Fitim Bozar

Here's a pattern I track: AI capex estimates keep getting revised up, not down. Goldman Sachs' own 2026 AI capex numbers came in 61% below what happened. When forecasters keep underestimating a trend, that tells you where the cycle is. Not financial advice — DYOR.

The data center semiconductor market was $86.8 billion in 2024. It's projected to hit $265.8 billion by 2029, a 25.1% compound annual growth rate. Most people are pricing this decade like the buildout is already priced in. I'm not so sure. Not financial advice — DYOR.

Chinese hedge funds that rode the AI rally are starting to look for exits. JPMorgan is telling clients to look past Big Tech into semiconductors, infrastructure, and utilities. When smart money disagrees this openly, I pay closest attention. Not financial advice — DYOR.

The Bank for International Settlements just warned the AI investment race could turn into a debt-fueled bust. I take that seriously. These warnings arrive at every stage of every buildout in history. What I track: capex and capacity utilization. Not financial advice — DYOR.

Microsoft is set to spend roughly $190 billion in capex this year, 61% higher than last year. Hyperscaler capex is tracking toward $650 billion in 2026, past $1.1 trillion in 2027. I track capex growth as the clearest signal in this sector. Not financial advice — DYOR.

Everyone is watching Nvidia. I'm watching the data center segment inside Marvell. It's now 76% of company revenue, up 27.6% year over year, with Q2 guidance implying 35% growth. The picks-and-shovels layer keeps compounding quietly. Not financial advice — DYOR.

Micron just posted revenue of $41.46 billion, up 345.7% year over year, beating estimates by 17.6%. Non-GAAP EPS hit $25.11 versus $20.28 expected. Seventh straight beat. The thesis I follow says memory is the next AI bottleneck. Not financial advice — DYOR.

Microsoft is rolling out Project Perception, a security tool that routes across Anthropic, OpenAI, and its own models. Cheaper than single-vendor stacks, with distribution most competitors can't match. Multi-model routing is becoming the default architecture, not the exception.

I keep seeing people debate Claude Code versus n8n like you have to pick one. Claude Code ships a full SDK for building custom agents. n8n now has memory, multi-agent coordination, and built-in guardrails. The winning setup uses both, not either.

Grok can now generate images from a voice message and edit video with two new tools: Extend and Upscale. X also started using AI to reorder what shows up in your Following feed. The distance between talking to your phone and shipping content is closing fast.

n8n stopped making you build workflows node by node. Its new AI Assistant reads a plain-language request, then builds, edits, and troubleshoots the workflow itself. I used to spend 40 minutes wiring a lead-gen flow. Last week it took under 5.

Mira Murati's lab just released Inkling. It's a large, open-weight model trained on multi-trillion tokens of web, code, and documents. Free weights, plus a hosted API for enterprises. Open-weight models are catching frontier labs faster than most people are tracking.

There is an open-source MCP server that connects Claude Code to 1,239 automation nodes and 2,646 pre-built configurations inside n8n. Type "build me a lead qualification workflow" and it creates, tests, and deploys it directly. No manual node-wiring required.

A new model is now dropping roughly every three days once you count open-source releases. GPT-5.6 Sol just set a new state-of-the-art on Terminal-Bench 2.1. Terra matches GPT-5.5 at half the cost. Luna is the fastest and cheapest of the three. The pace is the real story here.

Anthropic just shipped Claude Sonnet 5. It beats every prior Claude model on coding, debugging, and multi-step agent tasks. Intro pricing is $2 per million input tokens and $10 per million output, locked until August 31. I'm rebuilding client workflows around it this week.

Alphabet reports earnings next week. That call tells us more about the AI infrastructure cycle than every chip headline this month, because hyperscaler capex guidance is the real leading indicator. I'm watching one number: capex guidance for 2027. Not financial advice — DYOR.

The AI infrastructure trade everyone knows: Nvidia, TSMC, the obvious names. The trade I'm watching: power chipmakers like Texas Instruments and Vishay, who keep AI data centers from melting down. Nobody talks about power delivery until it fails. Not financial advice — DYOR.

The data center semiconductor market was $86.8 billion in 2024. By 2029, it's projected to hit $265.8 billion. A 25% annual growth rate for half a decade. Most of that growth won't come from compute chips. It comes from chips managing their power. Not financial advice — DYOR.

A number most people missed in Marvell's last earnings call: 76%. That's the share of revenue now coming from data center chips, up from a once-diversified chipmaker. Marvell didn't pivot to AI. It became an AI infrastructure company. Not financial advice — DYOR.

I read Micron's latest earnings line by line. Revenue: $41.46 billion, up 345.7% year over year. EPS beat estimates by nearly $5. Seventh straight beat. That is not a bubble stock. That is a company selling into demand it cannot fully meet. Not financial advice — DYOR.