CRDO · Credo Technology Group Holding Ltd · Stock research

Current Thesis

Credo sells the interconnect layer — active electrical cables, optical DSPs, SerDes IP — that hyperscalers buy by the rack to wire AI datacenters, and fiscal 2026 turned that into roughly $1.34B of revenue and $662M of non-GAAP net income. July and August tested whether that is a franchise bid or a flow bid. After the late-June initiation cluster carried price to a 52-week high of $302.52, the stock closed 2026-07-20 at $214.67 and traded to a reported $195.39 in early August, then rebuilt: +9.0% on 2026-08-04 to $237.92, and $249.89 at the 2026-08-07 close, still 17.4% below the high. The rebuild is macro-sourced — BofA's 2026-08-03 work put hyperscaler capex above $1.2T and named nine beneficiaries — and no company datapoint lands until the Q1 FY2027 print, calendar-listed for 2026-09-09, where the June guide of $465–475M meets a consensus of $470.38M. The narrative an investor is buying here is unchanged in substance and changed in stage: interconnect content per AI rack is still compounding, but the sell-side ceiling has not moved since 2026-06-22 and three officers sold into the July high.

Bullish and bearish views on Credo Technology Group Holding Ltd

The model's bull view on Credo Technology Group Holding Ltd (CRDO), in brief: Q4 FY2026 (2026-06-01): revenue $437.0M, +157% YoY, against $432.0M consensus; adjusted EPS $1.16 vs $1.03. The bear view: The price-target ceiling stopped rising. Both cases follow in full.

Bull Case

• Q4 FY2026 (2026-06-01): revenue $437.0M, +157% YoY, against $432.0M consensus; adjusted EPS $1.16 vs $1.03. One quarter outsold the whole of fiscal 2025.
• Full-year FY2026: ~$1.34B revenue, $662M non-GAAP net income (results summarized 2026-08-04) — the AEC/optical royalty is realized P&L, not a TAM slide.
• Guide vs consensus is not a stretch bar. Q1 FY2027 guided $465–475M on 2026-06-01 against $461.6M consensus then; the calendar consensus into 2026-09-09 sits at $470.38M revenue / $1.16 EPS — inside the company's own band.
• Sell-side marks remain above the tape. 19 analysts average $279.29 with a $290 median as of 2026-07-26; Stifel's $350 (2026-06-22) is the ceiling; Barclays raised to $300 from $260 on 2026-07-20 — that raise landed during the drawdown, not before it.
• Demand backdrop re-stated in August, not June: hyperscaler capex above $1.2T (BofA, 2026-08-03), with 2026-08-04 coverage framing the same number as a semiconductor-ETF-level bid.
• DustPhotonics $750M (cash plus ~0.92M shares), closing ~Oct–Nov 2026 (est.) extends the franchise from copper AEC into the optical/co-packaged layer where AVGO and MRVL concentrate.

Bear Case

• The price-target ceiling stopped rising. Top mark $350 set 2026-06-22; nothing above it through late July. Susquehanna's 2026-07-21 raise landed at $250 — at, not above, the 2026-08-07 close of $249.89.
• Officers sold into the high. Form 4 activity summarized in 2026-07-20 coverage: COO and director Yat Tung Lam ~56,000 shares (~$12.6M), CTO Chi Fung Cheng 27,500 shares on 2026-07-16 (~$6.6M), CFO Daniel W. Three insiders inside nine sessions.
• Multiple leaves no room for a mediocre quarter. P/E ~94.8 against a semiconductor industry median of 57.2, and a GF Value estimate of $219.90 versus $237.92 (both 2026-08-04).
• The July–August drawdown had no company cause. From $302.52 to a reported $195.39 with no guidance change: the name is a beta expression of AI-capex sentiment before it is a fundamental story.
• Customer concentration is unresolved. AWS has historically been over 40% of revenue; the 2026-04-28 OpenAI custom-silicon report produced a roughly -15% session on nothing but a headline.
• Interconnect supply chains sit inside that perimeter.

