Financebotresearch desk研究台

Discovery › screen

Screen gaming adtech consumer internet

Date: 2026-07-29 · Agent: Scout / Screener · Framework: analysis_notes.md §0, §1, §2, §3

Mandate: first-ever screen of consumer-facing software. 44 prior screens, ~15 on software — all enterprise/B2B. This lane is structurally different: IP libraries, recurring bookings, network effects, UGC platforms rather than enterprise seat licenses.

Thesis under test: both lanes were repriced on an "AI will disrupt this" narrative. The agency has already watched that exact discount unwind violently elsewhere (ACN +22%, PYPL +24%, TRI +20%, SAP +20%, EPAM +19%, CTSH +17% in 14 days). Is the same mispricing still unharvested here?

Headline finding — the thesis is half right, and the half that's wrong matters more

The AI-disruption discount in gaming is largely a false positive. Google's Project Genie (announced 30 Jan 2026 — generates playable 60-second interactive worlds from text prompts) did trigger a textbook narrative selloff: TTWO −9%, RBLX −12%, Unity −30% in a single session (GameSpot, Bloomberg, 30 Jan 2026).

But when I traced where the large drawdowns actually came from, the AI narrative is not what's holding these names down. The damage is fundamental and it is specific:

Name Off 52wk high Market says "AI" What actually happened
RBLX −68% Project Genie Age-verification rollout. Lost 12M DAUs (144M→132M). FY26 bookings growth guidance cut 22–26% → 8–12%. FCF guided down 6–22% YoY. (Qz, Reclaim, 30 Apr 2026)
NTDOY −53% Project Genie Memory-chip cost squeeze + Switch 2 cycle. FY27 net profit guided −27% to ¥310B; ~¥100B hit from DRAM prices driven by the AI datacenter buildout and tariffs. Switch 2 units guided 19.86M → 16.5M. (CNBC, 8 May 2026)
EA at high — Not investable. $210/sh take-private (PIF/Silver Lake/Affinity, $55B). Trades $209.49 = 0.24% merger spread, stuck at CFIUS, outside date 28 Sep 2026. This is merger arb, not a fundamental position. (EA IR)

There's a genuine irony worth naming: the clearest way AI is actually damaging a gaming company right now is Nintendo's bill of materials, not its content pipeline. AI datacenter demand bid up memory prices and squeezed console margins. That is the opposite of the disruption story being sold.

The cleanest surviving AI-narrative mispricing is in adtech, not gaming — but adtech's moats are genuinely thin, exactly as suspected, so the discount is only partly unearned.

Top 3

1. NTES — NetEase, Inc. · Gaming (publisher/live-service) · $84.4B cap · $131.85 · −17% off 52wk high

Why it's a gem. The single best business in this entire screen by a wide margin — 41% operating margin, 22% ROE, 25% FCF CAGR, net cash, a growing dividend, and a shrinking share count — trading at 12.5x forward earnings and a 8.1% FCF yield against a global gaming peer average near 26x. Critically, the discount here is a China discount, not an AI discount — which means it is the one name in this screen where I am confident the cheapness is not explained by the thing the market is currently panicking about.

Metric Value Note
P/E ttm / fwd 16.9 / 12.5 vs ~26x industry (SimplyWallSt)
FCF (FY25) ¥48.69B (~$6.8B) FCF yield 8.1%
FCF CAGR (3y) +24.8% OCF +22.3%
Revenue CAGR (3y) +5.3% Net income CAGR +18.4% — margin expansion, not volume
Gross / Op margin 66% / 41% Op margin ¥19.63B→¥35.83B over 3y
ROE / ROA 22% / 11%
Debt/Assets 3.0% Cash ¥47.2B vs debt ¥6.7B; true net cash higher (excludes time deposits)
Share count −0.8%/yr 3.22B → 3.19B
Dividend 2.33%, payout 40% ¥6.72B→¥13.83B over 3y — +27% CAGR

⚠️ Data flags: fin.py reports EV ¥259.22B and P/B 17.3 — both are ADR currency-unit bugs (statements CNY, cap USD; 1 ADR = 5 ordinary). Corrected: real P/B ≈ 3.8x, EV ≈ $79B, EV/EBIT ≈ 16x. Note EV/EBIT is only fair, not cheap — the cheapness lives in P/FCF (12.4x) and P/E, because deferred revenue and low capex make FCF run well ahead of accounting profit. Buybacks collapsed from ¥8.83B (2024) to ¥0.64B (2025) — cash is being routed to dividends instead. Worth a question.

