Kids' Texts: Excitement Quickly Turns to Worry
When your kids leave the house you get excited when they text you. But more often then not its... https://t.co/2fuBivB8YP
Cycles Return Even With Higher Dollar Floor
Memory will be cyclical again someday, even if the dollar floor is greater than past cycles.
ChatGPT Brand Kit Supercharges Visual Workflow Speed
Until you have tried it, having a codex skill that has your brand kit/style kit which ChatGPT made for you, for codex, is a top shelf workflow automation for visuals/decks/charts, etc. Time to visuals is now magnitudes faster.
Nostalgic Intel Fab Tour Highlights Bright Future Ahead
Bunny suit guy makes his return to my background. Got this at my first fab tour with Intel 04/05 ish I think. Pentium days were good days but even better days ahead for Intel. https://t.co/vvz4bA8THf
Semi Industry Facing Worse‑than‑expected Giga Cycle
From the report that kicked of @DiligenceStack in January, and why I call this moment a semis giga cycle. We knew it then, its even worse than we thought from what we know now. https://t.co/ltMhgyPJtV https://t.co/cAkAlM6aPY
NVIDIA‑Intel Feud Ends as Jensen’s Rival Departs
The bad blood between NVIDIA (Jensen) and Intel is over. The Intel he genuinely didn’t like is gone. Good times ahead.
Misinformation Accelerates Capital Fear in Semiconductor Market
One thing I keep seeing and do worry about is the amount of bad/wrong information that keeps coming out on the semiconductor industry. When the market is doing what it’s doing, capital spooks easy and the information that causes said...
Enterprise Software Spend Hits $6‑7K Per Employee, Likely Rising
So today the average enterprise pays $6-7,000 per year per employee on software and services. Once token budgets get sorted, which tokens will be used in that figure, I do think this annual spend per employee goes up. Question...
Never Let Agents Idle—Task Them Anytime
As my saying goes. Never let your agents sit idle. It’s a waste of your time and theirs. Sadly, sometimes, I wake up in the middle of the night with a thought or idea and need to task the agents...
Speed Wins: Time-to-Compute Becomes Currency
We’ve now mapped more than 25 constraints across the AI compute and power supply chain. Across conversations with operators in nearly every pocket of the industry, the message is consistent: speed is what matters. This may be less about the best...
Gemini's Recommendations Need Built‑in Purchase Links
Gemini really needs click to buy links. Gave Gemini a picture of my face asked for recs for skin care routine. Gave excellent recommendations but had to go search each on Amazon. Would absolutely have just clicked to buy...
AI Automation Reshapes Teachers' Workflow, Saving Planning Time
My wife is a teacher, lots of dense workflows mostly web service based. Tons of automation we are discovering with codex + chrome to help save her time from lesson planning, to IEP building, etc. Really amazing to...
AI Shines in Dense Workflows, Boosting Speed
I did not plan it this way yesterday, obviously, but Codex with Chrome only further validates what we laid out in this series where AI right now captures value where workflows are more dense, or where automation leads to faster...
CPU TAM Has Expanded Significantly Since the Recent Report
As I laid out months ago. And CPU TAM is now meaningfully larger via our models and others, TAM wise, then I when wrote this report two months ago. https://t.co/dKqm16myKH https://t.co/s0K2vC0J8C
Intel Emerges as Vital Second‑source Foundry Amid Capacity Crunch
Intel is the only foundry, and a now officially viable foundry second source, who has the capacity to expand to meet this DEEP foundry capacity shortage we have in the semiconductor industry. I'd expect any deal with Apple to...
New AI Features Spark Excitement and Exhaustion
We are at that stage where every new big feature release from @OpenAI and @AnthropicAI opens up new levels of unlock you just want to push both to the limits and try everything you can think of. This both simultaneously...
