科技推特精选 - 2026年3月18日

2026年3月18日科技每日简报

Today's top tech conversations are led by @NVIDIAAP, whose post about 'Physical AI and autonomous AI ...' garnered the highest engagement. Key themes trending across the top stories include nvidia, demand, openai, inference, revenue. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. NVIDIAAP (Group Score: 154.0 | Individual: 24.1)

Cluster: 9 tweets | Engagement: 2 (Avg: 69) | Type: Tech

Physical AI and autonomous AI agents are reinventing the world of manufacturing.

#NVIDIAGTC news: @Cadence, @Dassault3DS, @Siemens, and @Synopsys are building NVIDIA-powered AI agents to plan, optimize, and verify complex chip and system workflows.

🔗 https://t.co/e6gknMsaEe https://t.co/tCaCrpwt97

See 8 related tweets

@ReutersBiz: Nvidia raises its outlook for AI chip sales, while a startup aims to build robots for defense as wel...
@NVIDIAAP: NVIDIA Vera Rubin is opening the next frontier of AI.

#NVIDIAGTC news: The Vera Rubin platform’s se...

@NVIDIAAP: #NVIDIAGTC news: Physical AI leaders across robot brain developers, industrial robot giants and huma...
@Cadence: AI factories are redefining data centers.

With the Cadence Reality Digital Twin Platform + @NVIDIA ...

@NVIDIAOmniverse: RT @Cadence: Together with @NVIDIA, we're delivering accelerated engineering solutions purpose-built...

2. jukan05 (Group Score: 137.1 | Individual: 33.4)

Cluster: 5 tweets | Engagement: 69 (Avg: 187) | Type: Tech

《GF Overseas Electronics & Communications》 ☄️ March 16 GTC 2026 Keynote + Silicon Valley Channel Check Call Summary

☀️ Key Takeaways from the GTC Keynote

AI scaling laws have evolved from the traditional pre-train / post-train / inference stages toward Agent AI, with the longer-term direction pointing to Physical AI. We expect the Agent AI era to drive both GPU and CPU demand, and NVIDIA’s architectural messaging at GTC was clearly focused on improving inference performance.
Revenue visibility for NVIDIA Blackwell and Rubin from 2025 to 2027 has reached $1 trillion, implying roughly$ 500 billion in data center revenue by 2027, well above the market’s general expectation of $350–390 billion.
NVIDIA unveiled Groq 3 LPU, featuring 500MB of SRAM, built on Samsung 4nm, and offering extremely high bandwidth (150TB/s). The specifications were broadly in line with expectations. The system uses a 32 Compute Tray architecture, with 8 chips per tray, for a total of 256 chips.
The LPX Rack has been integrated into the Vera Rubin system product lineup, and this will also extend to Feynman later on, strengthening NVIDIA’s competitiveness in the inference market. Demand forecast: LPX racks are expected to reach 500 racks in 2026 and preliminarily 10,000 racks in 2027, which is positive for leading domestic PCB vendors.
Highlights of the NVIDIA Vera Rubin NVL72 architecture include the aforementioned LPX integration, a strong emphasis on inference, and mention of urgent demand for storage, especially SSDs.
NVIDIA Vera Rubin NVL144 uses an orthogonal backplane in place of copper cables, in line with expectations, and this is also beneficial for PCB and connector vendors.
NVIDIA ETL256 adopts an MGX design with 8 GPUs per tray, for a total of 256 GPUs. The middle 8 switch trays use Spectrum X rather than NVLink.
CPO scale-up came in below bullish expectations: NVIDIA indicated that both copper and optical solutions can be used for scale-up in the Oberon architecture (via Spectrum X CPO) and the Kyber architecture (up to NVL1152), rather than the market’s previous optimistic assumption that everything would shift fully to optical. Optical interconnect-related names may see some short-term profit-taking, but the medium- to long-term outlook remains positive.

☀️ Feedback from U.S. Institutional Roadshows

Investors had previously become more risk-averse due to the Middle East conflict, rising oil prices, and renewed inflation expectations. But after TACO, confidence gradually recovered, and capital has flowed back into optics (though GTC could trigger short-term profit-taking), memory, and CPU-related names such as substrates.
The broader tone remains constructive on AI, supported by:

accelerating commercialization of applications this year,
sufficient capital market funding for AI infrastructure,
strong infrastructure buildout through 2027,
patience for lagging AI names to strengthen, and
continued focus on TSMC, CPUs, CPO, optics vs. copper, memory, as well as the opportunities higher memory prices may create for Apple.