Moat — intangibles (IP) + switching costs, genuinely strong. Fantasy Westward Journey and Westward Journey Online are 20-year-old live-service franchises with multi-generational player communities, in-game social graphs and guild structures — the switching cost is social, not technical, which is the durable kind. Layered on top: exclusive China distribution and a global slate (Marvel Rivals, Naraka: Bladepoint, Destiny: Rising, Fragpunk), where the Marvel and Bungie licences are contractual intangibles a rival cannot replicate. §2 adversarial test: a well-funded rival can build a better-looking shooter; it cannot manufacture a 20-year guild community or licence Marvel twice.

The AI question — specifically. Project Genie generates environments. NetEase's economics come from live-service operations: seasonal content cadence, event balancing, monetisation tuning, and community management against a 20-year social graph. None of that is what a world-generator produces. More concretely, NetEase is an AI adopter, not a victim — CEO William Ding called AI "a core competence for development and operations," with proprietary CodeMaker and DreamMaker tools deployed across art, design, programming, animation and QA (Q4 2025 call). Cheaper content production raises the value of the scarce input NetEase already owns — distribution, licences, and installed communities. This is the bull side of the gaming AI debate, and NetEase is the cleanest expression of it.

Bear case (adversarial). (1) China is the whole risk — VIE structure, arbitrary regulatory intervention (the 2021 minors' playtime rules and the Dec-2023 draft monetisation rules both cratered the sector overnight), delisting tail risk, and soft domestic consumer spending. A 12.5x multiple on a Chinese asset may be a permanent structural rating, not a temporary discount — in which case there is no re-rating to harvest and you collect an 8% FCF yield forever. (2) Revenue CAGR is only 5.3%; earnings growth came from margin expansion, which mean-reverts and cannot repeat indefinitely. (3) The buyback stopped. (4) Global expansion is unproven at scale — Marvel Rivals is a hit, but hit-driven, and Marvel is a licence that can be lost.

Valuation (§3). DDM/DYT weak (short dividend history). Graham IV ≈ $78.5 vs $131.85 — but Graham systematically penalises asset-light, high-ROE businesses with modest book value; I weight it low here and flag it rather than lean on it. Weighting FCF-based: at a still-discounted 15x P/FCF → ~$160; at 12x → $128. Fair value range $135–$170. Entry zone <$125. Conviction 7/10.

2. DV — DoubleVerify Holdings · Adtech (ad verification) · $1.69B cap · $11.02 · −34% off 52wk high

Why it's a gem. The purest AI-narrative casualty I found that still has real cash economics: 10.2% FCF yield, 82% gross margin, net cash, and — rare in adtech — a genuinely shrinking share count. DV is the only small-cap adtech name in this screen buying back more stock than it issues; peers IAS and MGNI both dilute ~5%/yr. It is being priced as a melting ice cube while still growing and converting.

Metric Value Note
P/E ttm / fwd 33.4 / 9.0 ⚠️ fwd is non-GAAP — do not lean on it
EV / EV|EBITDA ~$1.53B / 12.6 Net cash $159M (cash $259M − debt $100M)
FCF (FY25) $172.7M FCF yield 10.2%; EV/FCF 8.9x
FCF CAGR (3y) +46.5% $54.9M → $172.7M
Revenue CAGR (3y) +18.3% ⚠️ FY26 guide 8–10% ($810–826M) — sharp decel
Gross margin 82% Adj EBITDA margin 34% guided
Share count −0.8%/yr 170.8M → 166.7M diluted. Buybacks $142M FY25, $100M YTD FY26
SBC $102–107M = ~12.6% of revenue High — but grants cut >40% vs 2025
Debt/Assets 7.4% Current ratio 4.77
ROE (GAAP) 5% Depressed by SBC + acquired-intangible amortisation

⚠️ The real problem, stated plainly: GAAP operating income has gone backwards — $85.7M (2023) → $82.4M (2024) → $79.2M (2025) — while revenue rose $573M → $748M. R&D went $125M → $178M. Diluted EPS fell $0.41 → $0.32 → $0.30. Revenue growth is not reaching the bottom line. FCF grows because SBC and amortisation are non-cash, but that is a real economic cost to me as an owner. This is the single most important number in the case and it cuts against the bulls.

