Tesla announced a reduction in drive unit cost, part of which will come from a 75% reduction in SiC. It’s surprising because it’s really the opposite direction to the rest of the industry – which is moving to add more SiC to increase range of EVs. Infineon has a very limited exposure to Tesla, but is benefitting from the widespread OEM adoption of SiC. The technology evolution is something we’ll continue to watch, but we think it’s unlikely the broader industry will move away from Silicon Carbide given its proven performance benefits (and the fact that it is continuing to become more cost effective for OEMs to implement). 

Portfolio view: Gallium Nitride is becoming a key material for power semiconductors, alongside silicon and silicon-carbide. This acquisition strengthens Infineon’s position in future core technology which we view as a positive. 

Network infrastructure strong alongside continued cloud build outs and new AI networks

• Broadcom (owned, small position) reported a stronger than expected EPS and sales inline with expectations, up 16% y/y.  Q1 Semiconductor Solutions sales increased 21% y/y. 
• Semiconductor sales growth is being helped by hyperscalers building out switching capacity and adoption of AI solutions. 
• Networks supporting large language models need to be lossless, low latency and able to scale – we think that means fuller, more frequent network infrastructure upgrades which the whole ecosystem (we spoke about Cisco and Arista Networks a couple of weeks ago) is benefitting from. 
• Broadcom semiconductors is fully booked for fiscal 2023 and lead times and visibility on semiconductors remain largely at 50 weeks. While there have been a small number of requests to push out certain orders, these are the exceptions and they have not had a material impact on their business.
• In 2022, Ethernet switch shipments deployed in AI was over $200 million (c. 10% of networking revenue). They forecast that given the demand they’re seeing from hyperscalers, this could grow to well over $800 million in 2023 – driving a significant part of the overall growth. 

Portfolio view: Broadcom continues to deliver on top and bottom line. The share price performance (and the reason we hold a relatively small position) is held back by the risk of losing Apple WIFI chip and the VMWare acquisition that’s under competition authorities scrutiny. More broadly we like Network infrastructure as a way to play AI and increasing bandwidth requirements and workload complexity. 

Other news.. 

Another Q1, another COVID hangover 

Zoom was “better than feared” but with a top line growth guide of +1%, the business is more like a telco than a tech company, on a much less attractive multiple.. We suspect that Q1 will bring more COVID enterprise deal churn (with each Q1 we lap the COVID signing quarter.. So for those businesses with annual contracts it’s the quarter with the biggest potential loss – that’s something to think about at Q1 reporting. We think the COVID hangover risk in the stocks we own is very minimal)

Chips Act coming with some more rules and guardrails for recipients 

On Tuesday, the commerce department issued their Notice of Funding for the $39bn semiconductor manufacturing funding program and set out new rules around businesses that accept CHIPS grants. We have to say some of the rules surprised us – companies must agree to not expand in any “foreign country of concern” (China – though it isn’t named explicitly); as well as commitments to limiting buybacks/dividends and an “upside sharing” agreement to share excess cash flows with the US government.. 

We’ll see how these get implemented in reality – but it does strike us as a neat way of getting foreign entities (like TSMC, Samsung) to further hamstring China. Before these guardrails were in place, it was up to their own governments to impose restrictions. Now, if TSMC (for example) wants to accept the funding, it looks like it must agree not to expand its Nanjing fab..? Let’s see..

Relatedly, Japan’s foundry venture (consortium) Rapidus announced it would built a plant on Japan’s northern island of Hokkaido, aiming to mass produce 2nm chips in the late 2020s with a test run planned in 2025. Small but it’s not just the US government incentives driving the industry from a geopolitical perspective. 

Meta mitigating ATT with AI?

Meta’s “Advantage+” advertising product it launched in August which uses AI to generate different advertisements (different text/image/products according to the specific objectives of the marketer) is said to be showing very good results. 

It’s one of the ways Meta hopes to mitigate the impact from Apple’s ATT (App tracking transparency) which effectively stops the use of third party cookies traditionally used to track users across the web. ATT has been one of the biggest structural issues we think over the past 18 months (it was introduced in Summer 2021). The capex spend on servers/AI is a separate bucket of spend to the metaverse R&D costs, and we think much more justified in terms of trying to retain Meta’s competitive moat. 

It’s a good example of AI having an impact beyond the ChatGPT hype and being used on the operational side of businesses. The issue for Meta (which we don’t own) as it relates to returns is that, even if they’re able to catch up with the old targeting success metrics,  this is likely a structural increase in cost/capex (server costs), because in order to stay ahead, there is a constant refresh cycle necessary – thinking about returns, a 5pp structurally higher capex/sales ratio brings asset turns down and leaves you with a structurally lower returns (which should feed through to valuation). 

For weekly insights on the latest market updates, please subscribe to our Tech Thoughts podcast.
For more information about the latest trends and forecasts, please visit our official Tech Thoughts page.

We provide investors with access to category leading technology companies, globally. Our assets under management have a total value of more than €1bn, and our limited partners include institutions, family offices and entrepreneurs. Learn more about our funds here.

Enquiries
For enquiries, please contact:
Inge Heydorn, Partner, at inge.heydorn@gpbullhound.com
Jenny Hardy, Portfolio Manager, at jenny.hardy@gpbullhound.com
Nejla-Selma Salkovic, Analyst, at nejla-selma.salkovic@gpbullhound.com

About GP Bullhound
GP Bullhound is a leading technology advisory and investment firm, providing transaction advice and capital to the world’s best entrepreneurs and founders. Founded in 1999 in London and Menlo Park, the firm today has 13 offices spanning Europe, the US and Asia. For more information, please visit www.gpbullhound.com.