科技推文精选 - 2026年5月4日

2026年5月4日科技每日简报

Today's top tech conversations are led by @pmarca, whose post about 'Co-sign.\n\nQT @DavidSacks: I’...' garnered the highest engagement. Key themes trending across the top stories include google, building, cloud, harness, https. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. pmarca (Group Score: 125.3 | Individual: 36.2)

Cluster: 4 tweets | Engagement: 564 (Avg: 532) | Type: Tech

Co-sign.\n\nQT @DavidSacks: I’ve been saying for awhile that AI capex will be a 2% tailwind to GDP growth this year. In fact, according to a new report from Morgan Stanley, the numbers are even stronger — more like 2.5% this year and over 3% next year.

And this understates the impact of AI for two reasons:

(1) This is just investment by 5 hyperscalers; it doesn’t include all the startups and other companies investing in AI.

(2) Capex is the investment to create the token factories; it doesn’t count the economic activity resulting from what happens inside the token factories. Those tokens are now being used to generate code (bespoke software) that will increase productivity throughout the economy. The ROI on capex is likely to dwarf the capex itself, which is why investment continues to grow.

In Q1, AI was already 75% of GDP growth. That trend is likely to continue. Technology leadership has always been America’s great strength, and it’s driving the economy forward.

Polls may show that AI is not popular, but economic growth is. At this point, stopping progress in AI would be equivalent to halting the U.S. economy.

See 3 related tweets

@DavidSacks: I’ve been saying for awhile that AI capex will be a 2% tailwind to GDP growth this year. In fact, ac...
@pmarca: RT @DavidSacks: I’ve been saying for awhile that AI capex will be a 2% tailwind to GDP growth this y...
@cryptopunk7213: my god so we basically have 5 of the richest companies in the world spending $800 Billion on ai cape...

2. StockSavvyShay (Group Score: 116.4 | Individual: 35.1)

Cluster: 4 tweets | Engagement: 369 (Avg: 472) | Type: Tech

$AMD CEO Lisa Su said at CES 2026 that the world may need 100x more compute over the next five years as AI workloads scale across industries.

This earnings cycle is proving the point: • $AMZN AWS reaccelerated to 28% (fastest growth in 15 quarters) while Bedrock customer spend grew 170% QoQ & processed more tokens in Q1 than all prior years combined •$ MSFT AI business is now at a $37B run-rate growing 123% YoY •$ GOOGL Cloud grew 63% (now nearly half the size of AWS) & has a $243B backlog while GenAI products grew 800% YoY •$ MU & $SNDK margins show memory demand is still compounding WAY faster than supply with margins now higher than many software companies

The 100x compute framework looks less crazy after this earnings cycle.\n\nQT @StockSavvyShay: THE COMPUTE CAPACITY BOTTLENECK

$GOOGL just admitted Google Cloud is leaving revenue on the table because it cannot build capacity fast enough with shifts the bottleneck to companies with the power, real estate & operational scale to deploy AI compute:

$NBIS building the AI-native cloud layer through vertically integrated GPU clusters & software optimized for training + inference.$ NVDA just wrote wrote a $2B check & Nebius now has a$ 46B contracted backlog, anchored by ~ $19B$ MSFT deal & ~$27B Meta partnership through 2032.
$IREN building the renewable-powered AI compute layer by turning low-cost power into GPU cloud capacity. The pivot to AI cloud is now backed by a ~$ 10B $MSFT contract, 2.9 GW of grid-connected power expanding to 4.5 GW+ & targeted 140K GPU buildout that could drive$ 3.4B of ARR by year-end 2026.
$DOCN building the agentic inference cloud layer for developers & long-tail AI workloads. AI customer ARR is up 150% YoY to$ 120M, over 70% comes from inference services & $1M+ customer ARR is up 123% to$ 133M.
$CRWV building the dedicated AI cloud platform for frontier model developers. The company has ~$ 67B of contracted revenue backlog (nearly $88B including Anthropic) with major commitments from OpenAI & Meta.
$CIFR building the Google-backed AI data center layer through contracted power & hyperscale leases. Barber Lake has 300 MW fully contracted with Fluidstack (Google backstops$ 1.4B with a ~5% equity stake) and AWS signed a separate $5.5B 15-year deal for another 300 MW of capacity.
$WULF building the power-backed AI compute layer through long-term data center leases. Lake Mariner has 360 MW tied to Fluidstack backed by a$ 3.2B Google guarantee & new Abernathy JV adds 168 MW over 25 years representing $9.5B in contracted revenue with$ 1.3B of Google lease support.
$APLD building the purpose-built AI data center layer through its Polaris Forge 1 campus in North Dakota. The full 400 MW critical IT load is contracted to CoreWeave under ~15-year leases worth ~$ 11B in expected revenue with first 100 MW delivered in Q4 2025 & another 300 MW targeted through 2027.

