In partnership with
Welcome to the iPrompt Newsletter
Friday Edition
What you get in the Friday Edition
Weekly Scoreboard
Top Headlines of the Week
Our Investing Angle
Three Ideas to Research This Weekend
AI Investment Framework
Your Move
…all in a FREE Weekly newsletter.
Sponsor:
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
iPrompt Signals
AI & robotics investing — explained so you can actually act on it.
ISSUE 04 · FRIDAY 24 APRIL 2026 · R. LAURITSEN
THIS WEEK
Google split the AI chip on Wednesday — training on one side, inference on the other, with a multi-gigawatt TPU commitment from Anthropic nobody priced in. A day later, SK Hynix printed a 72% margin and warned the memory shortage runs to 2030. IBM and ServiceNow beat earnings and got demolished anyway. Three stories. One pattern: capital has stopped concentrating. It’s sorting. The question isn’t whether to own AI — it’s which piece.
UPSTREAM | DOWNSTREAM | THE MIDDLE |
Memory + custom silicon winning. SK Hynix 72% margin. MU +8.5% WoW. | Physical AI gets funded. Tesla $25B capex. Nuclear and robotics pull forward. | Enterprise SaaS rerated as commodity. NOW −20%. Earnings were fine. |
Capital is sorting. Your job this weekend: pick a lane and test it.
WEEKLY SCOREBOARD
Ticker | Price (23 Apr) | Week % | What happened |
UPSTREAM · MEMORY + CUSTOM SILICON | |||
NVDA | $199.64 | −1.2% | Held $200 despite Google TPU split; nuclear tie-up with Oklo |
AVGO | $419.94 | +5.4% | Google partnership expanded; Anthropic + OpenAI now on TPU design list |
MU | $481.72 | +8.5% | HBM supercycle confirmed by SK Hynix; now +586% in 12 months |
GOOGL | $338.89 | +5.5% | TPU 8t + 8i launched at Cloud Next; training/inference split |
THE MIDDLE · ENTERPRISE SAAS | |||
NOW | $84.78 | −20.0% | Beat Q1 + blamed Middle East deal slippage; worst single day ever |
BENCHMARKS SMH $481.85 +3.8% · S&P 500 7,108.40 −0.4% · VIX 19.31 +13.0% |
VIX = the “fear gauge.” Measures expected S&P 500 swings over the next 30 days. Above 20 = real anxiety. It’s crashed 44% in three weeks — the fifth-biggest drop in history.
BOTTOM LINE Money isn’t leaving AI. It’s moving around Nvidia — into memory, custom silicon, and the few places that can still price their product. Software that sells AI? The air just got thin. |
TOP HEADLINES
→ Four stories. Three lanes. Match each one as you read — it speeds the Angle.
1. Google splits the chip and signs Anthropic for gigawatts.
At Cloud Next on Wednesday, Alphabet unveiled the TPU 8t (training) and TPU 8i (inference) — separate architectures, scaling to one million chips per cluster. Google claims 2.8× better training price-performance and 80% better inference price-performance vs the seventh-gen Ironwood. The real news was buried: Anthropic committed to multiple gigawatts of TPU capacity. OpenAI is also now on the customer list.
→ The custom-silicon story just stopped being theoretical. Google’s ecosystem expansion pulls Broadcom and Marvell up the stack alongside it.
2. SK Hynix prints $27B in net profit. The memory duopoly is pricing like it knows something.
Operating margin 72%. HBM share 57%. Profit growth +405% YoY. And Chairman Chey Tae-won said the quiet part out loud — HBM shortage extends to 2030, capacity gap above 20%. Samsung began shipping HBM4 in February; SK shipped HBM4 samples a year earlier and plans HBM4E samples in H2 2026.
→ The memory story is the AI infrastructure story nobody’s discounting enough.
🌱 New to investing? Here’s what this means HBM = High-Bandwidth Memory. It’s the specialised RAM that sits next to AI chips (like Nvidia’s GPUs) and feeds them data. Without HBM, the chip starves. Only two suppliers matter — SK Hynix and Samsung — and demand has run so far ahead of supply that SK Hynix is earning 72 cents on every dollar of revenue. When one link in the chain captures that kind of margin, it tells you where real scarcity lives. |
3. Tesla drops a $25B bomb, and the market hates the math.
Capex raised from $20B just three months ago. Roughly 3× last year’s spend. CFO Vaibhav Taneja said free cash flow turns negative for the rest of the year. Musk is betting the money on AI chips, Cybercab production, Optimus lines, and six factories simultaneously.
