0xrelogic

Analyzed Q3 2025 earnings data from UBS and found China's AI story is actually two separate cycles with 12-18 month lag between them.

• Mainland hardware suppliers (optical transceivers, servers): +34-57% earnings growth NOW

• Innolight hit $3.3B revenue (+122% YoY) from datacenter buildout

• Hong Kong internet platforms (Alibaba Cloud, Tencent): still in capex phase, monetization projected 2H26

• Alibaba e-commerce AI already break-even with +12% ROAS

The gap between these cycles matters for positioning. Hardware is paying off today, cloud is a 2026 call option.

Happy to discuss the optical transceiver data or enterprise adoption timelines.

I wrote this after Jensen Huang's Nov 7 statement that Nvidia now has "zero share" of China's AI datacenter market, down from 95%. The geopolitical split is creating three distinct compute ecosystems. - China revenue impact: Dropped from 21.4% of Nvidia's total (FY2023) to 13.1% (FY2025), with ~$15B in annual datacenter revenue now offline

- Huawei Ascend 910C performance: Delivers roughly 60% of H100 throughput for inference workloads, but training large-scale models remains challenging. The bigger gap is software CUDA's 15-year ecosystem vs Huawei's CANN framework

- Google TPU positioning: While Nvidia Blackwell has 12-18 month lead times and China is locked out, Google's TPU becomes more attractive for specific workloads (Transformer training, inference at scale) especially for GCP customers

- China's 30% mandate: State-funded datacenter projects under 30% completion must remove all foreign AI chips or lose funding (Reuters reporting from this week)

The article traces the export control timeline from Oct 2022 initial rules through Oct 2023 tightening, with all revenue figures verified against Nvidia's public financials and multiple sources.

Three parallel tracks are forming

1. Nvidia Blackwell (premium, supply-constrained, no China)

2. Huawei Ascend (domestic alternative, 60% H100 performance, ecosystem gap)

3. Google TPU (available hedge for workload-specific optimization)

Happy to discuss the geopolitics, technical performance comparisons, or supply chain implications. Also interested if anyone has hands-on experience with Ascend 910C vs H100 workloads.

Sources include Reuters, BIS.gov export control documents, Nvidia's FY2025 financials, and verified tech specs from multiple outlets.

Last week I wrote about Meta's $75B AI infrastructure spending. Seven days later: $30B bond offering (largest since 2023), stock dropped 11% (worst day in 3 years), and Wall Street analysts who were bullish 48 hours earlier all downgraded.

What changed? Nothing fundamental. Meta beat earnings on every metric. The only new info: 2026 capex will be "significantly larger" than 2025's $70-72B.

Key findings: - Oppenheimer downgraded from $696 target to no target, citing "Metaverse spending" concerns (the exact bear case they'd been dismissing) - 5 major firms cut price targets 3.6-10% within 24 hours - Fed Chair Powell had to address "circular financing" concerns - CoreWeave's $9B acquisition of Core Scientific collapsed same day

The irony: My Oct 28 article predicted bubble concerns, off-balance-sheet debt becoming a theme, and ROI timeline issues. Bloomberg published "Meta Starting Trend for Off-Balance Sheet Debt" on Oct 31.

Wall Street doesn't predict the future. It narrates the present.

Happy to discuss the analyst flip-flops, the $21.6B in hidden debt (Blue Owl SPV), or whether Zuckerberg's "better to overbuild" strategy is genius or gambling.

I tested Next.js 16 on a 200+ component project. The Turbopack numbers are real:

Development startup: 1083ms (Webpack) → 603ms (Turbopack) = 44% faster

Production builds: 45s (Webpack) → 17s (Turbopack) = 2.6x faster

This isn't a micro-optimization. It's a noticeable difference in daily workflow.

The article also covers: • Astro 5.15 deployment skew protection (solves version mismatch bugs)

• Node 22 LTS native proxy support (useful for enterprise environments)

• CSS View Transitions now Baseline (all major browsers)

• Vite+ announcement (unified JavaScript toolchain, preview early 2026)

October was packed with releases. Happy to discuss any of these updates.