See 3 related tweets

@alex_prompter: Your $20/month AI subscription is a lie.

Not a scam. A subsidy.

Every time you send a complex quer...

@pvergadia: 👀7 layers turn electricity into AI products. Most people only think about 3 of them. Here's the full...
@aakashgupta: Google passed Apple for the #2 spot in January. Four months later it sits $230 billion behind NVIDIA...

3. FredKSchott (Group Score: 107.9 | Individual: 42.0)

Cluster: 3 tweets | Engagement: 168 (Avg: 48) | Type: Tech

in case you’re wondering what the moat is in software these days: about 48hr\n\nQT @Shashikant86: 🎉Introducing PyFlue: The Python-Native Agent Harness Framework.🧰

💡Flue for Python: Fred K. Schott @FredKSchott CEO of HTML has launched Flue: The Agent Harness Framework for TypeScript. It brings programmable harness right into your agents rather than DIY plumbing. Python ecosystem already has 🦾 powerful AI/ML tools and frameworks and research initiatives but most frameworks asked users to build your own harness. Superagentic AI bringing this concept of Flue to Python 🐍 ecosystem. Here is PyFlue even even better 🤖 Agent = Model 💻 + Harness 🧰 + Memory 🧠 Almost all the feature of Flue plugged with @LangChain Deepagents harness built by @hwchase17 and team @Vtrivedy10 @sydneyrunkle and more coming soon. 👉 Stop building agent loops, start using a harness.

💻 Try PyFlue Now : https://t.co/GVo06OWBNB

⭐️ GitHub: https://t.co/HH2DXvzDyj 📚Docs: https://t.co/p6GXRyKuW9 📙 Blog Post: https://t.co/MZywMCeFQD

#HarnessEngineering #AgentHarnesses

See 2 related tweets

@hwchase17: Ha love to see this port of flue to python

Harness engineering is a fun time! Need more people expl...

@Shashikant86: 🎉Introducing PyFlue: The Python-Native Agent Harness Framework.🧰

💡Flue for Python: Fred K. Schott @...

4. chhddavid (Group Score: 106.5 | Individual: 32.3)

Cluster: 4 tweets | Engagement: 41 (Avg: 348) | Type: Tech

we are so cooked 😭

these guys let Claude run wild on vibe coding

Look at this Duolingo clone it built in 183 seconds that:

> automatically works on both mobile + desktop > teaches you new words in French 🇫🇷 > has live exercises with grammar, spelling etc https://t.co/7nrOjKrUn4\n\nQT @shipper_now: BREAKING: Today, vibe coding has come to an end.

I just watched my Mac build a full business in 173 seconds.

This is just absudr. https://t.co/fcHOeBRkjw

See 3 related tweets

@chddaniel: IT'S SO OVER....

CLAUDE JUST BUILT A COMPANY IN THREE MINUTES

WHAT IS EVEN HAPPENING??\n\nQT @ship...

@chddaniel: so you're telling me Claude Code Opus 4.6 can now...
scan an entire website
build it as a mobil...
@Shipper_now: vibe coding is over?\n\nQT @shipper_now: BREAKING: Today, vibe coding has come to an end.

I just wa...

5. kimmonismus (Group Score: 103.9 | Individual: 31.2)

Cluster: 4 tweets | Engagement: 404 (Avg: 326) | Type: Tech

Google Omni model incoming. Probably being prepared for google i/o.

However i assume they will launch a new video model with it instead of Veo 3.1 since Seedance jumped to the top a few months ago\n\nQT @testingcatalog: GOOGLE I/O 🚨: A NEW OMNI MODEL IS BEING TESTED ON GEMINI FOR VIDEO GENERATION!

"Start with an idea or try a template. Powered by Omni."

This is a new leaked headline from the video generation tab on Gemini.

Omni appears close to "Toucan", an internal name of the current video generation tool powered by Veo.

If Google plans to release Gemini Omni for video generation, it would likely outperform Veo 3.1.

If true (as it is still highly speculative), Gemini will be the first top-tier Omni model with video output!

Google I/O 2026 will be hot 🔥

See 3 related tweets

@TeksEdge: According to @testingcatalog , @Google is testing a new “Omni” model for video generation inside Gem...
@kimmonismus: Rumors so far:
Google Gemini Flash 3.2/3.5 (already being tested)
New Omni Model, maybe even up...
@testingcatalog: RT @testingcatalog: GOOGLE I/O 🚨: A NEW OMNI MODEL IS BEING TESTED ON GEMINI FOR VIDEO GENERATION!