☀️ Feedback from U.S. Supply Chain Checks

Google TPU demand remains very strong, and 2027 capacity is already sold out. Shipment demand is far above our previous forecast of 6 million cards, with upside coming from external sales of Broadcom-related projects.
HBM demand is extremely strong, with visibility extending through 2027. Samsung is making the fastest progress on HBM4. Architecturally, AI still depends heavily on HBM’s high bandwidth and speed, though it remains worth monitoring how NVIDIA software optimization and the launch of LPUs may affect HBM demand.
This year, the core theme of AI model iteration is still commercialization and Agentic AI. Groq is well-suited for 1-on-1 multi-turn conversational inference, while CUDA remains the core moat. The KV cache required for inference continues to drive demand for DRAM, SSDs, and context memory.

☀️ Feedback from U.S. Company Meetings

Coherent (COHR, Hold): NVIDIA’s $2 billion investment is mainly focused on CPO, in addition to supply agreements worth several more billions of dollars. The company is optimistic on the CPO opportunity and believes scale-up is accelerating, with both market opportunity and penetration potential higher than in scale-out. Its transition to 6-inch InP capacity, compared with the early stage of 3-inch production, has progressed more smoothly, and full-year capacity is expected to double. The outlook for OCS demand is also positive, and in addition to CSP customers, another CSP is currently under discussion. COHR occupies an important position in the optical module and CPO supply chain and is worth monitoring.
Intel (INTC, Buy): Yield progress on the 18A process is going smoothly, and it is expected to reach industry-standard levels by year-end. 14A is being developed in close cooperation with customers and is expected to progress even more smoothly than 18A did at the same stage. Structural demand for server CPUs remains strong, with double-digit growth expected this year. EMIB is expected to begin contributing revenue in 2H27. We remain positive on the stock, supported by CPU tailwinds, capacity conversion, and solid foundry progress.
SMCI (Hold): Revenue guidance has been raised, but shipment timing is being affected by major customers (xAI) and data center facility readiness. Its advantage lies in rapid deployment and engineering support. The company is relatively conservative on memory procurement in the second half. Given competitive pressure from Dell, we remain cautious on both its margins and revenue growth.

☀️ In addition, we have conducted supply chain and listed company checks across multiple non-expert companies. More details will be shared in follow-up calls and research reports.

See 4 related tweets

@FirstSquawk: NVIDIA CORP. CEO JENSEN HUANG FORECASTS AI PROCESSORS COULD GENERATE $1 TRILLION IN SALES BY 2027 AS...
@Reuters: At Nvidia's annual developer conference in San Jose, California, CEO Jensen Huang discussed how the ...
@ReutersBiz: WATCH: Nvidia said the revenue opportunity for its AI chips may reach at least $1 trillion through 2...
@Reuters: Nvidia said the revenue opportunity for its AI chips may reach at least $1 trillion through 2027, as...

3. theinformation (Group Score: 101.8 | Individual: 14.4)

Cluster: 9 tweets | Engagement: 29 (Avg: 9) | Type: Tech

Exclusive: OpenAI has signed a new AWS contract to sell AI to the U.S. government. A critical change to OpenAI’s partnership with Microsoft made it possible.

Read more from @SriMuppidi and @aaronpholmes 👇 https://t.co/tT5UyDwIf0

See 8 related tweets

@FirstSquawk: OPENAI SPOKESPERSON: CO SIGNED AN AGREEMENT WITH AWS TO ENABLE DEPLOYMENT OF OPENAI MODELS ON U.S. G...
@MSBIntel: BREAKING: OpenAI will provide AI to U.S. government and defense agencies through AWS, covering both ...
@theinformation: OpenAI just signed a deal with AWS for government contracts, @srimuppidi exclusively reported this m...
@theinformation: Exclusive: OpenAI has struck a deal with Amazon Web Services to sell AI tools to U.S. government age...
@techdaily24: OpenAI has reportedly partnered with AWS to provide its AI models to U.S. government agencies. This ...