Setup & Price Structure

• Last completed daily close $249.89 (2026-08-07). 52-week high $302.52, so -17.4%. RSI(14) 60.4. Three-month return +32.6%.
• The structure that matters is the early-August reaction low reported at $195.39. Above it, June–July reads as a deep higher-low retest of the late-June breakout; lost on a weekly basis, the June initiation spike and the July fade become a completed distribution top.
• Positioning observables, stated as observables: the entire recovery is about a week old (+9.0% on 2026-08-04 to $237.92, then $249.89 on 2026-08-07) and arrives directly into a scheduled print; RSI at 60.4 is mid-range, not extended; consensus median $290 is only modestly above the tape, so the sell-side offers little upgrade fuel without new estimates; and three C-suite/director sales cluster into 2026-07-08 through 2026-07-16, the window of the July highs.
• Life-cycle: MATURING. Dating it — mainstream retail coverage since 2026-06-15 (Cramer on air) and 2026-07-06 (inclusion in "7 stocks that win the $1.5T AI buildout"); price-target ceiling unchanged since 2026-06-22; the August bid arrives through basket lists (2026-08-03, 2026-08-04) rather than name-specific research. It is not SATURATED: Barclays still raised on 2026-07-20 and the stock reclaimed most of a large drawdown inside a week. It is no longer ACCELERATING: the June condition — fresh initiations printing new highs — has not repeated in seven weeks.

Catalyst Calendar (next 30 days)

• 2026-09-09 (calendar-listed, after close, not yet company-confirmed): Q1 FY2027 results. Consensus $470.38M revenue / $1.16 EPS against the 2026-06-01 guide of $465–475M. This sits just past the 30-day window, and nothing company-specific precedes it — the interval is traded on macro and sector read-through.
• ~2026-08-26 (est., prior-year cadence, unconfirmed): Nvidia Q2 print. Sets the sector's risk appetite and the AI-networking read-through into CRDO's own print two weeks later.
• ~2026-08-27 (est., prior-year cadence, unconfirmed): Marvell quarterly print — the closest listed comparable on custom silicon and interconnect attach rates.
• ~2026-09-04 (est., prior-year cadence, unconfirmed): Broadcom quarterly print, the largest single read on AI networking dollars and on how much of the optical layer stays in-house.
• ~2026-08-20 to 2026-08-22 (est.): Jackson Hole. Rate-path commentary is the transmission channel for the multiple, given ~94.8x trailing earnings.
• ~Oct–Nov 2026 (est.): DustPhotonics close, including the ~0.92M share component — outside the window, but the filing language on that component is checkable before it.

What Would Change Our Mind

The break is structural before it is fundamental: the early-August low at $195.39 is the entire basis for treating July as a retest rather than a top, and it was set in a week when no company news existed.

Second, the 2026-09-09 print is a genuine binary for the stage question. Revenue landing inside $465–475M with a Q2 FY2027 guide at or below the current quarter's growth rate would show the 800G-to-1.6T ramp arriving on schedule but no longer surprising — the condition under which MATURING becomes SATURATED. A guide above consensus with an AWS-concentration figure that fell rather than rose would argue the opposite and reopen the accelerating case.

Third, watch whether the price-target ceiling moves. It has been static at $350 since 2026-06-22 while the stock round-tripped roughly a third of its value. Revisions clustering below the $290 median after the print, or further officer sales filed post-print, would confirm the supply side of this name is the informed side.

Correlation Notes

• Tightest read-throughs are MRVL and AVGO (custom silicon plus networking), ALAB (connectivity), and the optical complex AAOI / LITE / COHR — the 2026-06-10 session had CRDO, AAOI and LITE moving together, which is how this name usually trades.
• Upstream driver is hyperscaler capex disclosure: BofA's $1.2T framing (2026-08-03) is the number the whole group is being marked against; a capex-digestion comment from any top-3 buyer resets the multiple for the cluster, not for one name.
• Macro sensitivity runs through the rate path — at ~94.8x trailing earnings versus a 57.2 sector median (2026-08-04), CRDO de-rates before lower-multiple semis on a hawkish repricing.
• Policy correlation is new since 2026-08-05: restrictions on Chinese AI components were framed as an AVGO/MRVL/NVDA problem, and CRDO's supply chain and end demand both sit downstream of the same rules.