...

6. seraleev (Group Score: 88.9 | Individual: 34.3)

Cluster: 3 tweets | Engagement: 45 (Avg: 56) | Type: Tech

How do you validate an idea?

The answer is simple: ship it to the App Store (thanks Claude for speed). If money comes in and users don’t cancel their trials, the idea works.

From there, it’s straightforward: improve the product → add more value → scale traffic.\n\nQT @seraleev: How do I come up with ideas for launching apps?

I don’t use any special niche analysis tools. Just two filters.

First: my own problems. The best path. If I faced a problem myself, at least a few thousand people around the world have it too. This includes requests from my wife, family (kids too) and friends.

Second: niches I’m personally interested in. For example, I have a video rendering framework. With it, I can quickly ship apps for collages, stories, boomerangs, etc.

See 2 related tweets

@seraleev: How do I come up with ideas for launching apps?

I don’t use any special niche analysis tools. Just ...

@seraleev: An app where I save all my app ideas https://t.co/IMPmPv918t\n\nQT @seraleev: How do I come up with ...

7. kimmonismus (Group Score: 72.9 | Individual: 39.5)

Cluster: 2 tweets | Engagement: 656 (Avg: 326) | Type: Tech

What I still don’t understand is why Europe seems to have so little ambition to play any meaningful role in the future.

There is no convincing strategy to solve the energy problem. There is no serious push to build out European data centers so AI training and inference can actually happen on this continent. There is no clear plan to support the emergence of globally relevant European tech companies.

I genuinely don’t get it.

Yes, the European Commission is now trying to soften parts of the EU AI Act. But that seems to be almost the only meaningful concession being made to address what companies actually need.

I’m open to criticism and different perspectives here. But at this point, I honestly struggle to see how European policymakers intend to tackle the big structural problems ahead.

While China is building dozens of nuclear reactors, the United States is also investing heavily in nuclear energy, solar capacity is booming, and China is rapidly scaling energy storage, Europe’s approach still feels erratic, vague, and fundamentally unserious.

See 1 related tweets

@kimmonismus: It's unfortunate how many people seem determined to misunderstand my point.

My argument is not abou...

8. teortaxesTex (Group Score: 68.4 | Individual: 43.5)

Cluster: 2 tweets | Engagement: 399 (Avg: 73) | Type: Tech

"AI will create thousands of new jobs we can't even imagine!" thousands of new jobs that we can't even imagine: https://t.co/61QoIq1fqK

See 1 related tweets

@0xSero: Going off to the good ol 9-5\n\nQT @teortaxesTex: "AI will create thousands of new jobs we can't eve...

9. thdxr (Group Score: 64.6 | Individual: 33.3)

Cluster: 2 tweets | Engagement: 1724 (Avg: 496) | Type: Tech

every ai company that has raised a lot of money is probably building a github competitor rn

See 1 related tweets

@cgtwts: Sam Altman is everywhere https://t.co/vNsm9sYb8v\n\nQT @thdxr: every ai company that has raised a lo...

10. reach_vb (Group Score: 61.9 | Individual: 34.6)

Cluster: 2 tweets | Engagement: 492 (Avg: 167) | Type: Tech

Codex for everything.

Over the last 2 weeks, we shipped a pretty big set of updates:

GPT-5.5, browser control, Sheets & Slides, Docs & PDFs, OS-wide dictation, auto-review mode, /pets, and a .tex plugin.

The whole app also got better: dynamic UI for the task at hand, ~20% faster computer/browser use, stronger slides and sheets, annotation across browser/artifacts/code, easier onboarding, cleaner design, performance improvements, and fewer clunky handoffs!!

See 1 related tweets

@reach_vb: https://t.co/FjspZaD9Ks\n\nQT @reach_vb: Codex for everything.

Over the last 2 weeks, we shipped a ...

11. codewithimanshu (Group Score: 61.3 | Individual: 33.1)

Cluster: 2 tweets | Engagement: 388 (Avg: 236) | Type: Tech

RT @codewithimanshu: I Make $8,000-$ 12,000/week Doing Nothing.

Claude + a Cheap MacBook + OpenClaw + Polymarket.

It takes 15 Minutes to Set Up & Runs 24/7 While I Sleep.

I've made the exact step-by-step guide to connect Claude Code & Openclaw with Polymarket and start earning on autopilot.

Giving This Free for 24 hours.

