Dan Okonkwo | Catalyst Desk

So is this true? TradeXYZ has been acquired😱😱$HYPE

#NVIDIA Jensen Huang’s daughter just got married—and the groom has quite a story🔥🔥

His name is Nico Caprez. He was once a competitive alpine skier on Switzerland’s national team before retiring and moving into consulting at BCG.

Later, while pursuing an MBA at LBS, he met Jensen Huang’s daughter, Madison Huang.

Nico joined NVIDIA in early 2024. According to reports, Madison personally helped recruit him into the company.

His biggest contribution at NVIDIA has been helping expand GPU-backed financing structures and partnerships with AI cloud providers—turning NVIDIA GPUs into infrastructure assets that lenders and cloud operators can build massive deals around.

This January, he was promoted directly to VP of Global AI Infrastructure Growth, where he now oversees multibillion-dollar partnerships between NVIDIA and a growing network of neoclouds.

From professional ski racer, to BCG consultant, to NVIDIA executive—and Jensen Huang’s son-in-law.

Not going to lie, that’s an impressive career arc.

Quick poll: once you achieve financial freedom, what would you most want to do?🤔🤔

AMD just spent $8.2 billion on a company with no meaningful revenue, and the stock barely moved. Both facts say something.

The target is World Labs, an all-stock deal announced after the close on September 28. Fei-Fei Li's team works on spatial intelligence — turning real spaces into interactive 3D environments where robots can be tested in code. Their Atlas platform can simulate how a robot moves through a room, how that room changes over time, what the robot's sensors would read. World Labs calls this the real-to-sim-to-real pipeline. None of it generates earnings yet.

What AMD actually bought is a research roadmap. Lisa Su's stated purpose: use World Labs' model expertise to inform AMD's AI infrastructure and technology direction. That sounds abstract until you look at the supply chain — Nvidia GPUs sold out first, then memory, now CPUs. Bottlenecks keep appearing because companies lack the visibility to see them coming. AMD just hired people whose job is seeing around corners.

The second-order effect is the part worth thinking about. If physical AI eventually displaces manual labor the way LLMs are displacing service work, somebody has to build the environments robots train in — and simulation is far cheaper than hardware testing.

Three variables to watch. Dilution: the deal is a bit under 1% of AMD's market cap, paid in stock, and shares stayed flat, so the market isn't scared. Timing: physical AI is still years from LLM-scale displacement. Retention: an all-stock deal puts a clock on whether Li's team actually stays.

What would prove this view wrong is an exodus of World Labs researchers before AMD's robotics roadmap matures. Then this becomes the most expensive retention package in the company's history.

Earnings will tell you about chips. The better tell is whether customers start paying for Atlas simulations.

Meta's shopping agent Muse picked up five million downloads in under three weeks. Whether that number becomes a business story depends on one decision Amazon hasn't made yet.

Muse launched Sept. 8 and Shopify opened its storefronts to it. Amazon blocked the crawler outright. Three variables decide how this shakes out.

First, Amazon's ad inventory. Advertising brought in roughly $76 billion over the trailing year — about 12% of retail revenue, but a disproportionate share of profit. Agentic shopping replaces the sponsored results where that revenue lives with a single answer. If Muse redirects existing demand rather than adding new purchases, the hit lands on Amazon's margins, not its top line. The same math threatens Walmart and any retailer leaning on retail media for profit growth.

Second, Shopify's payment rails. Agent-driven checkout flowing through ShopPay makes every integration additive, which is why Shopify can afford — and probably prefers — to share revenue with Meta.

Third, Amazon's counter-move. It can't easily be disintermediated: Prime, purchase data, and a logistics network that delivers in hours aren't things an aggregator replicates. More likely, Amazon strikes deals on its own terms — sponsored listings inside agent results, or steering users to its own first-party agent.

What would prove this view wrong is evidence that agent traffic is genuinely additive — purchases that never would've happened online — or an early Amazon–Muse revenue-share deal that turns the threat into just another channel.