Moat — thin, and I want to be explicit about it. Nominal sources: switching costs (verification tags embedded in ad workflows, MRC accreditation as a regulatory-ish barrier) and modest scale/data advantages. §2 adversarial test fails partially: the well-funded rival already exists and is the customer's own platform. Meta, Google, TikTok and Amazon can offer native brand-safety measurement for free inside their walls; DV's leverage exists mainly on the open web. This is not speculation — a shareholder derivative suit filed 9 Dec 2025 alleges DV concealed exactly this: that spend was shifting from open exchanges to closed platforms where DV's capability is "severely limited" (PPC Land). The bear case here is not hypothetical; it is litigated and partly conceded by the price. Real moat: narrow. Call it a toll booth with a shrinking road, offset by the fact that advertisers structurally want a neutral third-party grader — you don't let the player mark their own exam. That neutrality is the only durable moat DV has, and it is a real one.

The AI question — specifically. Two-sided, and unusually so. Bear: AI-native platforms grade their own inventory, and if walled gardens absorb ad spend, DV's addressable surface shrinks structurally. Bull: generative AI is creating DV's next market — an internet flooding with AI-generated junk inventory ("AI slop") makes third-party quality grading more valuable, not less. DV has shipped exactly this: AI SlopStopper, extended to social, and DV Authentic AdVantage extended to Meta and TikTok on 22 Jun 2026 — i.e. DV is getting inside the walled gardens rather than being locked out. ⚠️ Those pilot results are self-reported; treat with scepticism. My read: AI is net-neutral-to-positive on demand for DV's product, and net-negative on pricing power. The market is pricing only the negative.

Bear case (adversarial). Growth decelerating 18% → 8–10% is the tell that the walled-garden encroachment is already biting, not a future risk. SBC at 12.6% of revenue means reported FCF overstates owner earnings by roughly a third. GAAP operating income has been flat-to-down for three years. If verification becomes a commoditised checkbox, an 82% gross margin is indefensible and the terminal multiple is far below today's. A cheap-looking multiple on structurally decaying unit economics is §0's value trap quadrant, and I cannot fully rule it out.

Valuation (§3). Graham IV $7.18 vs $11.02 — Graham says expensive, driven by weak GAAP EPS; for an asset-light name I weight it moderately, and here it is a legitimate warning rather than a quirk. FCF-based: at 8–10% owner FCF yield after haircutting SBC (~$70–110M true owner FCF) → $8–13. Fair value range $10–14. Entry zone <$9.50 (demand a real margin of safety given the structural threat). Conviction 5.5/10.

3. CRTO — Criteo S.A. · Adtech (retail media / performance) · $1.10B cap · $21.80 · −14% off 52wk high

Why it's a gem — or the trap. The most extreme statistical value in the screen: ~19% FCF yield, 4.65x EV/FCF, 0.96x book, net cash equal to 18.5% of market cap, and management retiring 3.7% of shares a year. Criteo is being priced for terminal decline while generating $208M of annual free cash on a $1.1B market cap. I am including this as a top-3 finding but grading it as a coin-flip, and the reason is below.

Metric Value Note
P/E ttm / fwd 10.2 / 4.25 ⚠️ fwd is non-GAAP
EV / EV|EBITDA $970M / 3.31 Net cash $204M = 18.5% of cap
FCF (FY25) $208.5M FCF yield 19.0%; EV/FCF 4.65x
FCF CAGR (3y) +2.8% $192M → $208M — flat, not growing
Revenue CAGR (3y) −1.2% $2.02B → $1.94B — shrinking
Net income CAGR (3y) +152.8% ⚠️ artifact of a tiny 2022 base ($8.95M) — ignore
Gross margin 54% Adj EBITDA margin guided 32–34%
Share count −3.7%/yr 57.26M → 51.15M. Buybacks $152M = 13.8% of market cap
Debt/Assets 6.3% Very clean balance sheet
ROE 10%
Graham IV $32.92 vs $21.80 −34% to intrinsic — the one name where Graham says cheap

Moat — weakest in the screen. Essentially none, and the market just proved it. §2 adversarial test fails outright. The evidence is brutal and recent: two retail-media clients reduced scope, creating a $27M single-quarter headwind and a $75M FY26 headwind, driving Retail Media contribution down 31% and erasing a full year of growth (PPC Land, StockTitan, Q1 2026). A business where two customers can delete a year of growth does not have a moat — it has a client list. Legacy retargeting (the cash cow funding everything) is in secular decline as cookies and third-party identity erode. That is the honest §2 verdict.

