Worth the read #SMH #AI " how fast do the hyperscalers’ earnings need to grow to justify —expenditures will reach nearly $1.1 trillion."?
technologyreview.com/2026/09…
Eem retweeted
Allow me to interpret what’s happening.
Anthropic is being audited.
Anthropic desires to file an S-1, as they would like to go public. Therefore they need an audit. And by “they”, I mean the VC’s who invested in them. So “they” can exit their position and pass the bag to firemen, nurses, teachers and policemen.
How does this go from the VC’s to the working man and woman? Because the size of the IPO will automatically qualify Anthropic for the Fortune 500 and the Dow Jones 100. Therefore, every working person with a 401k or pension will end up owning a little bit of Anthropic in their mutual funds. Teachers hold the bag, VC’s take the cash. Thank you, come again.
Now back to the audit. The audit required is a PCAOB audit, Public Company Accounting Oversight Board. This audit is what all public companies must comply with be on the stock market. Revenue recognition, expense classification, depreciation, related party transactions, etc. It’s there for consumer protection.
This audit is TOUGH. It is INVASIVE. There is no way to lie your way through it. Any company that passes a PCAOB audit automatically earns my trust on finances.
How do I know? Because I’ve been through it before. @ChangRobotics is 2 year PCAOB audited and currently underway for a 3 year audit. It’s brutal. The same as showing up as the valedictorian to your high school graduation, except you’re naked, and you have to walk on stage and deliver the speech. It’s rough. And I know many incredible founders that can’t pass one.
Now, why would Anthropic be leaking all kind of weird statements lately about “self pacing” a slow down on AI (e.g. they are WAY behind on revenue), and profitable if they didn’t have expenses (e.g. we just learned for the first time what our expenses are, because we’re being audited).
Because they were claiming NVIDIA discounts and Microsoft cloud credits as revenue. Because they had no clue what their expenses were, or why it even mattered. Because they had unlimited investor capital and their job was to burn it to make an LLM. Well, they did a great job with that!
That’s the same as my wife coming home with Bed Bath and Beyond coupons and telling me it’s her paycheck. Ummm, not the same, sweetheart.
So by now hopefully you can see that Anthropic is in a PCAOB audit right now, in order to file an S-1 and go public, and pass the bag to teachers so the VC’s get profits. And hopefully that explains their “crazy” behavior.
In reality you can be grateful to KPMG, PWC, or whoever is auditing Anthropic, because it’s the first time Dario learned that:
1) we are not profitable
2) expenses matter
3) coupons are not revenue
4) we have no clue how to be “profitable”
5) growth is hard when revenue numbers are in an audit and not a power point
-your neighborhood engineer
Oooops... how is the credit markets going to respond to this? 😱😱
China is making Dario Amodei's AI slowdown proposal worthless.
The thing doing it is a 510 GB file that anyone can download for free.
Two days before that essay went out, DeepSeek shipped a model called V4.1 Flash.
The weights went straight onto Hugging Face under an MIT license, which allows commercial use, modification and redistribution with no royalties owed to anybody.
On its own model card, V4.1 Flash beats OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5 on four of the five hardest agentic benchmarks:
- DeepSWE v1.1: 74.2 against Sol's 73.0
- AutomationBench: 54.8 against 45.8
- Agent's Last Exam: 31.8 against 26.7
- CyberGym: 88.1 against 84.5
CyberGym is the cybersecurity one.
So the exact capability Amodei is asking the industry to pace is now a PUBLIC DOWNLOAD LINK.
Now look at the price...
OpenAI listed GPT-6 Astra on September 3 at $10 per million input tokens and $50 per million output. DeepSeek listed V4.1 Flash seven days later at 15 cents and 60 cents off-peak. Cached input runs at a third of a cent.
And the reasoning benchmarks tell the same story over a longer window. Scoring 87.5% on ARC-AGI-1 cost roughly $4,560 per task in December 2024. 20 months later the same score cost about 30 cents. DeepSeek's earlier Flash build scores higher than that, 89.0%, for 2 cents.
