The filings, read together, document a structural shift from GPU-centric AI training to custom-silicon AI inference at hyperscaler scale. This is not incremental — it is a supply chain reorganization.
Key finding this week (AVGO 10-Q, filed June 9): Broadcom reported Q2 FY2026 revenue of $22.19B (+48% YoY), with semiconductor revenue $15.0B (+79% YoY) driven explicitly by “custom AI accelerators and AI networking.” The forward signal sits in the commitments footnote: firmly-committed remaining performance obligations reached ~$164.6B — including a new long-term custom-AI-accelerator contract signed during the quarter — and Broadcom staged $55.2B (FY27) + $72.9B (FY28) of purchase commitments to fund the XPU ramp. This is the same custom-silicon thesis as MRVL, confirmed at roughly 6× the scale: hyperscalers are signing multi-year ASIC supply contracts and the dollar amounts are now enormous.
Prior-week anchor (MRVL 10-Q, filed May 28) — still valid: Marvell posted $2.42B in Q1 FY2027 revenue (+27.6% YoY), then immediately after quarter-close committed $870M in foundry deposits to secure wafer and substrate manufacturing capacity through FY2033. A company does not lock in 7-year foundry commitments unless customers have already signed multi-year purchase agreements. Marvell designs custom AI ASICs for hyperscalers (Amazon Trainium, Google TPUs, Microsoft). The deposits are the hyperscalers' AI capex showing up one layer down the supply chain.
Corroborating signals from other filings this week:
GOOGAlphabet stacked an extraordinary multi-instrument capital raise in a single week — a 6.25% mandatory convertible preferred (335M depositary shares), a ~$18B Class A/C block sale, and a $40B at-the-market equity program. A company with Alphabet's cash flow raising this much equity is funding an AI/datacenter capex surge at the source — the same demand that shows up one layer down in AVGO/MRVL silicon and storage.
HPEQ2 FY2026 net revenue $10.68B (+40% YoY) (10-Q, filed June 2). Server revenue +32.7%, driven predominantly by higher average selling price (the AI-server mix), and Networking +148% on the Juniper acquisition — HPE has re-segmented around a new “Cloud AI” reporting unit. The systems layer of AI capex is now showing up in legacy-OEM revenue, not just merchant silicon.
PANWQ3 FY2026 revenue $3.0B (+31% YoY); Next-Gen Security ARR hit $8.1B (~+45% since FY-end) and RPO reached $18.4B. The platformization push is landing even as GAAP margins dip on CyberArk deal costs.
RBRKCyber-resilience demand compounding: revenue +39% YoY ($387M), Cloud ARR $1.39B (+43% YoY), net dollar retention ~120%, operating-loss margin halved YoY. The security layer keeps repricing on AI-era attack surface.
CRMQ1 revenue $11.1B (+13.3% YoY). The "Data 360, Headless Platform and Other" segment — which contains AI agent infrastructure — grew +24.8% YoY ($3.68B vs $2.95B). Enterprise AI adoption is in production, not pilots.
SNOWAcquired Observe, Inc. for $596M — an AI-powered observability platform. Positioning as the runtime layer where AI workloads generate and consume data.
IBM$10B, 5-year quantum computing investment + Letter of Intent with the Department of Commerce to build a U.S. quantum chip foundry ("Anderon"), targeting the first fault-tolerant quantum computer by 2029. The next wave is being funded in parallel with AI.
WDCAppointed Manuvir Das — VP of Enterprise Computing at NVIDIA — to its board. An NVIDIA AI infrastructure executive joining a storage company is the clearest possible signal that AI training data storage is a bottleneck being discussed at the highest levels.
SNPSTime-based EDA revenue +14.2% YoY. Elliott Investment Management (Jesse Cohn) landed a board seat. Every AI chip that MRVL, NVDA, AMD and AMAT designs flows through Synopsys tools first.
The signal: AI inference is being industrialized. Hyperscalers are moving from buying GPUs off the shelf to commissioning vertically integrated silicon, and the whole supply chain — from EDA tools to foundry capacity to interconnect chips to storage — is repricing.
New this week (June 3-9 prints): The chip→cloud→data→security cascade kept printing, now at scale. AVGO's $164.6B committed RPO is the supply-side megaphone for hyperscaler custom silicon; GOOG raised ~$58B across a convertible preferred, a block sale, and a $40B ATM to fund the capex behind it. The security/data layer compounded: PANW NGS ARR $8.1B (~+45% since FY-end) on +31% revenue, RBRK Cloud ARR $1.39B (+43% YoY) on +39% revenue, and DOCU's AI-native IAM reached 12.6% of ARR. Component-cost inflation also surfaced: NTAP raised product prices in Q4 on rising memory/component costs, while WDC retired $858M of 2028 converts (deleveraging into the up-cycle). The systems layer printed too: HPE's Q2 net revenue hit $10.68B (+40% YoY) with Server +32.7% on AI-server ASP mix and Networking +148% on Juniper.
