Maku Akira

Exploring Science and Tech. Occasionally, I invest in stocks

[Maku Akira @Popinaplem]

@ShanghaoJin 我要confess… 我杠杆抄底了。🥲

[Maku Akira @Popinaplem]

@bboczeng 600%+ 谢谢。 😉

[Maku Akira @Popinaplem]

@ShanghaoJin @bboczeng 他思路没和你在一条线上，根本没看懂你说什么。 但这是大部分散户的思维，认定一个观点就往死里磕。很多人给自己设定的逻辑框架里是没什么问题的因为底层assumption或hypothesis就是错的或者biased。

[Maku Akira @Popinaplem]

@ShanghaoJin @bboczeng 他思路没和你在一条线上，根本没看懂你说什么。 但这是大部分散户的思维，认定一个观点就往死里磕。很多人给自己设定的逻辑框架里是没什么问题的因为底层assumption或hypothesis就是错的或者biased。

[Maku Akira @Popinaplem]

@bboczeng 你是没见过amd 2块钱长持到现在的人吗？ 哈哈 但你说的有部分是对的，对普通人来说定投或长持指数etf胜率是最高的。

[Maku Akira @Popinaplem]

Anthropic 是真缺算力

[Beth Kindig @Beth_Kindig]

Anthropic has signed a $10 billion deal with AI infrastructure startup Volta Infra, with the six year deal covering 133MW of $NVDA Vera Rubin capacity, with phase 1 targeted by year-end.

$GOOG $AVGO

[Maku Akira @Popinaplem]

@bboczeng 又应激了？ 😂

[Maku Akira @Popinaplem]

@ShanghaoJin 看了你和勃勃的comments。 验证了我之前说的，不是所有人都值得给信息。因为对信息解读会不一样。

[Maku Akira @Popinaplem]

很多人太高估自己的交易水平了。 赌对一两次就觉得自己是股神。 虽然想长期跑赢QQQ真不需要那么多有的没的。但你们真的能跑赢5-10年的年化收益吗？

[Maku Akira @Popinaplem]

@Liberty98043220 @RYANHINGSHING 确实是这样。

[Maku Akira @Popinaplem]

@ShanghaoJin 拉黑多好。省时省心。

[Maku Akira @Popinaplem]

@bboczeng 你自己交易决策不也是拍脑瓜。 还说别人赌狗😂

[Maku Akira @Popinaplem]

@leopoldasch Congrats!

[Maku Akira @Popinaplem]

exciting!

[AMD @AMD]

We're excited to announce our planned acquisition of @taalas_inc, bringing differentiated AI inference technology and world-class engineering expertise to AMD.

Together, we'll strengthen our AI roadmap and deliver faster, more efficient AI solutions across a growing range of inference workloads.

More on the news: https://t.co/2zA0wGL95P

[Maku Akira @Popinaplem]

@RYANHINGSHING @ShanghaoJin 才一次吗？ 那很少了。lol 我当年不知道做过多少次韭菜了。

[Maku Akira @Popinaplem]

@ShanghaoJin lol same. I sold all Chinese ADRs a few years ago when they wiped out the entire tutoring industry overnight.

[Jeff Pu @sssjeffpu]

Abstracts of Addressing Recent Debates on LT Capex and NVDA Platform Content Away from Memories

• Recent CSP Earnings Review on Capex: Amazon to $220bn, Google $195-205bn, Meta $130-145bn
• U.S. CSP capex forecast at $0.8tn/$1.2tn / $1.4tn in 2026/2027/28 (+45% / +20% YoY in 27/28), supported by healthy balance sheets
• Capex growth expected to moderate to +20% / +15% YoY in 2028/29 as CSPs prioritize returns & FCF & uncertainties over AI model competition
• Top 5 debt issuance seen at $257bn / $419bn in 2026/27 (31-35% of capex), still manageable
• Cumulative debt issuance could reach ~$421bn by 2028, yet remains well-supported
• Balance sheets stay very healthy (ex-Oracle): net leverage only ~0.5x (EBITDA basis) by 2028/29
• Net debt-to-equity ratio projected at just 18% by 2029 despite heavy capex