→ Physical AI is no longer a pitch deck. It’s a line item. But the cost of admission just tripled — and Tesla at 320× trailing earnings is paying it out of operating cash.
4. Enterprise SaaS had its worst day in two years. The earnings were fine.
IBM beat by $0.10, reiterated guidance, fell 9%. ServiceNow beat top and bottom lines, gave in-line subscription growth, fell 17%. Salesforce −9%. Adobe −7%. Workday −9%. Oracle −6%. None of these results were bad.
→ Investors have decided AI is a cost for enterprise SaaS, not a growth lever. Until one of them shows AI-attributed ACV growth accelerating, the sector is guilty until proven innocent.
OUR INVESTING ANGLE
→ The classification call. This is the centrepiece of the issue — read slowly.
Everyone’s watching whether Nvidia holds $200. The smarter bet is watching where the capital around Nvidia is going.
For most of 2025, the AI trade was a single stack: buy Nvidia, buy the hyperscalers, hold the ETF. That trade worked until it didn’t, and this week was the first clean signal that “didn’t” is now the base case.
The thesis. AI capital is sorting into three lanes, and they’re pricing completely differently.
UPSTREAM WINS Event: SK Hynix printed a 72% operating margin and Chey called the HBM shortage to 2030. Reaction: MU +8.5% WoW, AVGO +5.4%, Google TPU 8 pulled custom-silicon designers up alongside memory. Implication: Pricing power holds through H2 at minimum — capacity expansion lag is ~18 months. Watch: Samsung 12-layer HBM4 Nvidia qualification by Q3. |
DOWNSTREAM WINS (CONDITIONALLY) Event: Tesla raised 2026 capex from $20B to $25B and committed to negative FCF for the rest of the year. Reaction: TSLA sold off, but the number is now public — Optimus, Cybercab, and six factories are all funded. Implication: Capital flows to Japanese motion control (FANUC), surgical robotics (ISRG), and power infrastructure (Oklo, GEV). Watch: FANUC Q1 order book (mid-May) for demand confirmation. |
THE MIDDLE GETS CRUSHED Event: IBM and ServiceNow beat earnings and got demolished anyway. Salesforce, Adobe, Workday, Oracle followed. Reaction: Sector re-rated from growth to commodity — NOW at 55×, CRM at 40× suddenly look exposed. Implication: One more in-line quarter and these multiples compress further. Named losers, not categories. Watch: Any AI-attributed ACV growth print from CRM (28 May), NOW, or IBM Q2. |
If you’re long the middle, you’re paying growth-stock prices for what the market is now treating as commodity software. Think CRM at 40×, NOW at 55×, IBM’s software business propping up a 2% dividend. Named losers, names that could genuinely crack further if Q2 prints another in-line quarter.
Companion deep dive: The Three Lanes of AI Capital →
⚠ WHAT COULD GO WRONG? (THE BEAR CASE) 1. Hormuz escalates. WTI above $110 changes everything. Rate cuts go off the table, the dollar rallies, speculative tech flows reverse. Iran risk is currently priced as a blip — it’s an event, not a blip, and VIX knows it. Breaks: all three lanes, middle first. 2. Memory normalises faster. Samsung qualifies HBM4 with Nvidia by Q3, Chinese CXMT ramps DDR5, and SK’s 72% margin proves to be a peak, not a plateau. The shortage narrative is 18 months old — it can be wrong. Breaks: upstream directly. 3. Enterprise SaaS damage spreads upstream. If NOW and IBM can’t sell AI to enterprises, hyperscalers eventually can’t sell AI compute to them either. The bifurcation thesis assumes demand stays. If it doesn’t, everything above the middle layer gets re-rated down. Breaks: the whole thesis — middle contagion into upstream. Size your position for the possibility that the thesis takes longer than expected — and that oil is the variable that breaks first. |
THREE IDEAS TO RESEARCH THIS WEEKEND
→ One test per lane. Research the one you believe least — that’s where conviction gets built.