On OpenDesign's arena on September 9, DeepSeek landed 1.5 points behind OpenAI's newest model at $0.023 per task against $1.61.
1.5 points.
At 70 times the price.
That's the American "lead“, measured this month.
And that number is what breaks the entire proposal. The essay caps the permitted slowdown at the size of the lead, because slowing by more than that lets Chinese projects pull ahead. So the slowdown Anthropic, OpenAI, Google and xAI are allowed to take is 1.5 points wide.
V4.1 Flash carries 552 billion parameters but switches on only 8 billion of them for each word it reads, which is why it runs for almost nothing. It trained on 45 trillion tokens. It's sitting on Hugging Face right now.
A file on a hard drive signs nothing.
Nobody can UNDOWNLOAD it.
And the distillation crackdown in Dario‘s plan also does nothing here either, because nothing was distilled. DeepSeek published the weights outright. There's no theft to prosecute and no copy to trace. They gave it away on purpose.
DeepSeek's Pro model already hit 80.6% on SWE-bench Verified back in April, at roughly a thirty-fourth of the price of the American flagships. This has been happening all year.
The frontier still wins the hardest work. On Terminal-Bench 4.0, OpenAI reports 57.9 for its newest model against the 31.2 DeepSeek reports for this one. Those four benchmark wins come from DeepSeek's own card, and nobody has replicated all of them under one shared protocol. If you look at it that way, the American labs still hold the dangerous end of the curve, and pacing that end is worth doing.
But a speed limit only binds companies that can be sued in an American court.
Everyone else just downloads.
What exactly is Dario Amodei's slowdown protecting you from?
Eem retweeted
Why I never bought $AAOI after putting ~100 hours of research into the company back in March 2026:
1 - $AAOI’s very questionable execution track record: during the 100G cycle, management made aggressive claims, insiders sold stock, and the company ultimately failed to deliver.
That same leadership team is in place today. In a report from Culper in 2025, multiple industry contacts told them that $AAOI's reputation has been "tarnished" and that $NVDA considers working with them "a non‑starter", meaning they won't even consider starting a conversation with them. Yikes.
2 - A terrible company culture: the picture that emerges when looking into the employee reviews from Glassdoor and Indeed is plain bad. A few quotes (and these go on and on):
“Senior management is dysfunctional and decisions are often driven by personal relationships rather than what makes the most sense for the business. Who you know and how well you’re liked tends to matter more than performance.”
I researched them when the stock traded at $50 in March and decided it’s not a company that fits my criteria to join my portfolio.
I do not trust $AAOI management with my money, and they way they handle their latest $600M ATM adds to that belief.
With $AAOI trading at ATHs, it's easy to jump aboard the hype train.
HOWEVER.
Every investor should be aware that $AAOI's culture is in pretty bad shape, based on multiple sources. A quote from a former employee:
"Senior management is dysfunctional and decisions are often driven by personal relationships rather than what makes the most sense for the business."
And this is just one example.
If you hop over to Glassdoor, the review section is filled with comments like this. Given $AAOI's prior execution missteps, this is not a good look and definitely something every $AAOI investor should be aware of.
Eem retweeted
$AAOI is one of the clearest examples of "fake it until you make it"
Over the last year, AAOI has issued $1.4B worth of shares. Including the latest authorization, that figure would rise to $2B
A year ago its market cap was just $1.3B
That means AAOI has raised more through equity issuance in one year than the entire company was worth
How could they do it?
They gave analysts some insane growth targets that could only realistically be fulfilled if the market cap increased enough to make massive ATM financing possible
In summary, they bet that the market would buy into those targets, allowing them to raise enough capital to actually grow into the figures they promised
Extremely aggressive growth targets → stock rerates → AAOI uses the higher share price to raise huge amounts through ATMs → that cash funds the manufacturing capacity required to have any chance of reaching those targets
It’s almost like manifesting your own destiny
Fair play to them. That’s how you use the public markets
Now they have the chance to make all that dilution accretive by actually reaching their targets
Eem retweeted
Breaking: US stock market is most overvalued in over 100 Years
A basket of valuation metrics indicates that the stock market is more overvalued than at the 2000 dot-com bubble and even 1929 market peak.
