Bull Market Broadens — NVDA Corrects, Everything Else Bids
NVDA in correction (~$210, -11% from the $236 earnings high) — and the bid moves out, not down. Snowflake +40% one layer up, second-tier semis taking leadership inside.
Edition · 2026-05-28 · KW22
Thank you for reading this week’s edition of Closelook@Hypergrowth, dated Thursday, 28 May 2026. The next edition will be published on Thursday, 4 June 2026.
The Hypergrowth tape just gave us another clean broadening this week — in two directions at once.
Horizontally inside the semis: NVDA is in correction territory, down ~11% from the $236 earnings-print high to the ~$210 area, and the bid has rotated outward toward networking, packaging, and second-tier compute names rather than collapsing.
Vertically up the stack: Snowflake printed +40% on earnings — the first earnings-quality proof that consumption-priced software (Work-as-a-Service 3.0) is structurally re-rating.
Two regimes broadening simultaneously is not a bull-market top. It is a bull market maturing.
Table of Contents
1. This Week’s Action
2. Overview: The Hypergrowth Map
3. The State
4. The Outlook
5. What May Lie Ahead: Macro Setup
6. What May Lie Ahead: Technical Setup
7. The Hypergrowth Portfolio: What We Did
8. The Hypergrowth Portfolio: What We Plan To Do
9. Knowledge Corner — Work-as-a-Service: Why Snowflake’s Print Matters
10. Upcoming Transactions: Be Informed
11. What May Go Wrong: Risk & Change Triggers
12. Final Words
(1) This Week’s Action
The bull market in compute is no longer the only story — and inside compute, it has not been the NVDA story for quite a while. NVDA is in correction territory: ~$210, down roughly 11% from the $236 post-earnings high. That is a meaningful event.
What is more meaningful is the tape’s reaction: the bid did not collapse; it rotated outward. Memory, networking, packaging, and the second-tier compute names absorbed the flow that left NVDA.
PC hardware companies such as Lenovo and edge companies such as Qualcomm have joined the bull market.
One layer above the silicon, the surprise was even louder: Snowflake printed earnings and traded +40%, the largest single-week move in the agentic universe this year. The temptation is to call it a SaaS recovery.
This Week at a Glance — Closelook Indices (KW22)
The seismograph reads classic broadening, not weakness. AW25 holding while its single largest constituent (NVDA) corrects double-digit is precisely the signature of a healthy rotation regime.
The Snowflake move tells us the same broadening is now extending beyond the picks-and-shovels layer into the Work-as-a-Service tier. Two broadening vectors, one regime. That is a regime-maturing signal, not a regime-change signal. While writing, we are waiting for the MDB results to be published after the close today.
(2) Overview: The Hypergrowth Map
RubinBuildoutt 100 — 18 Sectors (KW22)
Memory still owns the YTD outlier print (+250-ish range).
The carriers this week were Memory, EDA & Chip Design (Arm Holding), Substrates, and Foundry.
Tech Thematics — 21 ETFs (KW22)
Semis are still inside, leading, but software and SaaS are breaking out. The interesting move sits inside software: the index broadened — not from legacy seat-based vendors but from consumption-tier names (Snowflake +40%, Datadog firm, MongoDB stable). Speculative pockets (genomics, quantum) flat — the bid stayed disciplined. AI-application thematics quietly bid through the week, telegraphing what the Snowflake print confirmed on Wednesday.
(3) The State
The Thesis
We are mid-cycle in the computer buildout — NVDA the pacesetter (still, even in correction), Rubin layers rotating leadership month to month, and capex visibility through 2027. The cycle is now visibly maturing on two fronts.
First: leadership within the buildout layer is broadening away from the single-stock NVDA dependency toward second-tier compute, networking, and packaging names.
That is precisely what mid-cycle behavior looks like — and the NVDA correction is the mechanism by which it happens, not a contradiction of the thesis. Second: outcome-priced software riding atop the agentic compute layer is no longer a framework — the Snowflake print is the first earnings-quality proof that this layer is structurally re-rating.
Linked entries: SaaSpocalypse · Work-as-a-Service · Agentic Accounting.
Regime Verdict
🟢 Hypergrowth regime intact — broadening, not breaking. NVDA’s correction is the orderly mechanism by which leadership widens; the absence of contagion into the rest of the buildout layer is the bullish tell.
The new wrinkle: the SaaS 3.0 cohort just got its first proof point. We treat Snowflake as a category signal, not a stock call. The cohort to watch is consumption- and outcome-priced names sitting on the agentic runtime. The cohort to keep underweighting is per-seat legacy SaaS — structurally pressured by the same dynamic.
TLT holds major support around 83. This is short-term and long-term bullish for risk-on assets.