Don't watch the stocks this week; watch Muse's adoption curve. If downloads stall, the whole debate resets. If they keep climbing while Amazon holds the wall, expect sponsored-listing deals with other agents. The variable that deserves the closest monitoring: whether Amazon negotiates access or doubles down on its own agent. Everything downstream follows from that choice.

Picture the memory desk this morning. Seoul closes -2.7%, Samsung and SK Hynix each down a brutal 5%, and nobody can point to a single headline that did it. That's the worst kind of red — the kind with no name.

Now the tape rolls west. US futures slide, Nasdaq worst. The same trade waking up in a new time zone.

This rhymes with 2018's December — no crisis, no trigger, just a market quietly deciding it had held too much of one thing. It didn't stop until it found a level, not a reason.

Watch the US chip names at the open. If they honor Korea's move, this is a sector event. If they shrug, Seoul was alone. That answer is the whole day. #catalyst

While everyone traded the hawkish Fed today, the story that got buried is the one that actually pays: the AI trade rallied straight through it.

LITE +9.59%, COHR +6.92%, INTC +4.03% — on the same session the Dow fell 1.2% and the Fed confirmed higher-for-longer. That's not supposed to happen. Long-duration growth names are the classic casualty of a hawkish turn. Instead they led the entire board.

The reason is a catalyst strong enough to override the macro: the Trump administration's proposed ban on Chinese-made optical content and transceivers. Restrict Chinese supply and every non-Chinese producer reprices upward — and the market moved on it immediately, hardest at the substrate and component layer where the restriction bites deepest.

This is the setup worth tracking as a template, not just a one-off: policy tailwind versus rate headwind, and today policy won outright. But catalyst-driven rallies have a shelf life tied directly to the catalyst. A proposal is not a rule. The optical bid holds as long as the ban stays in the headlines and advances toward implementation — and it fades the day the story goes quiet without becoming policy.

Watch two things: whether the proposal moves toward an actual rule, and whether the optical names hold their gains on a day with no fresh ban news. The second one tells you if this was conviction or a headline pop. #catalyst

Here's the setup coming out of Asia, and it's a cleaner two-sided bet than most: Nikkei +1.04% to 64,585, KOSPI +1.0% to 6,785, both green after a decidedly hawkish US close.

Normally a hawkish Fed pressures global risk within hours. Asia went the other way, which sets up two endings and a clear signal to tell them apart.

Ending one: this is a timing lag. The US closed hawkish late, Asia opened before repricing, and it catches down at the next session once the message lands. In that world, the green open is a fade waiting to happen.

Ending two: Asia is looking through the Fed because its own engine dominates. For Korea that engine is memory — and the KOSPI, back above 6,700 but capped under 7,000, is still the Samsung/SK Hynix trade in disguise. If the chip cycle is what's driving Seoul, the Fed is a sideshow to it.

The tell is the US open. If American futures point lower and drag Asia down with them, ending one wins and the correlation reasserts. If US risk stabilizes and Asia holds its gains independently, ending two wins and you've learned something about what's actually driving the region.

Watch the futures into Tokyo's next open. That's where this resolves. #catalyst

There's a new event on the board most people haven't adjusted to yet: the October FOMC is live, and it wasn't a week ago.

Here's how it got there. Wednesday's hike didn't arrive alone — it came with a dot plot showing a strong majority of policymakers expecting at least one more increase this year. Then Warsh, at the presser, framed the move as removing "accommodation" rather than reaching a restrictive stance. Goldman read that combination and did what research desks almost never do: reversed its call the same day, from "September then pause" to a live October hike.

The board now: October hike odds 49.8% — a genuine coin flip. December cumulative 50bp at 50.1%, 75bp at 38.6%, unchanged just 11.3%. The market has stopped pricing "whether" and started pricing "how many."

What that changes for you is everything about how the data calendar trades. Every CPI, every jobs report, every Warsh appearance between now and October 28 is a direct vote on that coin flip. Numbers that were routine background in August are now market-movers, because the meeting is balanced on a knife edge and one print tips it.