Codex in Chrome Feels Like Pure Magic
The only way to describe codex working in chrome is pure magic. "Any sufficiently advanced technology is indistinguishable from magic." - Arthur C. Clarke's Third Law
AAOI: Key Play in Copper‑to‑Fiber Transition
What I said to watch on $AAOI in my two part series on from copper to fiber. https://t.co/pCat48lvKI https://t.co/VkgvLSCH4T
Justifying Multiples Requires Consensus, Margin Growth, and Sustained Revenue Ramps
I know multiple justification is out the window these days but for these names. $COHR can justify the multiple mostly by hitting consensus and expanding margins. $LITE needs either a higher accepted scarcity multiple or material upside to 2028 revenue....
Shocker SiP Currently Volatile, Will Stabilize Within a Year
Shocker SiP is lumpy... In a year or so, much less, but for now, ebbs and flows.
Enterprise AI ROI Emerging—Who Captures Production Value?
Second report in our series on AI monetization: if ROI is starting to show up in enterprise workflows, who captures the value as those workloads move into production. Next week, we close the series with CIO/CTO survey results on AI...
Inference Margins Outpace Training, Driving ROIC Above 80%
As we have continually pointed out, inference margins are vastly higher than training margins. And as compute infra optimizes even more for inference workloads north of 80% will be the standard (often higher). This is the inflection for ROIC our...
Memory & Architecture Design Crucial for Inference Era
Good discussion here with @MicronTech Chief Business Officer Sumit Sadana, @AMD CTO Mark Papermaster on memory and architecture design needs in the inference era. https://t.co/H5d1zt1KbM
Data Center CPU Market May Exceed $200B by 2030
I'm now beginning to wonder if the now consensus data center CPU TAM ranging between ~110-175B in 2030 could be off and it ends up being north of $200b. Just looking at our model, unit volume trends, what we hear...
Timing Gap Hits AI Optics Stocks, COHR Slides 7%
$COHR off ~7% after a solid Q. Similar to $LITE, the market seems to be digesting the timing gap between real AI optics demand today and the larger architecture transition still ahead. As with everything, timing matters.
Compute Is Money: Race to Scale Inference Power
Hopefully, it is abundantly clear that Jensen was right when he said compute = $. The race is on to acquire and bring online as much compute as possible as fast as possible. Inference era economics $$$ 🚀
AI Shines Most in Dense Workflow Environments, CIO Survey Shows
Thought this was an interesting way to highlight our conclusion that AI's best value capture at the moment is in areas where workflows are dense. This will show up when we detail the CIO survey data. https://t.co/GO5Exh5d9w https://t.co/tV1dADV1Ga
Most AI Deployment Limits Lie Beyond the Chip
This is going to be an interesting report. Quick highlight: Our map identifies 25 constraints between announced AI capacity and billable compute. Ten can gate deployment outright, and most of those sit outside the chip itself.
AMD Poised to Dominate Expanding Data Center CPU Market
Shared some thoughts on $AMD along with how we see the data center CPU TAM expansion playing out.
Supply Chain Masters Reveal Upside in Constraint Map
I'm working on a report on the constraint map (I know it's an overused term) but still apt. But in this environment, the easiest way to see the upside is by those who are masters of the supply chain. https://t.co/9ZCdGgQZVs
CPU Market Set to Quadruple by 2030
We are tracking this as well going forward. This is the view of the pre-agentic CPU era, and when the CPU TAM is ~$25B. This may look very different as the TAM goes north of $100B in 2030....
Corning to Supply Passive Photonics for Inside‑Box SiPh
$GLW moving inside the box. I maintain that Corning becomes a key "co-design" partner for optical as SiPH really starts to ramp in 2028. But those conversations are happening now. "historically, we've had no inside the box content...
Corning's Optics Will Dominate GPU Rack Interconnects
$GLW having a presentation on optics. This slide.. Everywhere you see yellow you see Corning. "And when we add the optical scale up network to interconnect all the switches in every GPU rack, well we add a lot...