Not recommendations — starting points for your own research.
IDEA 1 MU — Micron Technology The memory printing press |
WHY NOW. SK Hynix printed 72% operating margin on HBM; shortage runs to 2030. Micron is the only US-listed HBM supplier with meaningful share. |
THE CASE. Capacity expansion lag of ~18 months means pricing power holds; margins should re-rate toward Korean comps. |
THE RISK. Cycle timing; Samsung getting 12-layer HBM4 qualified with Nvidia. |
TRIPWIRE. If Samsung 12-layer HBM4 passes Nvidia qualification by the 27 May Nvidia Q1 FY27 call, the duopoly premium compresses within the quarter. |
HOW TO RESEARCH. MU investor relations page; SK Hynix Q1 earnings call transcript (23 April); TrendForce HBM pricing data. |
IDEA 2 AVGO — Broadcom (Marvell quietly) Custom silicon designers |
WHY NOW. Look, I keep coming back to Broadcom and I get why people push back. It’s up 140% in 12 months. It’s a $2T company. You’re not early. |
THE CASE. Google Cloud Next added Anthropic AND OpenAI to the TPU co-design list. Every hyperscaler wanting to reduce Nvidia dependency is a Broadcom or Marvell customer. |
THE RISK. Customer concentration — Google alone is the lion’s share of custom-AI revenue. |
TRIPWIRE. If any credible report surfaces before the 5 June AVGO Q2 call that Google is moving meaningful TPU design work to Marvell, AVGO’s concentration premium unwinds fast. |
HOW TO RESEARCH. AVGO Q2 2026 earnings transcript (scheduled 5 June); Google Cloud Next keynote replay (22 April). |
IDEA 3 FANUC (6954.T), BOTZ ETF Japanese robotics (non-US) |
WHY NOW. Honest take — I’m not fully sold on this one. The gap between “Japan dominates robot motion control” and “Japanese robotics participates in the physical AI trade” has been wide for a decade. |
THE CASE. But Tesla’s $25B capex, Japan METI’s 30% global physical-AI share target by 2040, and NVIDIA’s Hannover Messe push collectively mean the sector finally has a funded catalyst. |
THE RISK. Sector traded on promise for 10 years without delivering. |
TRIPWIRE. FANUC’s Q1 FY27 order book (reported mid-May) below 15% YoY growth = the physical AI order flow is still theoretical. |
HOW TO RESEARCH. FANUC quarterly results calendar; METI physical-AI roadmap (March 2026 release). |
AI INVESTMENT FRAMEWORK
This is where each lane actually sits in a portfolio. Upstream gets the sizing. The middle gets the haircut. Downstream gets conditional weight — wait for the FANUC print before scaling. Not financial advice; research starting points.
CHANGES THIS WEEK. Physical AI moved from DEVELOPING to MEDIUM conviction on Tesla’s capex confirmation + Hannover Messe deployment evidence. Cybersecurity holds at MEDIUM — waiting on Q1 earnings from CRWD and PANW. No other layer changes. |
LAYER | TICKERS | YTD | CONV. | RISK | SIG. | SIZING |
INFRASTRUCTURE | NVDA · AVGO · MU | +31.0% | HIGH | ●●●○○ | ↑ | 15–20% |
PLATFORMS | MSFT · GOOG · AMZN | +13.4% | HIGH | ●●○○○ | ↑ | 15–20% |
APPLICATIONS | PLTR · CRM · NOW | −16.3% | DEVELOPING ↓ | ●●●●○ | ↓↓ | 5% |
PHYSICAL AI | BOTZ · ISRG · FANUC | +8.7% | MEDIUM ↑ | ●●●●○ | ↑↑ | 5–10% |
CYBERSECURITY | CRWD · PANW · ZS | +16.3% | MEDIUM | ●●●○○ | ↔ | 5–10% |
GLOBAL | BABA · 9984.T · SAP | +19.8% | MEDIUM | ●●●●○ | ↑ | 5% |
Per-layer notes.
INFRASTRUCTURE AVGO is the story this week — Google + Anthropic + OpenAI on the TPU design list. MU is the HBM proxy. NVDA at $199 is fine; the issue is relative performance vs custom-silicon designers.