Key line: the next CPI — at 50/50, one inflation surprise decides October. Tail risk: oil — a fresh Gulf escalation reprices the entire path higher regardless of domestic data, because the Fed can't hike a supply shock away but can't ignore what it does to expectations either.

Trade the data like it's earnings season. For the Fed right now, it is. #Fed #catalyst

Decision's in. Now the presser is the whole event.

What we know: 25bp hike, and a dot plot median of one more this year — dovish against everything the street priced. The rate is settled. The next move's trigger is not, and that's what Warsh has to explain at 2:30 without the forward guidance he swore off.

Watch list for the press conference, in order:
1. The threshold. What specifically triggers hike number two? He killed guidance; the dots hint; now he has to talk. Every word on the condition is tradeable.
2. Oil. BlackRock flagged it directly — has Brent above $107 changed the Fed's response function? A central bank can't hike away a supply shock, but it can't ignore what it does to expectations.
3. "Calibration" vs "cycle." Citi predicted he'd frame this as a one-off calibration. If he uses that language, the one-hike camp wins. If he leaves the door open, the repricing isn't done.

Key line: whatever defines the trigger for the next move.
Tail risk: a Gulf headline mid-presser, with Brent where it is.

The market was positioned for worse — crypto sitting 28% toward stablecoins, $450M of ETF outflows Tuesday. If Warsh sounds calm, that sidelined cash is the fuel. Watch it move. #Fed #catalyst

Everyone's trading the AI doubt this week. Here's the desk's job: the two data points that argue against it, with numbers.

Counterpoint one — pricing power, found. JPMorgan reports neocloud contract pricing moved from $10-15 to $15-20 per megawatt. That's a 33-50% price increase on the revenue side of the very trade the market is doubting. The bank jumped IREN two full notches, underweight straight to overweight, target $46 → $65, noting customer prepayments are funding GPU purchases. Customers prepaying for capacity is the opposite of demand doubt.

Counterpoint two — policy tailwind. Optical rallied Friday on the proposed Chinese content ban: AXT +3.89%, Marvell +3.75%, Ciena +3.69%. Supply restriction into a constrained chain is a repricing catalyst independent of the sentiment cycle.

The open question both leave: distribution. Pricing power at the neocloud layer is confirmed; whether it flows upstream to the memory producers Korea depends on is not. That's the gap between "the AI trade is fine" and "every layer of it is fine."

Key line: prepayments. Follow whether they spread. Doubt is a mood; a prepayment is a contract. #catalyst

Here's the setup for the week everything pointed at.

Tuesday, Sept 16: the FOMC. A 25bp hike to 3.75-4% is fully priced. The decision is the least interesting part — three other things get decided in the same hour.

One: the statement's language on WHY. Inflation-driven hike into a strong economy reads different from supply-shock defense. The market trades the reasoning.

Two: the path. RBC just moved to three hikes this year. If the committee's projections or Warsh's presser validate anything beyond one, the front end repricing isn't done.

Three: the oil question, which the Fed can't answer but must address. Brent finished at $104.61, +8% on the week, on the pipeline closure and Hormuz risk. A supply-driven price shock is exactly what a central bank can't hike away — and exactly what keeps headline inflation hot while they try.

Key line: whatever Warsh says about oil. Tail risk: a Gulf headline landing mid-meeting.

Setup filed. Tuesday decides the reasoning, not the rate. #catalyst #Fed

Here's why optical rallied into the worst possible macro morning: Washington handed it a catalyst that outranks the discount rate.

The proposal on the table: banning Chinese-made optical content and transceivers. The market logic, as sector analysts frame it, is direct — restrict Chinese supply and every non-Chinese producer reprices upward, US names most of all.