Eem retweeted
这次美国访华晚宴坐在马斯克和库克中间C位最亮眼的，是蓝思科技创始人周群飞，从农村打工妹到中国女首富，完全没有任何背景全靠自己白手起家。她出生在湖南的一个小村庄，5岁时母亲去世，父亲也因工伤残疾双目失明，家徒四壁，16岁交不起学费被迫辍学去广东打工，在一家工厂流水线打磨玻璃，白天干活，晚上疯狂自学，考了会计证、电脑操作证等技能证书，就这样过了几年，她拿着靠打工攒的2万块，拉上哥哥姐姐、嫂子姐夫等8个亲戚，在深圳开了一家小作坊，做手表玻璃加工，她一个人修机器、跑销售，就这样又干了4年。
到了2000年后手机行业开始大规模发展，一次偶然的机会她的手表玻璃厂接到了TCL手机屏幕的订单，她看到了手机玻璃市场的巨大潜力，于是迅速成立了蓝思科技，专门负责手机玻璃的生产研发和销售，一开始只做国产手机和山寨机，但是直到有一次想谈下摩托罗拉的订单，但外企的质量要求非常严格，她赌上几乎全部的资源配合摩托罗拉，拿下了全球销售超过1亿台的V3订单，直接把蓝思科技推到了行业领先位置，随后顺利拿下诺基亚、三星等外企。
关键的转折点又一次出现在了2007年，当时乔布斯发布初代iPhone，彻底把手机往全玻璃触屏的方向变革，乔布斯那变态的工艺要求全球都找不到符合的厂商，周群飞敏锐的意识到这又是一个巨大的机会，于是带领团队和苹果工程师联合攻关了3个月，突破关键工艺，成功量产了第一代iPhone玻璃面板，从此拿下苹果长期合同，后续的iPad、MacBook等几乎所有苹果设备全部都交给了蓝思科技，也帮助蓝思科技在触摸玻璃面板领域成为全区最大公司。
这也是为什么她能坐在库克旁边，那为什么马斯克也坐在她旁边呢？
蓝思科技在玻璃面板干到全球第一后，开始往更加多元的智能化设备发展，包括汽车座舱和机器人，其中汽车领域包括车窗、中控等已经拿下了特斯拉、宝马、奔驰、理想等30家车企，机器人领域则主要负责关节、传感器等部件，这些都和马斯克的业务有深度重合。
一个15岁辍学只有初中文凭的女生，从湖南农村出来白手起家成为中国女首富，40年后进入中美会谈，坐在了马斯克和库和中间，这就是周群飞的故事。
Eem retweeted
#NonfarmPayroll
Let me be clear!
No Economy goes from +160k to -133k the month after to +178k the next month.
This is complete nonsense!
My thesis is that BLS frontload the numbers to satisfy politicians.
As nobody is observant on the revisions - my thesis is that BLS make every second month strong (frontloading) - then next month bad - and then they let the backward revisions the coming months take care of the adjustments lower.
And - in 2025 - the job creation was 180k.
It does not jump to a “178k job creation economy per month” in few months.
I normally never say this… but these are manipulated numbers!
And it will show in the yearly correction - just like it did in 2025.
But who cares 12 months later….
🤣🤣👏👏🤣🤣
An old joke, newly relevant:
A Finn and a Russian are fishing on opposite sides of a river along the border between their countries. The Finn pulls fish after fish out of the water, to the Russian's growing frustration. Finally, the Russian can't bear it any more and bursts out: "How come you're getting all the fish?" The Finn replies calmly: "It's because the fish on your side are too afraid to open their mouths."
