Adam

exTencent (AI/IoT/MobileMarketplace Lead) | Junior Startup | Earlytech adopter

[Z.ai @Zai_org]

Introducing GLM-5.2: Frontier Intelligence, Open Weights

- Significant improvements in coding and agentic tasks
- Strong long-horizon capabilities with a 1M context window
- Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency
- MIT-licensed open weights
- Same API pricing as GLM-5.1

Tech Blog: https://t.co/LAsxUdN0JZ
Weights: https://t.co/g0A1C4UWx4
API: https://t.co/Kc3E22cbN7
Coding Plan: https://t.co/Nk8Y98HNhU
Chat: https://t.co/WCqWT0qCQb

[Watcher.Guru @WatcherGuru]

JUST IN: Elon Musk's SpaceX signs $920,000,000 per month deal with Google to provide AI computing power.

[Adam @adamlogs]

NVIDIA is no longer just a GPU company — it's becoming a full-stack compute system company.

We are seeing 3 structural moves happening in parallel:

1. CPU expansion (Arm-based “PC era” entry)
→ NVIDIA is moving from GPU-only into CPU + GPU integration

2. AI software stack tightening
→ Frameworks like Polar are not just tools — they are performance amplification layers
→ Reported benchmark gains (+594%) show how tightly hardware + software are now coupled

3. OEM ecosystem lock-in
→ Dell / Lenovo / enterprise PC stack is being pulled into NVIDIA’s architecture direction

---

What this actually means:

We are moving from:
“chip competition”
→ to “compute stack competition”

Where the winner is not the best chip,
but the most vertically integrated system:

CPU + GPU + AI framework + OEM alignment

---

Implication:

NVIDIA is increasingly behaving like:
→ Apple (hardware + software integration)
→ AWS (infrastructure dominance)
→ but for AI compute

This is a 5–10 year structural shift, not a product cycle.

Key question going forward:
Can any other company replicate a full-stack AI compute system?

#Nvidia #AI

[Adam @adamlogs]

Embodied Intelligence / World Model Wave: τ0-WM + Gamma-World + Robot World Models

Layer: Layer 1 — Models / Layer 2 — Platforms

τ0-WM: Largest pre-trained open-source embodied world model released by AGIbot https://t.co/W1B1xfVYQc
NVIDIA-Tsinghua team proposes Gamma-World: from "single-player" to "multi-agent coexistence"
Fudan team creates spatiotemporal unified architecture for robot world action models
World models succeed language models; physical AGI "dual pyramid" system proposed

#nvidia #tsinghua #fudan #Agibot #AI #AGI

[Adam @adamlogs]

NVIDIA Self-Designed CPU Enters PC Market — "NVIDIA's MacBook Pro" Revealed

NVIDIA is entering the PC processor market with a self-designed Arm-based CPU, dubbed the "NVIDIA MacBook Pro"
NVIDIA launches AI framework Polar, boosting Codex benchmark scores by 594.74%
Jensen Huang on AI applications: "Wasting a little money is fine; don't waste time"
Dell stock surges 32.8% on AI infrastructure demand; Michael Dell's wealth increases by $32.2B in a single day

Investment Analysis:

Layer: Layer 0 — Compute & Memory Infrastructure
Who's investing: NVIDIA (major corp ×1.2), Dell/Lenovo OEMs (major corp ×1.2), capital markets (sentiment ×0.6)
What: Self-designed Arm CPU, AI PC chips, GPU+CPU integrated architecture (https://t.co/oW6JLzaDTW)
Duration: Long-term (5+ years), chip design cycles typically 3-5 years
Intersection: NVIDIA + OEMs + OS ecosystem + AI inference demand = multi-party resonance
Dominant Factor: NVIDIA is the only company offering GPU+CPU+AI software stack simultaneously → Single-vendor dominant. Its decisions directly drive OEM alignment; irreplaceable. Chip architecture lifespan 10+ years.

[Adam @adamlogs]

BYD 4nm Self-Designed AI Chip + Huawei's 381-Chip "Overturning Moore's Law"

BYD releases a 4nm self-designed AI chip matching Nvidia's process node, claiming compute exceeding Tesla's FSD chip. Huawei's Eric Xu reveals 381 chips forming a domestic chip ecosystem, Jensen Huang comments on "Huawei's Tau Law."

#BYD #huawei #Nvidia #AIChips #China

[Adam @adamlogs]

SoftBank €75B Investment in French Data Centers — AI Infrastructure Arms Race Escalates

SoftBank announced plans to invest up to €75 billion to build data centers in France, the largest single investment commitment in European AI infrastructure to date.