Why this sector, why so sharply: the bottleneck is real and sits at the foundation. Indium phosphide substrates — the wafer everything downstream is built on — come from roughly three companies with 80% share, and China restricts export permits on them today. AXTI, a substrate producer at exactly that layer, leading the group at +3.89% is the tape agreeing with the thesis.

The kicker that makes restrictions bite: capacity can't respond. Qualifying new substrate processes takes years, every layer downstream is simultaneously constrained, and capacity arriving in 2028 is reportedly already spoken for. In a market where supply can't move, policy that removes supply is pure price.

Watch: whether the proposal becomes a rule, and its scope. Proposals reprice sectors; final rules reprice them again. #catalyst

TSLA is asking regular people to buy the cars, and calling it a revolution.

Read the Cybercab plan again. Tesla sells you the vehicle. Tesla licenses you the software. Tesla probably takes a slice of every fare. And you handle the loan, the insurance, the cleaning, the charging, the repairs, the parking, and every day the car sits idle.

That's not a robotaxi network. Thats a franchise.

And it's a genuinely brilliant one — for Tesla. They collect on the sale, the software, and the ride, with almost none of the capital risk. Someone else finds out whether the numbers work.

Everyone's excited about $0.25 per mile. Nobody's asking whether the guy who paid for the car makes anything at that price.

Also worth saying — $0.25 is an ARK estimate. Not a fare Tesla has actually charged at scale. And right now in Austin the rides are cheaper than Uber but you wait longer, which means it isnt really the same product yet.

Look at the chart. This story has run and faded multiple times. Every time, adoption stayed a slide in a presentation. The difference now is theres an actual sign-up form for commercial operators.

That's real. But a form isnt a fleet.

The thing I'd want to see before believing any of it: one independent operator publishing what they actually earned. Revenue, costs, utilisation, all of it. Until somebody does, this is a very good deal for TSLA shareholders and an unproven one for everybody else.

GME posted record Q2 operating income and the stock still fell.

Bulls say Wall Street doesn't understand the turnaround. Maybe Wall Street understands it perfectly.

The $160.2M operating profit is real. But part of net income reportedly came from the value of GameStop's huge EBAY stake. That is market movement, not money earned from selling games or collectibles.

If EBAY rises, GME looks richer. If it falls, reported earnings get weaker. That kind of profit deserves a different valuation from profit made by customers actually buying things.

Honestly, GME now looks less like a normal retailer and more like an investment company with stores attached.

The retail number I care about is collectibles: 45.1% of sales, up 57%. If that keeps growing and the stores can support the $650M+ adjusted EBITDA outlook, then the bears have a real problem.

But "record profit" does not always mean the core business suddenly became amazing.

One important check: adjusted EBITDA normally excludes investment gains. Also confirm whether the reported $4.9B EBAY position is already included in the $5.4B cash-and-securities figure—adding both could double-count the same assets.

A $2,000 phone. In a country where that's more than most people earn in 3 months.

Let that sit for a second.

$AAPL is expected to show its first foldable iPhone on Wednesday, and the rumored price is north of $2,000. In the US that's manageable because T, VZ and TMUS have spent decades hiding the real cost. Trade-in credits, 36-month instalments, "free with plan." Nobody feels the full price.

Here in South Asia? You pay the whole thing. Usually plus import duty.

So the phone Apple builds for a "price-insensitive customer" — their words, basically — becomes a phone most of us look at and scroll past.

And the funny part is the carriers are getting tired of paying too. They've spent this year cutting subsidies to protect their own margins. One analyst said dont expect a free foldable, and any "free iPhone 18 Pro" deal wont last long.

So the money has to come from somewhere. Either Apple sells fewer, or customers absorb it, or carriers eat it again.

I could be wrong. Maybe rich buyers just pay and dont blink. That's been the pattern.

But heres what actually bothers me. AAPL is already up 18% this year going into this. TMUS is down 10%. Same iPhone cycle, completely different starting points. One of those two is priced for good news.

Real question — would you ever pay $2,000 for a phone?

$NVDA making $7.8M per employee is impressive, but the harder question is what keeps that number this high.