4. Austen (Group Score: 80.2 | Individual: 30.0)

Cluster: 3 tweets | Engagement: 346 (Avg: 197) | Type: Tech

App Store review is one of the first columns of the software ecosystem to just completely buckle under the weight of AI.

It almost makes building apps not worth it until Apple gets its stuff in order.

See 2 related tweets

@seraleev: Apple: 90% of apps are reviewed in 24 hours

Reality: App Review now takes longer than building the ...

@Austen: “Why is App Store review taking so long?” He complained as his agent submitted the 5 new apps it had...

5. WSJ (Group Score: 77.5 | Individual: 25.5)

Cluster: 4 tweets | Engagement: 142 (Avg: 224) | Type: Tech

Nvidia CEO Jensen Huang unveiled NemoClaw, a new software tool kit to help businesses safely build and run autonomous agents with OpenClaw https://t.co/ZE3ER5uaCn

See 3 related tweets

@RoundtableSpace: NVIDIA JUST DROPPED NEMO CLAW TO MAKE OPENCLAW SAFE TO RUN ON YOUR OWN MACHINE.

It keeps private fi...

@NVIDIAAP: Ready to deploy AI agents? NVIDIA NemoClaw simplifies running OpenClaw always-on assistants with a s...
@NVIDIAAP: #NVIDIAGTC news: NVIDIA announces NemoClaw for the OpenClaw agent platform.

NVIDIA NemoClaw install...

6. NVIDIAAIDev (Group Score: 75.5 | Individual: 20.1)

Cluster: 5 tweets | Engagement: 250 (Avg: 230) | Type: Tech

RT @arthurmensch: Looking forward to building frontier open source AI models together with @Nvidia as we join the Nemotron Coalition and st…

See 4 related tweets

@NVIDIAAIDev: RT @MistralAI: 🚀Announcing a strategic partnership with NVIDIA to co-develop frontier open-source AI...
@theinformation: Jensen Huang’s remarks on Nvidia’s open-source progress “was a bit of a dig at Meta and DeepSeek”, F...
@NVIDIAAP: Space computing, the final frontier, has arrived 🛰️

#NVIDIAGTC news: NVIDIA is collaborating with @...

@Rajananandan: RT @chandrarsrikant: Nvidia taps Sarvam, Thinking Machines Lab, Mistral AI for a coalition to advanc...

7. IamEmily2050 (Group Score: 74.5 | Individual: 29.6)

Cluster: 3 tweets | Engagement: 54 (Avg: 183) | Type: Tech

The biggest mistake OpenAI made was focusing on the consumer market. There is no value from free accounts wanting to keep GPT4o; all the money and high quality data is in the enterprise sector. Hopefully this also means they will fix the political bias in their model and balance the models. No need to make a group of people happy, especially when they are not even paying for the service.

See 2 related tweets

@cryptopunk7213: f*ck yes openai just dropped 2 new models thats 2X faster and cheaper but AS GOOD as gpt-5.4 (it act...
@AndrewCurran_: The WSJ is reporting that OpenAI is about to take a hard turn into enterprise. https://t.co/wEfk5Ft9...

8. startupideaspod (Group Score: 74.1 | Individual: 48.1)

Cluster: 3 tweets | Engagement: 6187 (Avg: 754) | Type: Tech

"OpenClaw is the new computer." — Jensen Huang

This is the early PC era all over again.

A few power users see it.

Everyone else hasn't even started.

"It's the most popular open source project in the history of humanity, and it did so in just a few weeks. It exceeded what Linux did in 30 years."

A solo founder with OpenClaw can now build what used to take a 50-person team.

The leverage is absurd.

See 2 related tweets

@RoundtableSpace: JENSEN HUANG IS CALLING AGENTIC SYSTEMS THE NEW COMPUTER.

And OpenClaw is already being framed as o...

@hwchase17: RT @heyshrutimishra: The biggest software bet Jensen made today: NemoClaw.

He called OpenClaw "the ...

9. NVIDIAAP (Group Score: 63.9 | Individual: 29.8)

Cluster: 3 tweets | Engagement: 8 (Avg: 69) | Type: Tech

#NVIDIAGTC news: Introducing the new NVIDIA Vera Rubin DSX AI Factory reference design.

This new reference design and Omniverse DSX blueprint are providing the foundation to build the world’s most productive AI factories, accelerating time to first revenue and maximizing scale and energy efficiency.

