While the FY25 Q3 results constituted a beat, they came in quite close to what the share price implied. However, the guidance for Q4 surprised on the upside. Pelham Smithers reviews the performance and outlook, with a readacross to beneficiaries in Japan and the numbers from China.
As a follow up to our cheat sheet summarizing the cost structure of a 1GW datacenter, we estimate today on a single slide WFE required to produce semiconductors for a 1 GW datacenter, with our usual important caveat: WFE spending is driven by the growth of production capacity. In other words, one spends once on WFE to build a 1GW datacenter many times over.
We kicked off the 5th annual BCG × NSR Gen-AI Big Idea Series this month. We will host a slate of exceptional speakers through Feb-26 (full line-up on website). Our first session, a conversation with Mark Wade, CEO of Ayar Labs, went deep on optical I/O: progress in bringing optical I/O to market, the roadblocks to scaling co-packaged optics, and how optical interconnects could redefine the architecture and economics of datacentres. Please see the link below for our takes.
OpenAI announced a series of major contracts and partnerships in recent weeks, suggesting ambitions to deploy anywhere between 10 and 30GW of compute, with rich information about potential vendor positioning. We investigate in this research what can be inferred from these contracts and partnerships, and draw implications for the main names we cover, involved in the supply chain: Nvidia, Broadcom, AMD, TSMC, Micron, and semicap manufacturers.
Last week we attended the 2025 edition of the Semicon West conference, the key event for investors focused on the semiconductor equipment market. Over two days we hosted more than a dozen meetings with companies across the semiconductor value chain. Here are our key takeaways.
Memory fundamentals are strong, with HBM demand rising alongside a recovery in conventional memory, while supply stays tight. We expect a series of earnings beats over the coming quarters. However, Micron trades at peak multiples, and we expect those beats to be offset by continued multiple compression, consistent with past cycles—likely resulting in muted stock performance. We therefore downgrade Micron to Neutral, with a $190 target. For now, we prefer TSMC. It offers AI exposure comparable ...
The outlook for AI spending has strengthened in recent months, driving a rally in Semicap stocks. Our forecast embeds AI capex tripling by 2030, requiring ~$130bn of cumulative WFE spending, but driven by the first-order derivative of AI deployments, i.e. peak acceleration this year and peak spending next year; a trajectory in expectations already and resulting in a weak outlook beyond 2026. With near-term uncertainty, limited upside to 2026 forecasts, and valuations 2–12 turns above historic ...
Total capex required to meet OpenAI’s needs could reach ~$130bn by 2027, roughly equal to what consensus forecasts, on average, for each of the top four hyperscalers over the same horizon. In this note, we estimate how much of Nvidia’s datacenter revenue expectations this represents.
We publish today our comprehensive quarterly bible: 229 pages of detailed analyses on what happened in the last 3 months, and how we interpret it, in light of our current convictions. The first section acts as a PM summary, outlining our key findings, and latest thoughts on the semi cycle, in 6 slides. Please follow the link below for more details.
Today, we are publishing the Foundry section of our 28th Tech Infrastructure Quarterly Bible. The Tech Bible is a must-read for any tech investor, as it summarizes the quarterly earnings reports from the over 140 companies we track, providing an update on our key perspectives and convictions. In the coming weeks we will publish the last section on Enterprise IT. Driven by strong AI demand, TSMC increased its 2025 growth outlook to 30% (from 25%). This still leaves room for upside, as it implies...
Broadcom disclosed last week a $10bn AI rack order for 2H26 (widely linked to OpenAI) and signaled growth in their AI division could accelerate further in 2027. In this note, we break down what this could mean for the ASIC vs. GPU battle and the future structure of the AI compute market, looking in particular at the addressable market of in-house ASICs.
Unfortunately, this report is not available for the investor type or country you selected.
Report is subscription only.
Thank you, your report is ready.
Thank you, your report is ready.