Today's post:.1. AI on Fire with Cadence, Microsoft & Meta.2. An Under-the-Radar Play in the Biotech Value Chain.

$AMD began MI350 production ahead of schedule in June. Oracle is building a 27k node cluster with MI355X accelerators

$IFX - "Semi inventories at automotive OEMs could drop to a subcritical level". This obviously brings near term risks for mature semis, but also strong upswing potential once the cycle turns.

$LRCX - SAM per wafer in metal deposition going up 3x with GAA (TSMC N2), this is due to the slower and more complex atomic-layer-deposition of molybdenum (vs tungsten)

Nadella - " $SNOW accelerated this quarter".However, it's also clear that Databricks and MSFT Fabric are outperforming SNOW in terms of revenue growth

Samsung looking at inserting hybrid copper bonding into HBM volume production

"Plans for Moore's Law to continue until 2042" - $CDNS CEO

Agentic AI will design semis - from $CDNS:

$INTC remains a very difficult story:. - the biggest positive is that the company 'only' burned $1.1 billion of cash this quarter, the best result in almost two years. - the rest of the story is very muddy:. 1. Plans to go into AI with a more full stack approach - good luck.

Is the robotaxi market finally about to take off?. 1. Tesla will 10x the parameter count of its self-driving AI model. FSD is already extremely impressive, so this could bring the model actually close to being flawless. Especially with the massive data set Tesla has, which is a

SK Hynix taking up capex for this year due to stronger than expected HBM demand in '26

SAP is seeing 20 to 50% productivity gains with AI in coding, digital sales engagement and HR. SAP has always struggled with margin expansion as costs grew in line with revenues, but with AI productivity there is a clear bull case on margins. These are extremely sticky software

Goldman modeling on Chinese Semicap name AMEC - heavy growth in Etch

UBS on OpenAI potentially moving away from $DDOG:. We understand that OpenAI has at least two observability platforms, including a custom-built one for model monitoring and Datadog (and others) for monitoring how OpenAI’s products are interfacing with its GPU infrastructure.

How Goldman is currently modelling the HBM market:

Semis on Fire - TSMC, Broadcom & A New Niche Semi Winner From AI.

Netflix has started using AI to create special effects - 10x faster and massively lower cost. The AI generated a collapsing building in Buenos Aires, which otherwise would not have been possible with a show of this budget.

Account manager at neo-cloud Lambda: "No slowdown in Nvidia demand. Hyperscalers are trying to deploy as much Blackwell as they can. More available supply for H100, so pricing is coming down." (call this week via Tegus)

2) Despite these peak capex levels, TSMC’s capex-to-revenues ratio remains modest, suggesting plenty of upside potential: