Pulse Capital

Investor | Asymmetric AI infra & semis
Deep-dive theses + my real holdings

[Pulse Capital @pulsecapitalx]

I never introduced myself.

A few months ago I started writing investment theses.

No audience. No noise. Just research.

I came to X for one reason.

I’m not here to sell hopium.

I built a concentrated portfolio with a single target:

10x in the medium term.

Not 40 tiny positions hoping something works.

Not chasing whatever is trending this week.

Asymmetric bets.
Position sizing.
Timing.

What actually moves capital.

The same edge.

Here, in public.

I don’t chase stocks.

I hunt them.

— Pulse Capital

[Jonkoo Capital @JonkooTrades]

Everyone is laughing at $AAOI's stock price right now. The bears are celebrating, the shorts are getting louder, and suddenly everyone seems convinced that the entire investment thesis is falling apart.

It's funny how quickly sentiment changes when a stock starts going down. The same people who were excited about the company's growth prospects a few months ago are now questioning whether there's even a future here.

But I'm not investing in $AAOI based on what its stock price does this week, next week, or even next month. I'm investing based on what I believe this company could become over the next 12–18 months.

If $AAOI manages to scale toward $500M in monthly revenue next year, we're talking about a potential $6 BILLION annualized revenue run rate. That's an entirely different business from the one investors are looking at today.

And the opportunity isn't coming out of nowhere. AI infrastructure spending continues to accelerate, hyperscalers are investing billions into networking capacity, and the transition toward higher-speed optical connectivity is creating enormous opportunities for suppliers capable of scaling production.

Of course, execution matters. Manufacturing capacity needs to ramp, customer demand needs to translate into shipments, and margins need to improve alongside revenue. None of that is guaranteed.

But that's precisely why these opportunities exist. The market wants certainty, and by the time everyone agrees that the growth is real, a significant portion of the opportunity could already be priced in.

Maybe I'll look like an absolute idiot for being this bullish while everyone else is laughing.

But if my thesis plays out and $AAOI delivers anywhere near the numbers I believe it's capable of, I'll look like the GOAT.

And I'm perfectly comfortable being laughed at until then.

[Pulse Capital @pulsecapitalx]

Mid-2027 management framework is ~$5.65B annualized run-rate (not full-year revenue), vs ~$2.6-3B street consensus for FY27.

If they hit it, $68 looks cheap vs Meta Prometheus-scale optics demand — but it all hinges on execution of the 650k-unit ramp, yields, and no major dilution/competition issues.

[TopInvestor @Matrix_B0SS]

What if @aleabitoreddit posts again that he’s buying another 0.5% - 1% of the $SIVE float during the Halloween shopping season?

Could we see the May & June days come back and $SIVE back to $10 USD? 🚀

We’ll find out soon what Sivers management is cooking.

One thing is certain, the next surprise could shock the chart watchers once again with an unpredictable parabolic move.

$LITE $AXTI $LITE $COHR $SIVE $AMD

[Pulse Capital @pulsecapitalx]

Citadel just fully covered their short in $SIVE (0.55% → 0%).

This is not random. While the market is still panicking on AI names, the actual business is starting to deliver:

• Strategic LiDAR customer just placed first production orders. Shipments begin in Q4, volume production in 2027. $53-138 M lifetime potential is no longer just a design-win — it’s moving into real revenue.

• Yesterday’s EGM locked in a new management incentive plan and switched auditors to EY. Classic setup for a US dual listing in 2027.

Short interest remains high (~6%), the stock has corrected hard, and the photonics ramp (LiDAR + AI datacenters) is finally happening.

Smart money is covering into the inflection. The R/R is getting better, not worse. NFA

PS: long $SIVE

[TopInvestor @Matrix_B0SS]

@Citadel has exited $SIVE completely - from 0.55% to 0%.

What do they know that’s coming that could send $SIVE to the moon?

Could the November earnings call reveal:

New hyperscaler connections - $AMD / $NVDA?

New Chinese plug-and-play partners?

More major Photonics customer announcements?

Further progress toward a U.S. listing?

Everyone will find out soon…

[Pulse Capital @pulsecapitalx]

$IREN x $NVDA

This is the kind of thing people miss when they’re only looking at headline risk.

NVIDIA just dropped Minerva, an AIOps platform built for AI factories. It watches the GPUs, the network, the storage and the actual jobs running on them, then flags problems before they waste compute.

IREN is already using it.

Their CTO basically said the goal is to spend less time firefighting and more time keeping the GPUs productive. In a market that’s suddenly obsessed with utilization and whether all this AI spend is worth it, that matters.

It also fits the bigger picture. IREN already has the Microsoft contract, Exemplar status on GB300s, and the strategic NVIDIA partnership. Being early on Minerva just puts them deeper in the preferred stack.

Not the most explosive headline on its own, but a quiet signal that the operators who can actually maximize what they already have are the ones that should come out of this stronger.

PS: Still LONG and BULLISH on $IREN

NFA

[Shay Boloor @StockSavvyShay]

$IREN is partnering with $NVDA to launch its new Minerva AIOps platform on IREN AI Cloud helping catch problems before they disrupt GPU workloads.

That could help IREN generate more revenue from its existing GPUs without needing to spend more on hardware.

[Pulse Capital @pulsecapitalx]

$NBIS $SNDK $TSM

Hedge funds just had their worst day since COVID and retail is aggressively buying the dip in these three.

This isn’t the AI thesis breaking. It’s overcrowded funds being forced to deleverage hard.

Meta announced a new share class to fund more AI capex and the market read it as “this is getting expensive.” Result: forced selling across the sector.

Meanwhile:

• TSMC just reported another record quarter and says AI demand is still outstripping supply.

• Sandisk is in a real NAND shortage for data centers (not a phone cycle).

• Nebius is pure neocloud with hyperscaler deals.

When institutions sell on positioning and retail buys the actual bottlenecks, it’s usually an opportunity.

Are you buying or waiting for the war to ease?

[Serenity @aleabitoreddit]

Pretty crazy Reuters piece on the global selloff.

- October 22 marked the 1st time since the COVID crash that long/short, systematic, and multi-strategy hedge funds tracked by Goldman Sachs ALL lost more than 1% in a single day.

- Tech-focused hedge funds: down ~10% in October

- Asia-focused hedge funds: down ~9% in October

There’s separate interviews where Citadel/JP Morgan have been saying retail has been the ones buying the dip in semiconductor, neocloud, and memory stocks.

And the stocks bought by most retail were $NBIS, $SNDK, and $TSM.

Reuters stated this downturn triggered institutional deleveraging events among hedge funds.

Quite the divergence, I do think there’s been a ton of irrational selling now, hedge fund deleveraging seems to be a major contributor.

[Outlier Capital @outliercapx]

This post is going very viral. The #1 question I’m getting:

“What if competitors just flood the market with laser capacity and kill $SIVE and $AAOI?”

They can’t. Not in the timeframe that matters.

1. InP is the real constraint. Specialized epi, yields and process IP that don’t scale like silicon. Even $LITE and $COHR are still constrained.

2. Capacity takes years. $LITE already ramped 12x and is sold out through 2029, missing big chunks of demand. New capacity is 18–36+ months.

3. Qualification stickiness is extreme. Hyperscalers don’t switch laser suppliers casually once qualified.

4. The leader has no clear path. $LITE is fully booked, stopped taking orders, and lacks concrete expansion plans.

$AAOI is already ramping 1.6T + in-house lasers with demand exceeding supply through mid-2027.

$SIVE is the pure-play scaling to 100M+ CW DFB lasers/year.

Next-gen TPUs in 2028 are deeply optical. The overflow is real.

2027 can be a huge year for both. We are positioned.

NFA.

[Carrio @carrioresearch]

$NVDA $COHR $AVGO $LITE

直接回答你的问题：2027 年前完蛋不了。如果硬要说，真正的风险在 3.2T 那一代，时间点在 2028 年以后，要看的是到时候中际旭创和新易盛在不在新方案的供应商名单上。

最近走势很差劲，拿今天说：中际旭创 (SZSE: 300308) 收跌 8.8%；新易盛 (SZSE: 300502) 收跌 12.3%。

如果这是「CPO 要替代可插拔」的交易，那 CPO 的受益方应该涨才对。实际情况是同一天美股盘前，Coherent、Lumentum、Credo、Ciena 全都跌超 5%，康宁跌超 4%。这几家里有做 CPO 激光器的，有做光纤的，也有做铜缆的，一起跌，说明资金抛售的是整个 AI 硬件，不是某一条技术路线。

两家今年最大回撤已经到 48% 和 43%。从融资上看，拥挤的票在情绪转弱的时候先跌、跌得多，这很正常，我今天也没找到任何把下跌归到英伟达 CPO 新消息上的报道。但是有一点是事实：中国大陆股市在令人失望上从未令人失望😅

我们还是回到基本面，拆一下 CPO 到底替代谁。我提三点：

第一：CPO 替代的那段链路，今天用的是什么？

英伟达的 CPO 是分两步走的。第一步进交换机，Spectrum-X 的 CPO 版本已经量产，这部分确实会吃掉一些原本插在交换机上的可插拔模块，这是真替代。

第二步进 scale-up，也就是 GPU 之间的 NVLink。Rubin Ultra 会用 CPO 的 NVLink 交换托盘，把 576 颗 GPU 连成一个域。英伟达在本月 OCP 上说得很清楚：机柜里面继续用铜，机柜之间才换成 CPO。机柜之间这段今天走的是铜缆，换成光以后是新冒出来的光需求，它本来就不是光模块厂的订单，谈不上被抢。

可插拔也没有被英伟达放弃。每个 NVL72 计算托盘照样配 8 个 OSFP 可插拔口，Spectrum 6 另出一个全可插拔版本，英伟达在 OCP 上的讲者原话是「我们也在继续投资可插拔」。

第二：换代发生在哪一代速率？

光模块是按速率一代一代换的，替代只会在新一代发生，不会把正在出货的那一代砍掉。

现在出货的主力是 800G 和 1.6T，这两代还是可插拔的天下。本月 OCP 上，Meta 的讲者按每颗 GPU 约 6 只估算，2027 年数据中心要超过 1 亿只；Arista 创始人 Bechtolsheim 说交期已经从 3 个月拉长到 9 个月到一年，现在下的几百万只订单要到明年夏天才交货。英伟达还跟思科一起定了新的可插拔规格 USFP，首版 6.4T，说明年量产。一个准备替代可插拔的公司，不会再去牵头定一个新的可插拔标准。

替代真正发生的那一代是 3.2T。LightCounting 在 OCP 上预计 3.2T 端口里 CPO 和 NPO 合计约占 75%，3.2T 规模放量按摩根士丹利的判断在 2028 年以后。所以从现在到 2027 年，两家吃的是 800G 和 1.6T 的饭，这个碗 CPO 端不走。

第三：新方案的供应商名单里有没有它？

这一条才是需要注意的点，也是我觉得最该拿出来讲的。

英伟达发布 CPO 交换机的时候列过一份硅光生态名单，一共 11 家：台积电、波若威、Coherent、康宁、Fabrinet、富士康、Lumentum、SENKO、矽品、住友电工、天孚通信。A 股只有天孚通信 (SZSE: 300394) 一家。中际旭创和新易盛出现在同一篇新闻稿的另一句里，身份是「可插拔光模块」供应商。

博通 Tomahawk 6 的 CPO 伙伴页上，外置光源模块列了七家供应商，新易盛是其中之一，中际旭创没出现。

为什么名单这么重要？因为 CPO 把价值重新分了一遍。光引擎直接封装在交换芯片旁边，这一块落在台积电的 COUPE 封装和矽品这条线上。光模块厂在 CPO 里还能做的，是外置光源、光纤阵列、组装和测试，单价和毛利都比整只 1.6T 模块小一截。

所以两家的处境其实不一样。新易盛已经拿到博通 CPO 外置光源的位置，算是一只脚踩进去了。中际旭创押的是 NPO，也就是光引擎不封进芯片、放在旁边的近封装方案，公司 8 月电话会说 NPO、XPO、2.4T 在按客户要求开发送样，有望 2027 年下半年量产。NPO 有一个对它有利的地方：买方喜欢它。Meta 在 OCP 上说，从多家供应商的可插拔转到单一供应商的 CPO 是「巨大的风险」，NPO 可以保留多供应商，也能沿用光模块厂现有的制造体系。

说说看业绩这边，两家现在的底子很雄厚：

中际旭创上半年营收 417.78 亿元，同比增 182.5%；归母净利 136.51 亿元，同比增 241.7%，已经超过去年全年。二季度毛利率 46.40%，主要是 1.6T 出货多了。

新易盛上半年营收 209.10 亿元，同比增 100.3%；归母净利 75.29 亿元，同比增 91.0%。二季度单季 47.49 亿元，环比增 70.8%。

公司自己也在用真金白银表态。中际旭创 9 月花 49.97 亿元回购了 565.31 万股，成交价在 806.93 元到 944.68 元之间，比今天收盘高出约 37% 到 60%，这批股份 10 月 29 日上董事会审议全部注销。新易盛董事长提议回购 8 亿到 10 亿元。

接下来怎么看？

近的看三季报。重点看 1.6T 的放量有没有接上，毛利率能不能守住二季度的水平。这将比任何 CPO 新闻都更能说明这两家现在的生意到底怎样。

远的看 2027 年下半年。中际旭创的 NPO 送样能不能转成量产订单，新易盛的外置光源能不能跟着博通 CPO 一起放量。

我的看法是：CPO 是 2028 年以后的事，现在两家的股价反映的是情绪和仓位，跟生意本身关系不大。下面三种情况出现任何一种，我会改判断：

1、博通或英伟达下一代 CPO 的伙伴名单里还是没有中际旭创，而且到 2027 年下半年 NPO 仍拿不到量产订单。

2、超大厂把已经规划好的 1.6T 可插拔部署直接改成 CPO 或 NPO。这个可能性 OCP 上有人提过，一旦出现，替代时间就会从 3.2T 往前挪一代。

3、三季报里 1.6T 放量没接上，收入环比转负或者毛利率掉下来。

真正值得问的是，到 3.2T 那一代，中国光模块厂能在 CPO 和 NPO 里分到多大一块？

[Pulse Capital @pulsecapitalx]

@ivancop27932 Contracts de-risk demand and lock in volume. That’s already a very different setup from past cycles.

[Pulse Capital @pulsecapitalx]

Been thinking about this comparison to the late 90s a lot lately.

Serenity just dropped the cleanest breakdown:

Memory names under 6x fwd P/E with multi-year LTAs. $MU alone has 26 strategic agreements covering >35% of revenue through 2030, $32B in customer prepayments, and more than 75% of next year’s output already committed. Some talks even extending past 2031.

Optical compressing hard despite 500-1000%+ growth. $LITE is effectively sold out through early 2029 on key products, with a large portion of next year’s demand still unmet.

$NVDA , $AVGO and $TSM are printing real cash while hyperscalers are guiding $700B + in Capex this year alone, most of it locked into the supply chain.

1999 was 100-600x+ on NOTHING while 2026 is EARNINGS + CAPACITY locked in.

Which of these are you adding on weakness?

$NVDA $MU $LITE $COHR $AAOI $AVGO $TSM $AMD

[Serenity @aleabitoreddit]

AI names are backed by actual earnings and contractual demand.

Memory names are under <3x fwd P/E, with LTAs extending 5-10 years from your richest companies in the world like $GOOGL.