The counter-evidence, which is real: stripping those two clients, the underlying retail media base grew 24% in Q1 2026. So the engine works; the customer concentration is what broke. Management guides FY26 contribution ex-TAC down low-single-digits, but up mid-single-digits excluding the $75M client headwind, with 32–34% adj EBITDA margins held via cost control. Criteo is also redomiciling to Luxembourg specifically to loosen buyback constraints — a shareholder-friendly signal, with $190M still authorised.

The AI question — specifically. Genuinely negative here, more than for DV. Criteo's historical edge was a shopper-intent dataset built on third-party cookies and a bidding engine — precisely the kind of pattern-matching that commoditises fastest as AI capability diffuses, and precisely the identity substrate that is disappearing. Its answer is "agentic AI" commerce media, which as of now is a slide, not a revenue line. Retail media is also a battleground Criteo does not control — Amazon owns the largest retail media network outright, and large retailers are increasingly capable of in-housing the exact tech Criteo rents them. The two client "scope reductions" are most plausibly read as early in-housing, which would make this a structural bleed rather than a one-off.

Bear case (adversarial). This is a melting ice cube that management is liquidating elegantly via buybacks. Revenue has declined three straight years. The moat is nil, the largest customers are also the most credible future competitors, and the growth pivot just took a 31% hit from two phone calls. §0's warning applies with full force: poor business + cheap price = value trap. The 19% FCF yield is only worth something if the cash flows persist ~5+ years; if contribution ex-TAC compounds down mid-single-digits, today's multiple is roughly fair and there is no re-rating.

Bull rebuttal, weighted honestly. At 4.65x EV/FCF you are paid for a lot of decay. Retiring 13.8% of the market cap annually means per-share FCF can grow even with revenue shrinking — this is the Bogle "growth comes in many forms" case, and §0 explicitly warns against fixating on revenue growth alone. The whole thesis is: does per-share FCF hold flat while share count falls 4%/yr? If yes, this is genuinely mispriced. If the two-client loss is the leading edge of in-housing, it isn't.

Valuation (§3). Graham IV $32.92 vs $21.80 (Graham weighted higher here — this is an asset-backed value play with real book and net cash, exactly its intended use case). At 7–8x EV/FCF on modestly declining FCF → $26–31. Liquidation-ish floor: net cash + 4x FCF ≈ $20. Fair value range $24–32. Entry zone <$20. Conviction 4.5/10 — genuinely a coin-flip, sized small if at all.

What I rejected, and why

Name Verdict Reason
RBLX ❌ Reject The −68% is earned, not narrative. 12M DAUs lost to age verification; bookings growth cut 22–26%→8–12%; FCF guided down. Still 6.5x EV/Rev, negative GAAP margins, and +5.0%/yr dilution. Great asset, deteriorating trajectory, not cheap. The UGC network effect is real — revisit if bookings re-accelerate and dilution stops.
NTDOY ⚠️ Watch — best moat, wrong price Finest IP intangibles in the industry (Mario/Zelda/Pokémon), ~¥1.79T net cash, 0.13 beta. But on FY27 guided net profit of ¥310B (−27%), P/E ≈ 26x, EV/FCF ≈ 24x. −53% off the high and still not cheap. §0: great company, bad entry. Memory-cost squeeze is cyclical — this becomes very interesting if DRAM normalises and the price doesn't. ⚠️ fin.py share-count data is internally inconsistent (1.15B out vs 4.65B diluted).
EA ❌ Not investable Merger arb at 0.24% spread.
TTWO ❌ Reject Only −7.8% off high, EV/EBITDA 59x, negative ttm EPS, D/E 84. Entire 2026 sector bull case rests on GTA VI — maximum release-slate concentration at a maximum multiple. Fails §0 badly.
PLTK ❌ Reject 9.9% yield on negative book value, payout 1.67x, negative op margin, float only 59M of 380M shares (85% insider). Illiquid, levered, declining. Yield trap.
SE ⚠️ Watch Genuinely strong (+47% rev growth, −47% off high) but 20x fwd / 24x EV/EBITDA isn't value yet.
MGNI ❌ Reject The CTV story is right, the security isn't: +5.0%/yr dilution, FCF CAGR +0.8% (flat), Debt/Assets 20%, EV/EBITDA 21. FY25 net income flattered by a tax item (op inc $98M vs net $145M).
IAS ❌ Reject vs DV Same business, worse execution: +4.9%/yr dilution vs DV's −0.8%, FCF declining ($97.9M→$77.4M), no buyback. ⚠️ Yahoo annuals stale (through 2024 only).
PUBM ❌ Reject Revenue −2%, negative operating margin, $593M cap. Subscale in a scale business.
EBAY / MTCH ❌ Reject Both near 52wk highs. EBAY 26x ttm / 18x EV/EBITDA. MTCH optically cheap (9.4x fwd) but +4% revenue with declining payers — a different value trap, and not a drawdown play.