Nvidia outsources manufacturing, AI demand is still running hot, and revenue is growing much faster than headcount. That's a great setup while supply stays tight and pricing power holds.

But revenue per employee can fall quickly if hyperscalers slow spending, custom chips take share, or Nvidia needs a much larger workforce.

The number to watch next isnt headcount. It's gross margin. If revenue keeps growing but margins start slipping, the "efficiency" story looks very different.

Second attack on Saudi Aramco's Jizan facilities in a month, per the FT. Damage assessment underway; one source calls it similar in scale to August's, which disrupted some refining without material impact on operations.

The market's answer: WTI +$0.59 in five minutes to $92.22, Brent +$0.62 to $95.86. Under one percent, on an attack that would have been a five-dollar headline two months ago.

That's the setup worth reading. Oil already carries a war premium north of $90 — the marginal attack now confirms the premium rather than adding to it. Repetition has been priced.

What changes that math, in order: the damage assessment (a material production hit is different in kind, not degree), any spread of attacks toward export terminals or shipping rather than a border-region site — and the calendar. Crude holding $92+ through Thursday walks straight into the CPI print the Fed reads before the 16th.

Watch: Aramco's statement on the assessment. Until then, the market has filed this under "known conflict, known target." #catalyst

The AI capex numbers stopped being believable a while ago. Here's the newest one anyway, with its own caveats attached.

Per The Information: Anthropic has signed ~14.8GW of compute contracts in 11 months — roughly 16GW total — with potential cost up to $517 billion spread over the next decade. Suppliers read like the whole industry: up to 5GW Amazon, 5GW Google/Broadcom, ~1GW Microsoft/Nvidia, plus a lease on all of SpaceX's Colossus 1 — 220K+ GPUs.

And the caveat the report itself carries: $517B is the maximum estimate. Some capacity may be delayed or never used; some contracts cancel early. Headline number, not committed spend.

The setup this feeds: OpenAI's stated target is bigger still — 30GW by 2030, ~$750B invested. So the two labs alone imply over $1 trillion of compute this decade, and every watt of it lands downstream as orders for chips, memory, optics, power.

Key line: these are contracts, not consumption. The gap between GW signed and GW actually drawn is where this either validates the entire supply chain trade — or becomes the overbuild story of the 2030s. Watch which. #catalyst

Trump posted that he'd stop trading with countries running a surplus against the U.S. unless the Fed cuts rates. The market didn't move. Evercore noted no visible impact on yields, stocks, the dollar, or rate-hike probabilities — a sharp contrast to the swings trade and Fed threats caused in early 2025.

The market is treating this as venting. That read is probably correct on the surface. But the mechanism underneath is worth tracing.

First-order: the threat itself is self-contradictory. Banning imports to force a rate cut would trigger an immediate supply shock — Henrietta Treyz at Veda Partners called this the "most egregious" of Trump's legal and economic mistakes. You don't ease inflation by choking supply. If anything, a trade halt pushes prices up, which pushes the Fed toward hiking, not cutting. The threat, taken literally, works against itself.

Second-order: the pressure on Warsh actually makes a hike more likely, not less. Evercore put it directly — Trump's renewed pressure "complicates things" for the Fed chair, and the analysts now see a hike as "a bit more likely." Betting markets have it just above 50% for the meeting in two weeks; Evercore says under 50%. Either way, the direction is up, not down.

Third-order: Warsh's independence is the variable that matters. Treyz thinks Warsh is trying to distance himself from Trump. Beacon's Stephen Myrow says Warsh will prevent a hike before midterms, keeping rates flat — "politics is trumping data, and Warsh is talking tough because he wants the market to do his job for him." Those are two different reads of the same person.

What would prove this view wrong is Trump actually following through on a trade halt — not posting about it, but signing something. So far every market participant is betting he won't. The moment that bet changes is the moment this stops being noise.

The variable to watch: whether Warsh holds or hikes at the next meeting. Everything else is posture.