🔗 https://t.co/xkHmISxjoa

See 2 related tweets

@nvidia: RT @nvidiaomniverse: AI factories are essential infrastructure in the age of AI, but building them a...
@NVIDIAAP: #NVIDIAGTC news: NVIDIA announces the new NVIDIA Physical AI Data Factory Blueprint turns accelerate...

10. ns123abc (Group Score: 61.8 | Individual: 27.7)

Cluster: 3 tweets | Engagement: 682 (Avg: 1303) | Type: Tech

🚨 OpenAI CEO of apps tells staff Anthropic success should be a "WAKE-UP CALL"

"we are very much acting as if it's a code red" claude code has become default tool at every tech company claude cowork absolutely dominating enterprise anthropic's enterprise share has doubled openai's dropped from 50% to 34% altman bet on "a series of startups" inside openai sora, atlas web browser, ChatGPT e-commerce... all of them failed

meanwhile anthropic just focused on 2 things and won silently shipping new products everyday openai investoors absolutely seething fidji simo tells staff this should be a "wake-up call" "we cannot miss this moment because we are distracted by side quests"

sora, atlas browser, ChatGPT e-commerce, jony ive hardware device ALL getting deprioritized refocusing entire company on coding and enterprise deploying engineers with consulting firms to try accelerate adoption

OpenAI new strategy = copy Anthropic

See 2 related tweets

@FirstSquawk: OpenAI’s top executives are finalizing plans for a major strategy shift to refocus the company aroun...
@Cointelegraph: 🚨 NOW: OpenAI is shifting focus away from side projects to double down on coding and enterprise busi...

11. CNBC (Group Score: 61.5 | Individual: 24.3)

Cluster: 3 tweets | Engagement: 62 (Avg: 61) | Type: Tech

Alibaba launches agentic AI tool for businesses with Slack, Teams integration plans https://t.co/Xfa7c4T6M7

See 2 related tweets

@techdaily24: https://t.co/ozFgsNC2PV is an AI employee tool designed to help users "hire" an AI for various roles...
@Reuters: Alibaba launches AI platform for enterprises as agent craze sweeps China https://t.co/hxALmuMicr htt...

12. KirkDBorne (Group Score: 60.9 | Individual: 60.9)

Cluster: 1 tweets | Engagement: 1019 (Avg: 21) | Type: Tech

Hermes Agent v0.3.0 ☤

248 PRs. 15 contributors. 5 days.

• Real-time streaming across CLI and all platforms • First-class plugin architecture, package and share tools+commands+skills • /browser connect to live Chrome via CDP • @vercel AI Gateway model provider • @browser_use browser tool provider • VS Code, Zed, and JetBrains integration • Voice mode with local Whisper • PII redaction everywhere

9 new skills. 50+ bug fixes. Much more in the full changelog.

13. theinformation (Group Score: 59.2 | Individual: 29.9)

Cluster: 2 tweets | Engagement: 2 (Avg: 9) | Type: Tech

Today on TITV:

-Nvidia projects $1T chip revenue through 2027 | @nickpatience, The Futurum Group; @aginnt, Hydra Host

-OpenAI secures AWS deal for government contracts | @srimuppidi, OpenAI & Anthropic Reporter at The Information

-Mastercard to acquire stablecoin startup BVNK | @kenbrown12, Finance Editor

-@Asana CEO Dan Rogers on its new AI agent tool

📺 Tune in at 10 am PT / 1 pm ET on https://t.co/iyO8ht485m

See 1 related tweets

@theinformation: Today on TITV:

-Nvidia projects $1T chip revenue through 2027 | @nickpatience, The Futurum Group; @...

14. aakashgupta (Group Score: 59.1 | Individual: 30.8)

Cluster: 2 tweets | Engagement: 33 (Avg: 330) | Type: Tech

Most product managers are still using AI the wrong way.

They open ChatGPT. They ask a question. They get an answer. They move on. The interaction is reactive, manual, and one-off.

Our Guest explained why @openclaw represents a fundamentally different model - one where AI agents run autonomously on your behalf, 24/7, without being prompted.

The example that stood out:

"You can configure OpenClaw to wake up every 30 minutes and look up our competitors website and see what changes they have made without you having to manually do that."