To get it:

Comment "Money" & Like and Retweet This Post
Follow me @codewithimanshu ( then only, i can DM you)

The setup is embarrassingly simple:

Grab a cheap MacBook - Subscribe to Claude Pro
Connect it to Toobit + Polymarket in 15 minutes
Let AI trade spot and futures
Let AI bet on events while you sleep - Save $8,000-$ 12,000/month doing absolutely nothing

Note: You Must Follow me @codewithimanshu

, so i can send you DM.

That's it. That's the whole system for you to start making money online using ai & polymarket.

No coding. No experience. No daily monitoring.

AI trades. AI bets. AI compounds. You sleep.

Why isn't everyone doing this?

Because everyone is too busy:

Watching YouTube videos about making money. Saving Twitter posts about making money. Buying courses about making money. Talking about making money.

Instead of actually setting up the thing that makes money.

15 minutes of setup. That's the barrier. Not skill. Not money. Not knowledge.

15 minutes.

You spent longer today deciding what to eat for lunch. The MacBook on your desk right now could be earning $8,000-$ 12,000/week. Instead it runs Chrome with 47 tabs open and Spotify.

Same machine. Same capability. Different owner. One person set it up.

One person didn't. Guess which one is you. You Must Follow me

@codewithimanshu , so i can send you DM.

See 1 related tweets

@codewithimanshu: RT @codewithimanshu: CANCEL Your Weekend Plans, & Learn Claude Code Today.

This Claude Code teaches...

12. aakashgupta (Group Score: 61.1 | Individual: 35.0)

Cluster: 2 tweets | Engagement: 4 (Avg: 49) | Type: Tech

I've been telling people AI PM is the highest-leverage role in product. The comp data finally caught up to the thesis.

OpenAI: median PM total comp around $800K. The range runs from$ 300-400K up past $1M. Staff level PMs are routinely clearing seven figures.

Google: Senior PM comp at $500K median. Director and VP roles run into multiple millions. The equity is public, so you can verify any of this in 30 seconds on https://t.co/WuSXBPfjpl.

Anthropic: starts at half a million, ranges into high six figures. Pre-IPO equity at a valuation that keeps stepping up.

Meta: half a million for senior PMs, multiple millions at the top of the band.

This isn't speculation. The numbers come from actual offers verified against compensation data, not what someone says they "could be making" if their RSUs vested at peak.

Here's what's actually happening.

Three years ago, the top of the PM market was roughly $400K total comp at FAANG. Today's median for AI PMs at frontier labs is double that. The ceiling moved from "great PM career" to "founder-level outcomes without the founder risk."

The reason is leverage. An AI PM at OpenAI or Anthropic isn't shipping a feature for an existing surface. They're shaping a product layer that touches hundreds of millions of users in its first quarter. Companies are paying for the upside that one good decision compounds across that base.

Which is why the interview bar moved with the comp.

The mock with Ankit shows what the new bar looks like in practice. The "AI product sense" round evaluates whether you can reason about model behavior, eval design, and the failure modes that determine whether the product actually works. That's the round that's increasingly deciding who lands the offer at the median and who gets cut in the first loop.

The companies writing $800K offers aren't running standard PM loops anymore.

If you're a PM and you haven't seen what AI product sense actually looks like under interview conditions, the comp data above is the reason to fix that this week.\n\nQT @aakashgupta: AI PM interviews are now testing "AI product sense."

So I recorded a mock to demystify what it is with Ankit Virmani, who just nabbed AI PM offers at Uber, Atlassian, and Cisco.

6:59 3 tiers running it 12:04 Live mock 52:30 9/10 breakdown https://t.co/Oinn6sNmok

See 1 related tweets

@aakashgupta: Most candidates think "AI product sense" means knowing more about AI. That's the wrong target.

Afte...

13. petergyang (Group Score: 58.4 | Individual: 41.0)

Cluster: 2 tweets | Engagement: 434 (Avg: 104) | Type: Tech

How to keep your agents running even when your macbook lid is closed:

Go to Mac App Store and download Amphetamine
Go to Settings → Session Defaults → uncheck everything below

You're welcome :) https://t.co/mRVEq7c7Ly\n\nQT @cormachayden_: software engineers before vs after agents https://t.co/jJp75lO8O7

See 1 related tweets

@addyosmani: Software engineers before vs. after agents :)

If this is you: try Cloud agents (they're pretty good...

14. ShriramKMurthi (Group Score: 56.7 | Individual: 46.2)

Cluster: 2 tweets | Engagement: 3590 (Avg: 324) | Type: Tech

RT @samkhawase: What if India built GitHub? https://t.co/c3UG6oqyCg

See 1 related tweets

@starbuxman: RT @s13k_: What if Valve built GitHub? https://t.co/m2CRXZsCXs...