AI Survey Shows CIOs Rely on Claude for Slides
Working through survey results for our report next week on CIO/CTOs and AI investment and deployment strategy. Claude still makes prettier PPT FYI. https://t.co/HdvQUk4jew
Wafer Shortage Forces Companies to Fight for Compute Capacity
What is clear in semis right now is that we are in a wafer and capacity scarcity moment. A key part of the analysis is which companies are best positioned to secure the capacity they need to meet unprecedented compute...
AMD Forecasts 35% YoY Rise in Server CPU Market
$AMD @LisaSu conviction that CPU server TAM 35% YoY driven by agentic. This is actually going to be more fun to track than GPU/ASIC volumes YoY :)
Optical Hype Outpaces Market
There has been too much talking of optical technology, but not mapping to market timing or vendor deployment analysis. This is generally where supply chain checks are useful.
Neoclouds' CPU Shortfall Threatens Agentic Inference Competitiveness
What gets missed in the neocloud narrative is the CPU side of the equation. If neoclouds intend to compete with hyperscalers on inference workloads at scale, and particularly on agentic inference, they will need substantially more CPU capacity than their...
Agentic CPU Market Forecast May Become Obsolete in Six Months
The agentic CPU, as a specific product, will evolve over this year. It's possible much of what we know now will be out of date in six months. That said... This was a stab at detailing the market...
Agentic AI Enters First Investable Cycle via Measurable Work
Latest report @DiligenceStack From AI Usage to AI Earnings Power Agentic AI’s first investable cycle is forming around measurable work, workflow control, and budget formation. https://t.co/GO5Exh5d9w
UBS Predicts TAM Expansion, Shared Gains for Intel, AMD, ARM
Now this is the UBS chart that will be HOT debated. $INTC, $AMD, $ARM. Remember, this is TAM expansion, so dollars for everyone. This is share grab not share taking. For now... https://t.co/KfyXsEftNa
Apple's Foundry Talks With Intel Are More Advanced Than Expected
Agree with this, we think Apple is much farther along than just "discussions" with Intel on foundry.
IREN's Mirantis Acquisition Could Lift Its Benchmark Rating
Definitely have to see the $IREN acquisition of Mirantis as a way to move them up the @SemiAnalysis_ benchmark list. Time will tell, but I'm sure this will move them from not-recommended. WIth all neoclouds' ability to leverage software for...
Microsoft’s Datacenter Rollout Far From Canceled, Many Projects Underway
Microsoft has clearly paused or canceled some projects and leases, but “probably didn’t get built” is not right. Fairwater broke ground in Wisconsin in 2023 and Microsoft says it came online ahead of schedule. Saudi Arabia construction was completed in...
AI Revenue Could Surge Past $300B This Year
Best estimates I've seen for dedicated revs tied to AI was around ~$100b for 2025. That includes hyperscalers, neoclouds, labs, etc. It is likely that the number will be above $300B this year as this number inevitably catches up...
Power Semiconductors and Supply Chain Layers Gain Spotlight
I heard power semis, and layers deeper in the power supply chain, are starting to get more attention. I should have known given this report drove the second highest paid subs rate after my neocloud one. https://t.co/Ispz18CUmu
Custom Skills Power Codex Automation and Agentic Growth
Skills are where it’s at on codex. Been building so many custom skills and automations for my workflows and research process. Next unlock for agentic.
Inference Shift Set to Drive 60% Gross Margin by 2026
As we have been saying, inference will generate much better margins. In our Anthropic note, we project a 60ish % blended GM average for 2026, and the only barrier to full upside in blended is the shift to inference...
All Viable Accelerators Could Win Hyperscaler Adoption.
I maintain my thesis that all viable compute (in this context accelerators) has a chance of being adopted by hyperscalers and frontier labs. Hence why @cerebras is getting more attention and now @tenstorrent. BUT, @AMD has more upside, and maybe...
Anthropic Set to Outpace 2027 Revenue Forecasts by 2026
Yep, I laid out the whole revenue growth thesis in this report. Even two months ago they were the fastest scaling enteprise software company ever. https://t.co/kHzGuJf03g And, even our bullish revenue forecasts were low as Anthropic will do in 2026 what...