Layer: Layer 0 — Compute & Memory Infrastructure

Investment Analysis:

Who's Investing: Government (French backing) × Enterprise (SoftBank) × Cross-border capital → Tri-party resonance
What: AI compute infrastructure → data center clusters
How Long: Long-term (5+ year construction cycle)
Intersection: EU AI sovereignty strategy + SoftBank global AI infra + French low-cost nuclear power
Dominant Factor: 🛡️ Policy/Regulatory Dominance — EU AI sovereignty laws driving nations to compete for AI compute landing, policy enforcement 5+ years

French nuclear advantage (low-cost clean power) is the core attraction, irreplaceable
EU AI Act requires local data processing, policy rigidity drives demand

#softbank #datacenter #AI

[Adam @adamlogs]

DeepSeek V4 Chip-Model Co-Design + BYD's 4nm Custom AI Chip: China's Domestic Compute Ecosystem Accelerates

> QbitAI reports DeepSeek V4 achieves chip-model co-design, with China's domestic compute ecosystem entering an accelerating flywheel. Separately, QbitAI reports BYD's custom AI chip uses 4nm process, matching Nvidia's process node and surpassing Tesla in compute power.

Layer: Layer C (China AI Ecosystem) / Layer 0 (Compute)
Investors: DeepSeek (major ecosystem) + BYD (Major Tech ×1.2) + Government policy (×1.5)
Investment Scale: Hundreds of millions in R&D + policy funding, long-term (5+ years)
Intersection: Enterprise + Government dual investment → two-party resonance
Dominant Factor: 🏛️ Policy/regulatory dominated — Domestic substitution is a national-level strategy with 10+ year execution cycle, extremely high irreplaceability

[Adam @adamlogs]

Groq Raises $650M + XCENA Raises $135M: AI Inference Chips & Memory Bandwidth — Two Parallel Bets

> TechCrunch reports that after Nvidia's $20B non-acqui-hire, AI inference chip startup Groq is raising $650M. On the same day, chip startup XCENA closed $135M, betting that AI's biggest bottleneck is memory bandwidth, not compute.

Layer: Layer 0 (Compute & Memory Infrastructure)
Investors: VC (×1.0) + Big Tech ecosystem (Nvidia-adjacent)
Investment Scale: Groq $650M + XCENA $135M = ~$800M in a single day
Intersection: VC + Big Tech + Academia all investing in AI chips → three-party resonance
Dominant Factor: 🔧 Demand-driven — AI inference cost and memory bandwidth are industry-wide technical bottlenecks, not dependent on any single vendor

#groq #XCENA #techcrunch #AI #AIChip

[Adam @adamlogs]

Exactly — it's a pretty big refactor. My guess is they're leaning on a shared graph / vector + relational backbone (probably built on top of their existing Fabric / Azure infrastructure) for cross-product state and memory. That would let agents in the super app have consistent context across GitHub, Chatbot, Coworker, etc.We'll know more next week at Build when they show the actual architecture. I'm expecting some kind of unified memory layer + semantic index to handle consistency.

[Adam @adamlogs]

Microsoft Building AI "Super App" — Consolidating All Copilot Products + New "Autopilot" Agent Workflow

> The Verge exclusively reports Microsoft is developing an AI super app combining GitHub Copilot, Copilot Chatbot, Copilot Cowork, and a new agentic workflow capability internally named "Autopilot."

Expected to debut at Microsoft Build next week.

Layer: Layer 2 (Agent Platform) / Layer 3 (Application)
Investors: Microsoft (Major Tech ×1.2)
Investment Scale: Cross-divisional consolidation, long-term (5+ years)
Intersection: OpenAI also pursuing super app → two-party resonance
Dominant Factor: 🏢 Single-vendor dominated (Microsoft) — Owns the full stack: OS (Windows), Office Suite, Developer Platform (GitHub), Cloud (Azure). Their decisions directly pull the entire enterprise AI ecosystem.

#microsoft #githubcopilot #AI #AIAgent

[Adam @adamlogs]

@pangshuo1981 respect to pragmatic builders 🫡

[Adam @adamlogs]

AI coding startup Cognition (maker of Devin) closes $1B funding at $25B pre-money valuation. DeepSeek reportedly raised ¥70B RMB, led by ACM gold medalist Cui Tianyi building DeepSeek Code. OpenAI Codex self-distillation techniques go viral in developer community.

Layer: Layer 2/3 (Agent Platform + Application)

Who's investing: Big Tech ×1.2 (OpenAI/Google) + VC ×1.0 (top-tier funds backing Cognition) + Community ×0.8 (Codex tutorials spreading)

Timeline: Mid-to-long term (2-5 years, AI coding from assistive → autonomous → full-stack Agent)
Intersection density: Multi-party ×1.5 (OpenAI, Cognition, DeepSeek, Cursor, Windsurf, ByteDance Trae all placing bets)

Dominant factor: Demand-driven ⭐⭐⭐⭐ — All software companies have inherent needs for developer productivity

#deepseek #cognition #codex #openai