No prompt. No conversation. No manual check-in. The agent just runs.

Then the line that reframed it all:

"Instead of being conversational, this is executive. So you're not just talking, you're taking charge."

Lesson: The next generation of product managers will not be defined by how well they prompt AI. They will be defined by how well they deploy autonomous agents to do the work before they even think to ask.

See 1 related tweets

@rauchg: Every month I periodically see the recycled take that “something better than chat” is coming for AI....

15. BitcoinNews (Group Score: 56.2 | Individual: 26.7)

Cluster: 3 tweets | Engagement: 539 (Avg: 153) | Type: Tech

JUST IN: 🌐 Coinbase 🤝 Sam Altman's World partner to verify real human identity behind AI transactions.

AgentKit, enables World-verified individuals to delegate their World IDs to AI agents on Coinbase's x402 protocol. https://t.co/OnNTlAagXv

See 2 related tweets

@CoinMarketCap: LATEST: ⚡ Sam Altman's World has launched AgentKit, a developer toolkit that allows AI agents to cry...
@arstechnica: World ID wants you to put a cryptographically unique human identity behind your AI agents https://t....

16. Reuters (Group Score: 54.7 | Individual: 32.8)

Cluster: 2 tweets | Engagement: 227 (Avg: 225) | Type: Tech

Nvidia CEO Jensen Huang said the company could generate at least $1 trillion in revenue from its AI chips by 2027, as it ramps up competition in the fast-growing real-time AI computing market. Read: https://t.co/t4GT38L4kZ https://t.co/qYAjNCuRFk

See 1 related tweets

@business: Nvidia CEO Jensen Huang said the $1 trillion projection for the company’s AI chip sales doesn’t capt...

17. dkundel (Group Score: 54.4 | Individual: 25.7)

Cluster: 4 tweets | Engagement: 456 (Avg: 117) | Type: Tech

GPT-5.4-mini is a wildly capable model and gives you ~3.3x more usage on Codex tasks compared to GPT-5.4. It's excellent for spinning up new subagents! https://t.co/OX8GtEMevR

See 3 related tweets

@kimmonismus: GPT-5.4 mini available today, looks very good so far. Going to dig into it later https://t.co/9Tw7U3...
@scaling01: GPT-5.4-mini looks really good for computer-use https://t.co/NfYb8pcPME...
@OpenAIDevs: RT @dkundel: GPT-5.4-mini is a wildly capable model and gives you ~3.3x more usage on Codex tasks co...

18. UnslothAI (Group Score: 51.3 | Individual: 25.9)

Cluster: 2 tweets | Engagement: 76 (Avg: 904) | Type: Tech

Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs.

• Run models locally on Mac, Windows, Linux • Train 500+ models 2x faster with 70% less VRAM • Supports GGUF, vision, audio, embedding models • Auto-create datasets from PDF, CSV, DOCX • Self-healing tool calling and code execution • Compare models side by side + export to GGUF GitHub: https://t.co/2kXqhhvLsb Blog and Guide: https://t.co/ENuTWal5AA

Available now on Hugging Face, NVIDIA, Docker and Colab.

See 1 related tweets

@SergioPaniego: RT @UnslothAI: Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs.

• Run m...

19. business (Group Score: 50.8 | Individual: 19.8)

Cluster: 3 tweets | Engagement: 95 (Avg: 147) | Type: Tech

Microsoft is uniting the teams that build versions of its Copilot for individuals and corporations, streamlining product development as the company works to get its AI assistant into the hands of more customers https://t.co/UOIxhQCqWt

See 2 related tweets

@FirstSquawk: MICROSOFT IS REORGANIZING COPILOT AI PRODUCT TEAMS TO UNIFY COMMERCIAL AND CONSUMER VERSIONS - WSJ...
@edzitron: hearing microsoft is once again reorganizing its AI team under the banner of "the Copilot System." A...

20. RoundtableSpace (Group Score: 46.9 | Individual: 46.9)

Cluster: 1 tweets | Engagement: 4684 (Avg: 366) | Type: Tech

A STUDENT USED ANTHROPIC AI TO TURN ~ $1K INTO$ 1.5M ON POLYMARKET WITH 44K TRADES AND A QUANT-DRIVEN BOT.

https://t.co/nfWym4rOPW