(4) The Outlook
Forward Reads — Cohorts & Names
• NVDA specifically: in correction at ~$210, -11% from the $236 earnings high. Thesis unchanged. We treat this as a controlled multiple-reset inside a structurally bid name, not a top — but we respect the tape and do not add aggressively until the broader correction completes.
• Memory (Micron, SK Hynix, Samsung): the immediate beneficiaries of the NVDA broadening. The flow that leaves the cap-heavy name is rotating here, not exiting the layer.
• Networking / Interconnects (ANET, CRDO, CIEN): the other broadening direction. Bid all week.
• Picks-and-shovels at the wafer level (ASE, BESI, TSM and more): thesis intact, earnings remain the gravity. TSM monthly sales the next read.
• AI-Native infrastructure (Anthropic-adjacent, model-runtime plays): bid. Sovereign-AI tendering picking up — read it through TSM monthly sales.
• Software 3.0 / Consumption-tier (SNOW, DDOG, MDB, NET): the new cohort to track. Snowflake set the bar. The next proof point comes from CRM (Agentforce metrics) or NOW (agent-conversation cadence).
• Consumer-AI: startim to show signs of life .
• Per-seat legacy SaaS: structurally pressured. Trim on strength rather than buy weakness.
Watchlist — What Would Flip the Read
• NVDA correction depth: -11% so far. A -15% to -18% pullback would still sit inside normal bull-market correction parameters. A break through that into the -25% range escalates from “correction” to “regime question” — and would pull the second-tier compute names down with it. We monitor the depth, not just the level.
• Contagion into the second-tier names: the current bull tell is that other stocks are absorbing the NVDA flow. If they roll over too, the broadening read is wrong and we have a layer-wide correction, not a single-stock one.
• Memory ASP rollover: Micron’s next monthly tells us whether HBM pricing is bending. The YTD outlier becomes the YTD problem if it bends.
• A second WaaS proof point: Salesforce Agentforce paid-credit run-rate or ServiceNow agent-conversation volume. One print is a stock move. Two prints is a category.
• Power constraint signal: Constellation/Vistra revenue cadence + AEP utility capex commentary. If power becomes the binding constraint, Compute multiples compress regardless of fundamental demand.
(5) What May Lie Ahead: Macro Setup
We see a calm macro setup in the week ahead.
(6) What May Lie Ahead: Technical Setup
The Technical Reading on NASDAQ 100
This is an earnings-driven bull market with many more legs.
Key Levels · Alignment Check
═══ Constructive Pullback — Bullish bias, defensive execution ═══
Trend long-term up · short-term correcting toward the channel midline. Momentum reset — exactly the cooling that mid-cycle corrections produce, and exactly the cooling that the next leg requires before it can run. Volume on the way down is moderate, not distribution-class — the bullish tell is still that NVDA is correcting on rotation, not on capitulation. Indicator alignment is bullish on the weekly and neutral on the daily. The setup invites add in tranches, not a single-shot entry.
(7) The Hypergrowth Portfolio: What We Did
We trimmed Infineon.
(8) The Hypergrowth Portfolio: What We Plan To Do
Moves under active consideration — conditional on signal confirmation.
• Snowflake (SNOW): do not chase the +40% print. We look to add only on a constructive pullback — the thesis is multi-year, the entry does not need to be today.
• Packaging-layer add: monitoring BESI, TER on cycle confirmation. We want one more print of CoWoS supply normalisation before scaling.
• NVDA: the correction is the opportunity, but tranched. We add in slices into weakness inside the channel — first slice now in the $210 area, second slice reserved for a move toward the channel midline near $200 if it gets there. We do not deploy the full add at one level.
(9) Knowledge Corner — Work-as-a-Service: Why Snowflake’s Print Matters
The interesting thing about Snowflake’s +40% week is not the magnitude. It is what the market was paying for.
Snowflake does not sell seats. It sells compute credits — you pay per query, per workload, per byte processed. That model was unfashionable for three years because it created revenue volatility: when customer compute spend went down, so did Snowflake’s. This week, the same volatility flipped the other way. AI workloads are consuming data warehouses at a rate the consensus model did not price in, and the print made that visible.
This is the first concrete evidence that the Post-SaaS thesis — what we call Work-as-a-Service — is moving from a framework into actual revenue prints. The shift is from selling tool access (per-seat SaaS) to selling completed outcomes (per-workload, per-resolved-ticket, per-generated-report). Three models are emerging. Success credits — customers buy blocks of agent conversations that get consumed when work is completed (Salesforce Agentforce is the loudest example). Outcome-based pricing — you pay only when the measurable outcome is delivered (Palantir AIP is testing this in enterprise). Consumption-based hybrid — base platform fee plus per-action charges. That third model is the Snowflake template, and the one the market just validated.