Who Benefits — and Whose Trajectory It Moves
The demand signals above lift the obvious beneficiaries — AVGO / MRVL (direct custom-AI silicon), CRM / SNOW (AI software), SNPS / TSM (EDA + foundry), WDC / STX (AI nearline storage), and IBM (quantum optionality). But on a mega-cap base the signal barely bends the trajectory — AVGO’s $164.6B RPO is a durability signal, not a needle-mover on a ~$1.9T base. The names below are the ones still small enough that a single customer, contract, or print can meaningfully change the growth curve:
Primary Pick
CRDOCredo Technology
~$40B market cap (up from ~$4-6B a year ago — the AI interconnect rally already pulled the stock 7-10x). Makes high-speed SerDes (serializer/deserializer) connectivity chips that sit between GPU clusters and their network fabrics. As AI clusters scale from hundreds to hundreds of thousands of GPUs, every node needs Credo-type interconnects. CRDO's revenue base is still small relative to its mid-cap valuation, so a single new hyperscaler customer can still meaningfully move the needle — but the valuation cushion is gone.
PCIe Gen 6 · CXL interconnect · AI networking
Secondary Pick
ALABAstera Labs
~$55B market cap (up from ~$10-15B at IPO). Makes PCIe and CXL retimer/connectivity chips. CXL is the memory expansion fabric that AI systems will use to give GPUs access to far more memory than they physically carry — critical for large language model inference. ALAB is pre-revenue at scale on CXL; a trajectory change could be dramatic once adoption inflects, though much of the upside on AI retimers is already in the stock.
CXL memory expansion · PCIe retimers · AI inference enablement
Data Layer Pick (new this cycle)
MDBMongoDB
Mid-cap (cap pending SUM tab refresh; historically ~$15-18B range). Q1 FY2027 10-Q (filed May 28) shows the clearest "AI workloads hitting the data layer" evidence so far: Atlas mix climbed 72% → 75% YoY, net ARR retention at 121% driven by large existing customers expanding consumption, and Atlas now natively exposes Voyage AI embedding + reranking models for direct RAG/vector productization. The big inflection: swung to GAAP profitability ($0.05 diluted EPS vs -$0.46 PY). If the street still models MDB as a "growth slowing" pure DB play, the AI-vector mix shift + first GAAP-positive print is the re-rating catalyst.
Atlas vector search · RAG infrastructure · Voyage AI embedded
Institutional Accumulation Signal
NBISNebius Group
~$52B market cap. Schedule 13G (filed May 27) discloses Situational Awareness LP — the AI-thesis fund founded by Leopold Aschenbrenner — holding 5.6% of Nebius’ Class A ordinary shares (event date May 19). Nebius is a European/sovereign AI cloud infrastructure provider building GPU compute capacity where no AWS or Azure equivalent exists at scale. The buyer matters here: a dedicated AI-compute fund taking a passive 5%+ stake is a higher-signal accumulation tell than a generic index flow.
Sovereign AI cloud · Situational Awareness LP 5.6% · 13G filed May 27
Actionable Ideas
Ranked by near-term actionability — the valuation / priced-in read is folded into each idea.
1AVGO — Contracted demand: $164.6B RPO + a new multi-year custom-AI-accelerator contract turn the hyperscaler-silicon thesis into committed revenue. Largely priced (~$1.9T), but the RPO de-risks forward models. Risk: contract durability + top-5 customers ~45% of revenue. Confirm: Q3 FY26 (~Sep).
2MRVL — Likely mispriced: the $870M FY2033 foundry deposit sits in a 10-Q footnote, beyond any trailing model and not yet in consensus. Purest mid-cap read on the thesis AVGO just confirmed; the next guidance print should “beat” on visibility. Confirm: Aug 2026 earnings.
3SNPS — Mispriced the other way: $10B Ansys debt + a $300-350M restructuring compress near-term FCF, but Elliott’s board seat + the EDA monopoly = a compression-then-expansion setup, and Ansys broadens the moat. Resolves over 3-4 quarters. Starter now, add on restructuring-complete.
4MDB — Data-layer AI inflection: Atlas mix 72→75% YoY, 121% NRR, first GAAP-positive print, Voyage AI embedded for RAG. If the street still models “growth slowing,” the AI-vector mix shift is the re-rating catalyst. Confirm: Q2 FY27 Atlas mix breaks 76%+.
5PANW / RBRK — Security & data-resilience repricing on the AI attack surface. PANW NGS ARR $8.1B (~+45% since FY-end) on +31% revenue; RBRK Cloud ARR $1.39B (+43% YoY), ~120% NDR. Both richly valued — size accordingly; watch PANW’s CyberArk-driven GAAP-margin dip.
6WDC — Underappreciated: the NVIDIA-exec board appointment is the tell, and the $858M convert retirement is balance-sheet cleanup into the up-cycle. The market prices it as commodity storage; AI nearline demand adds a new leg. Asymmetric if AI storage inflects.
7CRDO / ALAB — Small-cap AI interconnect, higher risk/return. Size small; wait for a tier-1 hyperscaler design-win to add materially.
8IBM — Long-duration quantum call option: $10B + DoC foundry partnership is the biggest quantum policy news in years. A 3-5 year thesis, not an earnings-quarter one. Smallest position, longest hold.
Wait on the insider-supply names.CRWV (another 10 Form-144s + a universal shelf), DELL (Silver Lake trim to ~7.2% + 21 Form-144s), and DDOG (9 Form-144s in one week) all show clustered pre-arranged insider selling. Real businesses, but the supply overhang argues for waiting before initiating.
Source: SEC EDGAR filings.
Disclaimer: This page is for educational purposes only. NO investment advice.