15. DataChaz (Group Score: 56.6 | Individual: 29.1)

Cluster: 2 tweets | Engagement: 62 (Avg: 271) | Type: Tech

"If you really want to make money... found an Agentic AI company."

Eric Schmidt

the market is still starved for builders.

but how do you actually hit $10k/month doing it?

@eng_khairallah1 just dropped the exact playbook to build and sell AI automations.

bookmark this 👇 https://t.co/9RGPvMo4Gl\n\nQT @eng_khairallah1: https://t.co/IgV4uj0iDx

See 1 related tweets

@aiedge_: The fastest way to make $10k/month with AI.

If I were looking to make my first thousands with AI, t...

16. chddaniel (Group Score: 53.9 | Individual: 25.3)

Cluster: 3 tweets | Engagement: 2 (Avg: 30) | Type: Tech

this is like watching 13 devs working on my codebase at the same time.

vibe coding is cooked fr https://t.co/ztCpAiNJi8\n\nQT @shipper_now: BREAKING: vibe code @telegram bots!

Shipper now builds complete Telegram bots from prompts https://t.co/7YEdI9FGJW

See 2 related tweets

@chhddavid: babe wake up.

Telegram bots are literally one Claude prompt away now.\n\nQT @shipper_now: BREAKING:...

@chddaniel: vibe coding is over?\n\nQT @shipper_now: BREAKING: vibe code @telegram bots!

Shipper now builds com...

17. KobeissiLetter (Group Score: 53.0 | Individual: 22.0)

Cluster: 3 tweets | Engagement: 2119 (Avg: 3413) | Type: Tech

Hedge funds are rushing to reduce tech exposure:

Hedge funds just posted their largest 2-week reduction in US information technology exposure over the last decade, excluding the meme stock frenzy in early 2021.

This was driven by long sales outpacing short covers at a ratio of 1.5 to 1.

Nearly every subsector saw exposure cuts, led by Semiconductors and Semi Equipment via long sales.

Furthermore, hedge funds sold Magnificent 7 stocks in 4 of the last 5 trading sessions.

Hedge funds are cashing-in massive profits in tech.

See 2 related tweets

@Mayhem4Markets: Hedge funds are doing that thing again.

They must hate tech.

Because they just sold it hard.

The ...

@FirstSquawk: GOLDMAN SACHS SAYS HEDGE FUNDS CUT TECH EXPOSURE AT SECOND FASTEST PACE IN A DECADE...

18. damianplayer (Group Score: 51.7 | Individual: 20.6)

Cluster: 3 tweets | Engagement: 60 (Avg: 50) | Type: Tech

Sam Altman in 2015:

“AI will MOST likely lead to the end of the world, but in the meantime, there’ll be great companies.”

today - he’s the CEO of OpenAI. https://t.co/7zHOZ7ygvk

See 2 related tweets

@WSJ: Pressure on the OpenAI CEO is mounting as the onetime AI front-runner prepares for a public offering...
@theinformation: RT @erinkwoo: at the time of this scoop from the information (october 2022, a month pre chatgpt) thi...

19. dair_ai (Group Score: 51.1 | Individual: 28.5)

Cluster: 2 tweets | Engagement: 29 (Avg: 106) | Type: Tech

We have released a little skill to help you build LLM Wikis with your agents.\n\nQT @omarsar0: Releasing a little skill to help build LLM Wikis.

More people should be building LLM wikis.

I hope the skill is useful for your agents.

Skill, examples, blog, and even a video tutorial included. Have fun. https://t.co/V3lIfjNLFu

See 1 related tweets

@omarsar0: RT @omarsar0: Releasing a little skill to help build LLM Wikis.

More people should be building LLM ...

20. scaling01 (Group Score: 49.9 | Individual: 28.2)

Cluster: 2 tweets | Engagement: 75 (Avg: 162) | Type: Tech

I one-shot myself with this post and I'm writing the second article today

This time about the value of optionality

I got really hung up on that "preserving optionality" phrase and how it connects to intelligence and many other topics\n\nQT @scaling01: I think returns to intelligence are nonlinear because decisions are path-dependent

early choices in code, experiments, or strategy can compound positively or negatively over time

for example by avoiding dead ends or preserving optionality

it's why I am a big fan of very long running tasks and massive benchmarking budgets

GPT-5.5 and Mythos Preview are only marginally more intelligent than previous models and have pretty much the same performance up to 10M tokens, but after that they go absolutely ballistic

See 1 related tweets

@garrytan: RT @scaling01: I think returns to intelligence are nonlinear because decisions are path-dependent

e...