Optical names are compressing to <10x fwd P/E despite 500-1000%+ growth next year

Back in the day companies like Cisco/Yahoo were in 100x-600x+ earnings range.

[Outlier Capital @outliercapx]

As I said, everything comes down to oil.

Today made it obvious. $SPCX and $META just posted some of the strongest earnings in their history, AI exploding and turning profitable; Meta raising FY26 capex to $135-145B and staying ultra-aggressive on the buildout.

Both confirming the AI infra supercycle is accelerating.

And $QQQ is still -1.6%.

Until the US starts thinking for himself instead of being a puppet, we’re cooked. Higher oil keeps inflation sticky, rates elevated, and financing tight.

That’s not a broken AI trade. That’s oil and rates overriding everything else. The setup hasn’t changed. Only the price has.

Stay disciplined and take advantage of dips.

NFA.

[Dan 🤖🌥❤️ 🚀 @mcF_dan]

$IREN is about to double to $60. This is just how it trades. Anyone who has been following it for the last 3 years and longer knows this.

[Pulse Capital @pulsecapitalx]

@StockSavvyShay Prediction markets are pricing the silence itself. When Elon has nothing to say, the market starts writing the story for him 🤨

[Potentia Finance | The Insider @FinancePotentia]

$META JUST EXPLODED AFTER HOURS 🚀

Our members saw how the Whales moves BEFORE:
🟢 Trump
🟢 Funds
🟢 Analysts
🟢 Lobbies

$BE tomorrow. $MSFT, $GOOGL, $AMZN next week.

Next print is tomorrow. Don’t miss it.👇🏽

[Pulse Capital @pulsecapitalx]

$META $META $META

Everyone’s staring at the EPS miss.

They’re missing the real story.

Meta just printed $64.6B in revende, up 26% YoY and a clean beat. The $3.83 EPS (+265% YoY) only looks weak because of the one-time $9.9B AG charge. Take that out and it was roughly $7.40.

Ad pricing up 14% YoY, Family DAP at 3.64B, and Q4 guidance of $73-76 B. Muse and Muse Wonder are out too.

Stock is green after hours for a reason.

Who else is still long here? NFA

[The Earnings Correspondent @earnings_guy]

$META (Meta Platforms) #earnings are out:

[Joseph Invests @Joseph_Invests]

$IREN WILL MAKE MORE REVENUE IN 2028 THAN THEIR MARKET CAP TODAY AND TURN EPS POSITIVE
$SPCX just said there is a HUGE DEMAND FOR COMPUTE and they are pricing at 50 M/MW
This company is being valued like they are about to default but when you run the numbers:

contract pricing going up to 20-25 million per mw
1000 mw will likely be brought on during CY 2027 and 2028 in addition to the 300 mw brought on this year
Contracts with more hyper scalers and leading ai frontier labs like anthropic
almost the same amount of cash as debt
45-55% customer prepayments
ONLY A 2.4 PRICE TO BOOK RATIO vs competitors are 4.5 and 7 price to book ratio, the re rating will happen with execution of Horizon 2-4 delivery, "anthropic" deal announcement, and early 2027 deal announcements

[Michael Sikand @michaelsikand]

So you're telling me $SPCX.

- Will be my car
- Will be my company's workforce via Grokbot
- Will be my internet
- Will be my cell phone provider
- Will be my Uber driver
- Will be my chef and cleaner with Optimus

Yeah it's not priced in guys...

[Pulse Capital @pulsecapitalx]

$IREN $NVDA

Honestly this is one of the more BULLISH things IREN has done lately and the stock is still acting like nothing moves.

They just announced IREN Europe at NVIDIA GTC Berlin, put former Santander CEO Ángel Corcóstegui in charge of the board, and got the civil works license for the 300MW site in Badajoz. Power was already secured and they’re targeting first operations in 2028. It’s also set up for NVIDIA’s DSX platform.