PLATFORMS GOOG broke out on TPU 8. MSFT flat on Copilot monetisation questions. AMZN Q1 call next week (1 May earnings).
APPLICATIONS Conviction cut. NOW at 55× after a beat-and-miss-guidance print is a category warning, not a stock warning.
PHYSICAL AI Tesla $25B capex is the catalyst; FANUC Q1 order book (mid-May) is the confirmation.
CYBERSECURITY Project Glasswing narrative intact but needs Q1 earnings (CRWD 27 May) to confirm revenue attach rates.
GLOBAL SAP at record high on Hannover Messe AI factory positioning. 9984.T strong on SpaceX/xAI upside.
What we’re watching.
DATE | EVENT | QUESTION TO TRACK |
30 Apr | MSFT Q3 earnings | Is Copilot ACV growth accelerating or stalling? |
20 May | NVDA Q1 FY27 earnings | Q2 guidance vs the $78.6B consensus |
Mid-May | FANUC Q1 FY27 order book | Growth rate — the physical AI tripwire |
5 Jun | AVGO Q2 earnings | Custom silicon revenue share + Google concentration |
Disclaimer: This newsletter is for informational and educational purposes only and does not constitute financial advice. iPrompt Signals is not a registered investment advisor. Always conduct your own research and consult a qualified financial professional before making investment decisions. |
YOUR MOVE
Three things to watch between now and June. Each tests one lane directly:
1. Samsung’s HBM4 Nvidia qualification, expected Q3. One headline breaks the memory duopoly premium. SK Hynix at 72% margin won’t survive that news intact. MU is the US proxy.
2. AVGO’s Google concentration disclosure, 5 June Q2 call. If Broadcom reveals any TPU design work moving to Marvell, the concentration premium unwinds inside a quarter.
3. FANUC’s Q1 order book, reported mid-May. Below 15% YoY growth = the Tesla-capex physical-AI narrative outran actual order flow. Above 20% = the downstream lane is real.
Research one. Not all three. One.
Pick the lane you believe least and spend an hour trying to falsify it — that’s where conviction gets built.
P.S. Three weeks ago I flagged the Japanese robotics discount — 28× vs 58× for US peers — and the BYD/Huawei industrial-robotics tripwire. Neither has fired yet. The gap’s narrowed slightly. Watch Tesla’s Optimus production update at the next all-hands.
🌱 SHORT TAKE — new here? This week: AI capital stopped piling into one stack and started choosing sides. Easiest takeaway — for broad exposure without stock-picking, something like VGT (tech sector) or SMH (semis) covers most of the upstream story, without the risk of picking one winner. Not a recommendation — a starting point. |
Stay curious — and stay qualified.
— R. Lauritsen
Know someone building an AI position? Forward this — they’ll thank you by Friday.
QUICK GLOSSARY
ACV | Annual Contract Value. The revenue a software company books from a customer in a year. When investors ask “is AI driving ACV?” they mean: are customers paying more because of AI features? |
Capex | Capital expenditure. Money spent building physical assets (factories, data centres, chip lines). Tesla’s $25B number is capex. |
Custom silicon | Chips designed for a specific use case instead of general-purpose (like Nvidia GPUs). Google’s TPU, Amazon’s Trainium, Meta’s MTIA are examples. |
Free cash flow | Cash from operations minus capex. When Tesla says FCF turns negative, they mean they’re spending more on factories than they’re earning. |
HBM | High-Bandwidth Memory. The specialised RAM that sits next to AI chips. SK Hynix and Samsung make 95%+ of it globally. |
Hyperscaler | The giants who run most of the world’s cloud compute: Amazon (AWS), Microsoft (Azure), Google (GCP). They’re the biggest AI spenders. |
Inference | Running an already-trained AI model to answer a prompt. Cheaper than training, but happens billions of times a day. |
TPU | Tensor Processing Unit. Google’s in-house AI chip. Until this week, one chip did training and inference. Now it’s two. |
iPROMPT SIGNALS · FrontWave Media Ltd · 24 APR 2026 Page of
Sponsor:
A Senior Analyst Sees Half a Billion Dollar Potential.
Kingscrowd Capital's senior analyst reviewed RISE Robotics and projected potential growth to a $500 million valuation. The community round is open now on Wefunder. You don't have to be an institutional investor to get in at today's price.