• Reiterate our July 2’s report of Nvidia cutting memory: Vera CPU SOCAMM from 192GB → 96GB → selective SKUs to 64GB by 1Q27
• Rubin Ultra HBM stepped down from 384GB to 192GB/256GB (HBM4/4E); 8-Hi design finalized, yet 12-Hi remains on line up
• Competitors already on lower configs: AMD MI450 Custom and Google TPU Humufish already 8-Hi
• Scale-up expanding via NVL576 w NPO optics to offset reduced memory content
• Nvidia platform OE shipments estimated at 5.8m / 8.1m units in 2027/28
• Still positive on AI trends; prefer GPU / CPU / Optics, cautious on memory commodities

Recommendations:
1) Nvidia GPU: NVDA (Buy), Hon Hai (2317 TT), SMCI (Buy)
2) CPU: INTC (Buy), AMD (Buy), ASE (3711 TT, ASX, Buy)
3) Optics NPO: LITE (Buy), MRVL (Buy), SMTC (Buy), TSEM (Buy), Bizlink (3665 TT Buy), Browave (3163 TT)

#capex #NVDA #RUBIN #Memory #HBM

[Jeff Pu @sssjeffpu]

Earnings Takeaways- AMD

• Raised 2026/27 EPS by +8%/+23%; TP lifted from $640 → $700 on 45x 2027E or 31x 2028E
• After-hours weakness as market digests high expectations despite solid results
• 2Q26 revenue $11.5bn (+50% YoY) in line; EPS $1.66 slightly above consensus
• 3Q26 guide: $13bn revenue above consensus, GM 56% soft
• Mgmt sees datacenter >100% YoY growth in 2027 (CPU + GPU)
• MI450/MI455X on track. Helios ramping in 3Q/4Q26 with further step-up in 1Q27
• We see Helios rack supply chain build plan ROBUST (>20k racks 2H26-27) 
• MI500 expected to outperform Nvidia’s Rubin Ultra in hardware performance

• Business tracking well ahead of long-term model (>35% revenue CAGR, EPS >$20)
• Strength from AI GPUs (MI450/Helios/MI500) + server CPU TAM/share gains
• Margin wise, AI GPU margins to stay below corporate avg due to large-memory strategy, while CPU and Embedded recovery are positive drivers

#AMD

[fin @fi56622380]

两个月前因为供应不足，NV rack DRAM配置削减，现在HBM上又发生了一次：客户要求Nvidia Rubin Ultra GPU 12层堆叠变成8层，造成HBM减配

我的观点很明确，如果真的让这个减配成为主流，那下一代产品token效率还不如上一代，谁还买啊

现在正处于明年HBM合约价谈判拉锯战时期，这个时机就很值得玩味了

这次的客户集体施压做减配版，根本原因可能有几个：

1. 供应确实短缺，12hi -> 8hi确实能让供应量变大

2. 给HBM三家供应商明年的长约价格谈判施压，如果做12hi -> 8hi型号，对三家的股价肯定有压力，属于谈判凭空造牌了，而且是明谋造牌

就算真的做出来了这样的减配版，顶多是某一种sku型号，12 hi的sku型号是不可能放弃的，而且仍然是主力

而8 hi这样的型号用在大模型推理端token效率是不划算的，顶多只能用在compute heavy而对hbm size不那么敏感的地方，比如prefill，不然的话，还不如买上一代产品rubin呢

每一代GPU单位电力的token throughput必须要指数型进步，最好是翻倍，才能大卖

所以rubin ultra在架构没有什么大改动的情况下，决定token throughput的HBM必须要在带宽或者size上指数增长

看看这次市场会不会还是像上次那样，把供应短缺解读成需求减少