There is a second mechanism underneath this re-rating that is rarely discussed: Agentic Accounting. When AI agents move from experimental tools to production systems that directly serve customers, auditors push companies to reclassify the inference compute from R&D to Cost of Goods Sold. A company that spends $500M on AI as R&D shows healthy gross margins. The same company reclassifying that spend as COGS suddenly looks margin-challenged. Per-seat SaaS vendors are walking into this trap — their per-seat revenue does not scale with inference, but their COGS now does. Snowflake walks past this trap because its revenue scales with the same compute that drives its costs. That is the structural asymmetry the market just paid for.
The losers in the WaaS transition are companies whose moat was built on per-seat lock-in. Once an AI agent replaces 10 seats with a single credit-consuming workflow, the seat-based revenue base is structurally at risk. That is the SaaSpocalypse mechanism — not “SaaS dies” but “per-seat SaaS shrinks unless it pivots to outcome or consumption pricing.” The winners are companies that own two layers: the outcome measurement infrastructure (proving that an agent resolved a ticket, generated a lead, or completed an analysis) and the runtime where agents execute, and the state is managed. Snowflake sits at the data-runtime layer. Palantir at the outcome-verification layer. Salesforce at the workflow-completion layer.
For the reader: stop reading SaaS as a single category. Split it into per-seat (structurally pressured by both the agent-replaces-seat dynamic and the COGS-reclassification dynamic) versus per-outcome / per-consumption (the new growth tier, structurally insulated from both). Snowflake’s print is the first earnings-quality proof. The second proof point likely comes from Salesforce Agentforce credit consumption or ServiceNow agent-conversation cadence in the next two earnings cycles. Watch for it. The first analyst to write a note titled “AI Inference is Now COGS” triggers the re-rating across the per-seat cohort.
(10) Upcoming Transactions: Be Informed
• Free: LinkedIn Newsletter (Closelook Hypergrowth) · closelook.net Free tier → closelook.net/subscribe
• Coming Soon · Paid: Real-time alerts · C+ Premium → closelook.net/c-plus
(11) What May Go Wrong: Risk & Change Triggers
The bull case is straightforward; the warning signals come from three places — NVDA’s channel, the memory layer, and a power-side surprise.
Trigger 1 · Memory ASP Rollover
If Micron’s next monthly print shows HBM ASP softening, the YTD outlier in Rubin (+250-ish) becomes the YTD problem. Memory funds the buildout margins for everyone in the stack — once ASPs roll, the gross-margin tailwind across Compute / Networking / Packaging unwinds with it. We watch the monthly print more closely than the quarterly print here.
Trigger 2 · NVDA Correction Goes from Healthy to Structural.
NVDA at ~$210 is -11% from the $236 high — that is a correction, not a top. Threshold to watch: a weekly close below the long-term channel base (currently the structural line the tape itself drew, running roughly 10–12% below current price). Below that line, the read shifts from “controlled rotation broadening the bid” to “leadership-stock break that pulls the layer down.” We measure depth and tape behavior, not single closing prints. The bullish tell remains intact as long as second-tier compute names continue to absorb NVDA’s flow rather than roll with it.
Trigger 3 · A Second WaaS Print That Disappoints
The Snowflake print is one piece of evidence. If the next read — Agentforce paid-credit metrics or ServiceNow agent-conversation volume — comes in soft, the SaaS 3.0 cohort cools fast. One print is enough for the move we just had. Two prints are needed for the cohort to compound. Keep the conviction calibrated to the data we actually have.
(12) Final Words
“The market is paying for outcomes, not access — and inside compute, it is paying for breadth, not concentration.”
This was the week the bull market grew up. NVDA, the single name that has carried the headline narrative for two years, went into correction, and instead of pulling the layer down with it, the bid moved out. Second-tier compute, networking, and packaging absorbed the rotation. That is what mid-cycle broadening looks like; it is the opposite of what a top looks like. The first Post-SaaS print, sitting one layer above the silicon, arrived. Snowflake’s +40% is not a SaaS comeback — it is the market telling us that outcome-priced software, running on top of the agentic compute layer, has its first earnings-quality proof: two broadening vectors, one regime, both bid.
We do not chase. We add NVDA in tranches into weakness inside the channel — first slice now, second reserved for the channel midline. We let the SaaS 3.0 thesis consolidate before adding to Snowflake. The setup runs for years if it runs at all, and the second proof point — Salesforce or ServiceNow — is what would let us scale conviction into position size. Until then: stay long the buildout layer, lean into the second-tier compute beneficiaries of the NVDA rotation, watch the consumption-tier cohort, and trim per-seat legacy on strength.
Thank you for reading. See you next Thursday.
Disclaimer: Closelook Venture GmbH is not a BaFin-licensed investment advisor. All content is investment-diary framing for educational purposes only. Nothing in this newsletter is investment advice.