Lane verdicts

🎮 Gaming — the AI discount is mostly a false positive. One name worth owning, and it's cheap for a different reason.

The Project Genie panic was real and the "durable IP owners benefit from cheaper content production" bull case is, I think, correct — content abundance raises the value of scarce distribution, licences, and installed communities. But the market has already worked this out. The names still deeply depressed (RBLX, NTDOY) are depressed by age-verification regulation and memory-chip inflation respectively, not by AI content fear, and both drawdowns are fundamentally justified. There is no unharvested AI-narrative discount left in gaming that also clears §0.

The exception is NTES, and I want to be precise about why: it is cheap because it is Chinese, not because of AI. That's a real risk and possibly a permanent rating — but it is a risk the user can actually underwrite and size, and it is decoupled from the AI question entirely. Verdict: one qualified gem (NTES), for reasons other than the screen's original thesis.

📊 Adtech — the AI discount is real and partly unearned, but the moats are as thin as feared. Trade, don't marry.

Here the narrative discount genuinely persists. But the brief's suspicion is confirmed by the evidence: adtech moats are weak and often illusory. Criteo lost a year of growth to two client decisions. DV's central vulnerability — walled gardens absorbing spend where its tech is limited — is documented in active litigation. Neither has a moat I would underwrite for ten years.

The nuance that keeps this lane alive: an internet flooding with AI-generated inventory increases demand for neutral third-party grading, and advertisers structurally will not let platforms mark their own homework. That is DV's one durable advantage. Verdict: DV is a legitimate mispricing with a capped holding period — own the re-rating, not the terminal value. CRTO is a genuine coin-flip. Neither is a compounder.

Cross-lane note on §0

Applying the framework honestly, this screen produces one 7/10 and two sub-6 names. That is a thin harvest from a 200+ name universe, and it should be read as a signal: cheap consumer internet is frequently cheap for real structural reasons, and this screen mostly confirmed that rather than overturning it. I have deliberately not manufactured a third high-conviction pick.

🚩 Most worth a full /analyze

NTES — NetEase

It is the only name in this screen that is simultaneously a great business (41% op margin, 22% ROE, 25% FCF CAGR, net cash, growing dividend, shrinking share count) and priced below its peer group (12.5x fwd vs ~26x) — the only candidate for §0's top-left "Value" quadrant. It also fills two documented portfolio gaps at once: consumer-facing software, and the 🌏 Structural-Risk sleeve. A full /analyze should focus on: (1) correcting the ADR unit errors across all valuation multiples; (2) why buybacks collapsed from ¥8.83B to ¥0.64B; (3) VIE/delisting structural risk and China regulatory precedent; (4) the durability of the global slate (Marvel Rivals licence terms, hit-concentration); (5) whether the 12.5x multiple is a temporary discount or a permanent structural rating — which is the entire thesis.

Secondary: DV merits a /value (not a full /analyze) to stress-test whether SBC-adjusted owner earnings support the price. NTDOY merits a watchlist entry with a price trigger — it is the best business in gaming and only needs a better entry.

Data gaps & caveats

  • ⚠️ ADR currency-unit bugs confirmed in fin.py for NTES (EV ¥259B, P/B 17.3) and NTDOY (EV −¥2.16T). All ADR EV/EBITDA and P/B figures in this screen were recomputed manually from market cap + debt − cash and flagged inline.
  • ⚠️ Forward P/E figures are non-GAAP for CRTO (4.25) and DV (8.97) and diverge wildly from GAAP. I did not weight them.
  • ⚠️ IAS Yahoo annuals only run through FY2024.
  • ⚠️ NTDOY share count internally inconsistent (1.15B out vs 4.65B diluted) — per-share figures unreliable; roic.ai cross-check recommended before any action.
  • ⚠️ CPCAY resolved to Cathay Pacific Airways, not Capcom (correct ticker: CCOEY). Capcom was not screened — a gap. Square Enix, Konami, Bandai Namco, CD Projekt, Embracer, Krafton and Nexon were also not screened (limited/unreliable US-listed data).
  • Bookings data for RBLX taken from company disclosure; all other gaming names assessed on GAAP revenue, which lags bookings for live-service titles — a known limitation of this pass.
  • Insider-transaction data not pulled for any name; recommend --holders before acting on NTES or DV.

Sources