That’s not another slide in a deck. That’s actual POWER and PERMITS in Europe, which is usually the hardest part. And with the Mirantis software layer they can actually handle the data residency and control stuff European customers care about.

The Real Madrid AI Cloud partnership a few days ago fits the same picture. They’re not just talking about Spain, they’re building there.

Market is still focused on financing noise and sector fear while the company keeps converting megawatts into real capacity.

[IREN @IREN_Ltd]

IREN is advancing sovereign AI in Europe, bringing together its European data center capacity and managed cloud services for customers seeking more choice over where their AI workloads run and how their data and infrastructure are managed.

Announced today at NVIDIA GTC Berlin, IREN has established IREN Europe and appointed Ángel Corcóstegui as Chair of the newly established board of directors.

IREN also secured the civil works license for its planned 300MW campus in Badajoz, Spain. Following the completion of its acquisition of Mirantis, IREN’s software capabilities include workload isolation and operational controls that can help customers address data residency, security and governance requirements.

Read more: https://t.co/46hRxoIDu0

[Pulse Capital @pulsecapitalx]

@ChartingU Risks are real (dilution, concentration, cash). Execution on the 800G/1.6T ramp is what matters now. We’ll see in a few months.

[Pulse Capital @pulsecapitalx]

BULLISH $AAOI at these levels.

At ~$77 (down >65% from highs) this is a clear buy on the extreme fundamental/valuation divergence.

Demand for 800G/1.6T remains supply-constrained through mid-2027. Capacity is ramping hard: ~200k units/month now → 650k by YE26 → >930k by YE27 (>50% US).

Q3 guide $255-290 M (vs $192M Q2) with 800G expected ~5x sequential. $200M + 1.6T order already locked in.
SMCI just raised guidance massively to $18.5-19.5 B (from $14.5-15.5 B) — the procurement acceleration is real and visible.

You’re looking at a ~$6.5B company on a path to well over $1B this year with a much bigger ramp coming. The market is pricing in an AI capex collapse that the data doesn’t support yet.

Earnings Nov 5. High conviction setup.

What am I missing?

PS: very long $AAOI . NFA

[Serenity @aleabitoreddit]

Divergence between fundamental demand and valuations are now at extreme levels.

You're seeing accelerating procurement from $SMCI to $SPCX vs. collapsing equity prices on $AAOI to colo players.

After ERs, I think Oct 2026 will be seen as a DeepSeek / April tariff opportunity

[Carrio @carrioresearch]

$ANET $LITE $COHR $AAOI $AXTI

ARISTA'S ANDY BECHTOLSHEIM JUST PUT THE NEXT FULL OPTICS UPGRADE ON THE CALENDAR: TWICE THE SWITCH BANDWIDTH, SAME LANE COUNT, EVERY LANE MOVING UP TO 400G

Last month in Riga he said the industry can't build enough lasers for this year's modules. Now he's asking every one of those links to run twice as fast.

His Oct. 20 slides at PECC go from 204.8 Tbps on 1,024 lanes at 200G to a potential 409.6 Tbps on 1,024 lanes at 400G, with first silicon projected for 2028 and volume ramp in 2029, according to Converge Digest's report from the room.

He also lined up Arista's ($ANET) own XPO module, 25.6 Tbps with a 500 W design thermal envelope, against the NVIDIA and Cisco USFP at 6.4 Tbps and 100 W. In his measurements both connectors look like a credible path to 400G PAM4. XPO is Arista's design, so that part is his case for his own module.

The Riga numbers: OSFP demand this year of about 150 million units against just over 120 million shipping, because module makers can't get enough lasers. Lumentum ($LITE), Coherent ($COHR) and Applied Optoelectronics ($AAOI) make the indium phosphide lasers that go into these modules, and AXT ($AXTI) sells the InP substrates they're grown on. The roadmap he showed keeps the lane count at 1,024 and doubles what each lane carries, so I'd expect the laser makers to stay at the center of this through the 400G generation. Their 400G lasers should be ready in volume by the time that switch silicon shows up in 2028.