Former senior executive at Microsoft Corp
- Cloud purchasing momentum for software, electronics components and semiconductors
- Positioning of key cloud data centre suppliers such as Intel (NASDAQ: INTC), AMD (NASDAQ: AMD) and Nvidia (NASDAQ: NVDA)
- Adoption of custom-designed semiconductors, AI inference accelerators and ARM (Advanced Reduced Instruction Set Computer Machines)-based server CPUs (central processing units)
- Supply chain update, disintermediation risks, vendor relationships and market share shifts
What are some themes and trends in component procurement for hyperscaler buyers? What should we focus on over the past couple of quarters since your last Interview [see Cloud Data Centre Semiconductors – Custom Design & Disaggregation – 10 December 2021] and as we look out into H2 2022?
What are your best-, base- and worst-case outlooks for server demand through H2 2022 and into 2023? What might happen if the US economy entered a recession? As you mentioned, it seems hyperscaler growth remains pretty robust, but we’re definitely seeing some consumer products and tier 2 vendors under significant pressure.
What do you think would cause such a big moderation in hyperscaler or server growth, as you outlined?
How would you expect hyperscaler CAPEX to evolve in H2 2022, primarily around tier 1 and cloud vendors?
Do you think shift spend is increasing in line with the overall spend on CAPEX?
How could deployment construction requirements evolve over the next 3-5 years?
How much of the consumption increases over the next 5-10 years could be handled through simple hardware replacement vs broad-based build-outs?
You said there seems to be caution from major hyperscalers as the world recovers from coronavirus. What do you think this means in terms of CAPEX spend in dollars? What does the continued emergence from coronavirus mean for the overall sector, including hyperscalers’ growth caution?
How might the custom design trend play out over the next 3-5 years, in terms of the proportion of workloads that will or could run within data centres on internally designed chips?
What might be the best- and worst-case scenarios for the custom design trend and its adoption over the next five years? You don’t seem to think this will run out of steam any time soon.
How should we think about the budget allocation spend on cloud networking equipment going forward?
What are your thoughts on the market’s annual growth or CAGR in the aggregate?
What is the spending strength at the tier 2 clouds you mentioned, primarily Baidu and Alibaba? What dynamics and trends should we monitor here?
To what extent are supply chain issues constraining hyperscalers’ abilities to build out at the pace they would like to? Where are we currently and how might this evolve over the next couple of quarters?
It seems the forecasting is no longer considered accurate enough. How might this impact more refresh-focused players over the next 12-18 months, given the strength of the buying?
Could you expand on where leading hyperscalers’ inventories are at today for CPU [central processing unit], GPU [graphics processing unit] and memory? Do you think inventories are likely to continue to grow or shrink QoQ through H2 2022?
What is your prediction for hyperscaler server purchasing volumes over the next couple of quarters, should prices reduce as expected?
How are hyperscalers pushing to optimise their data centres’ cost and performance?
To what extent is the software offering important to the cloud and enterprise data centre operators relative to the hardware offering? Could you provide context around CAPEX spending in dollars, so we can better understand the importance of priorities on both sides of the spend?
Would you expect self-development of software from major cloud and hyperscaler providers to continue with as much aggression as the hardware custom design trends? Is there a way to compare or contrast the two in terms of the operators’ priorities?
What do you think about the introduction of microservers and other innovations in hardware, such as SmartNIC [network interface controller] or additional accelerators for offloaded compute? How do these impact dynamics and spend in the space?
Do you think hyperscalers will have the same appetite for hardware innovation in their procurement over the next 6-12 months as we possibly enter a tighter economic environment? What might this mean for the market? Does it materially shift anything?
Could you outline some key trends in training inference and how they impact the positions of AMD, Intel and Nvidia?
What would make a vendor likely to win the AI inference market, or across GPUs, ASICs [application-specific integrated circuits] or CPUs? You mentioned competition stepping up.
Could you expand on the strength of Nvidia’s software moat? Could AI chip start-ups such as Graphcore or Cerebras realistically win deals over the next 1-3 years in the data centre training market?
To what extent do you think the hyperscalers are testing the AI chip start-ups today? What functionality would it take for hyperscalers to test and potentially continue to work with them?
You mentioned the specifications and the TCO [total cost of ownership] model. Does anything else within the models you mentioned factor into the vendor decision-making criteria? Might any factors become more important over the next two years? Should we monitor anything around adoption of AI chips coming from start-ups vs what’s available on the market?
How are you assessing the state of the merchant switch silicon market in the data centre now and going forwards? What are your general thoughts around merchant adoption vs networking hardware equipment?
Why are Google and AWS [Amazon Web Services] deciding to design their own custom silicon? Is it to solve a very specific niche problem or is it to keep main chip suppliers competitive?
Is anything else important to understand about when AWS decides whether to create its own chips or rely on merchant chips? Are custom chips workload-specific? What share of workloads are likely to move to custom silicon?
Can you quantify the lead time rate across key categories, whether they’re growing or shrinking? How might components such as microcontrollers, passives, DRAM and display ICs [integrated circuits] grow or shrink over the next couple of quarters?
What’s your outlook for NAND and DRAM pricing through the end of Q3 2022 into Q4? Might we continue to see upward or downward price movement?
Can you discuss the softening in the price curves for NAND and DRAM? You seem to be expecting a continued flattening in the price curve for both, and I think we’ve seen that with some of the past softening in the market based on Micron’s Q3 FY22 earnings report. What are your expectations for differences between NAND and DRAM? How do you see competitors and their roadmaps positioned over the next 3-6 months or so?
When do you expect decision-makers to start purchasing DDR5 [Double Data Rate 5]? Where are we on the adoption curve?
How are customers viewing Nvidia’s move into new areas beyond GPU, so DPU [data processing unit] and the new CPU offerings?
Could you discuss the developments in non-GPU accelerators? Do these threaten the position of GPUs in data centres, and which companies are mostly likely to benefit from these trends?
Could you discuss the pricing dynamics for AMD vs Intel CPUs? Are they equally priced today? What direction is pricing heading in?
What are your thoughts on the CHIPS [Creating Helpful Incentives to Produce Semiconductors] Act seemingly passing a Senate test vote? This seems to disproportionately benefit Intel over the fabless players, particularly Nvidia and AMD. How would the bill passing impact hyperscaler CAPEX and procurement over the next couple of quarters, if at all?
Should we highlight anything else on data centre and hyperscaler chip and software procurement? Do you have any other predictions about the state of procurement over the next couple of quarters?
Gain access to Premium Content
Submit your details to access up to 5 Forum Transcripts or to request a complimentary one week trial.
The information, material and content contained in this transcript (“Content”) is for information purposes only and does not constitute advice of any type or a trade recommendation and should not form the basis of any investment decision.This transcript has been edited by Third Bridge for ease of reading. Third Bridge Group Limited and its affiliates (together “Third Bridge”) make no representation and accept no liability for the Contentor for any errors, omissions or inaccuracies in respect of it. The views of the specialist expressed in the Content are those of the specialist and they are not endorsed by, nor do they represent the opinion of, Third Bridge. Third Bridge reserves all copyright, intellectual and other property rights in the Content. Any modification, reformatting, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, transferring or selling any Content is strictly prohibited