You're touching on the core tension in Meta's strategy. I think you're partially right, but there's more to it.

On "replacing expensive humans" agree that's part of it, but the bigger play is augmenting existing products. Meta's Q3 2025 guidance shows ad revenue still growing 21.6% YoY. They're using AI to make existing ads more effective (better targeting, higher conversion), not replacing the ad model entirely.

On the moat question this is where the infrastructure spending makes sense. You're right that wrapping an LLM has no moat, but owning the infrastructure to train and serve your own models does. Meta has three advantages: (1) 3B+ daily users generating training data competitors can't access, (2) owning 2GW of infrastructure means $0 marginal cost for inference vs paying OpenAI/Anthropic, and (3) AI embedded in Instagram/WhatsApp/Facebook is stickier than standalone chat.

On ads behind chat interface this is the real risk. But Meta's bet seems to be: short-term AI improves existing ad products (already working), mid-term AI creates new surfaces for ads (AI-generated content, business tools), and long-term if chat wins, Meta wants to own the chat interface (Meta AI), not lose to ChatGPT.

The $75B question is whether they're building a moat or just burning cash on commodity infrastructure. Time will tell, but the data advantage plus vertical integration gives them a shot.

What's your take do you think the data moat is real, or can competitors train equally good models on synthetic/public data?

Interesting theory, but the current numbers and recent developments suggest a different picture.

Meta's Q3 2025 ad revenue is expected at 48.5B(21.666-72B, so still a significant cost-to-revenue gap.

Regarding advertising in AI assistants specifically OpenAI's Sam Altman actually addressed this in June 2025, calling ads in ChatGPT a "trust-destroying moment" (per The Decoder). There's been internal pushback at OpenAI over "engagement farming" tactics, and users already assume product suggestions are sponsored, which creates trust issues.

The more realistic monetization path for Meta (based on their earnings guidance):

Better ad targeting through AI (already happening - 11% increase in ad impressions Q2 2025) AI business tools (Meta AI for businesses) Infrastructure-as-a-service (selling excess capacity) Direct advertising in AI responses faces major regulatory and user trust hurdles. ChatGPT reached 800M weekly users by Sept 2025, but monetization through subtle product placement has sparked backlash even internally at OpenAI.

Sources: Meta Q3 2025 earnings preview (LSEG, Nasdaq), The Decoder (June 2025), Yahoo Finance marketing analysis

Good question. The financing structure is actually quite interesting it's not traditional equity investment.

For the Hyperion deal specifically (confirmed by Meta's official announcement Oct 21, 2025):

80% owned by Blue Owl Capital (private credit firm) Financed through $27B+ debt arranged by Morgan Stanley PIMCO as anchor lender (144A bonds, maturing 2049) Meta has 20% equity + 16-year residual value guarantee So Meta's "guarantee" is essentially a long-term lease commitment with a 4-year initial term + extension options. If it goes bust, Meta is still on the hook for 16 years of payments, but Blue Owl/PIMCO absorb most of the asset risk.

The other deals (CoreWeave 14.2B,Oracle 20B) are traditional service contracts Meta pays for capacity, vendors own the infrastructure.

This is actually the largest private capital deal on record according to Bloomberg (Oct 16, 2025).

Sources: Meta official announcement, Bloomberg, Data Center Dynamics

I spent weeks tracking Meta's AI infrastructure deals from September-October 2025. The scale is unprecedented: $75.5 billion across 4 deals in just 3 months.

Key findings: • CoreWeave: $14.2B (6+ years, Nvidia GB300 GPUs) • Oracle: ~$20B (multi-year cloud deal) • Blue Owl/Hyperion: $27B (joint venture, private credit financing) • Scale AI: $14.3B (49% stake)

What's interesting isn't just the size—it's the structure. Meta is using private credit and joint ventures instead of traditional CapEx. The Hyperion deal: Meta owns 20%, Blue Owl owns 80%, but Meta is on the hook for 16 years.

The math is uncomfortable: $75B in infrastructure spending vs ~$7B in AI revenue. That's a 10:1 cost-to-revenue ratio.

Happy to discuss the economics, the private credit angle, or whether this is sustainable.