Best Long-Term AI Stocks for a Recession: Top Picks

Planning beats panic: AI stocks when the economy slows

Planning beats panic. Every cycle. The investor who jots down a two-page playbook before the storm, trim some froth, add to the durable AI names, keep cash for ugly days, usually ends up compounding right through the mess. The investor who refreshes headlines every ten minutes? Sells winners into the hole and buys them back higher. I’ve done both in my 20+ years on the Street. One feels smarter in the moment; the other actually makes money.

Here’s the frame for Q3 2025: stop trying to timestamp a recession. Time the process, not the point. A written plan beats whatever the market is yelling about this week. Why? Because headlines are loud, but cash flows are quieter, and way more useful. In practical terms, “long-term” means 5-10 years, not 5-10 weeks. If you own AI exposure you’d be proud to hold through 2027, or 2030, your decisions get simpler when volatility spikes.

Quick reminder from history: recessions tend to hit in layers, cyclical revenue first, speculation second, quality last.

• Cyclical revenue first: ad budgets, discretionary gadgets, and capex for non-essential projects get cut within a few quarters of slowdown. In the 2009 downturn, Gartner reported worldwide IT spending fell about 4.6% (2009 data). SEMI tracked semiconductor equipment revenue down roughly 46% in 2009 versus 2008, violent, fast, and merciless.
• Speculation second: concept stocks and zero-profit moonshots lag as funding tightens. That pattern showed up in 2001-2002 and again in 2022 when unprofitable tech widely underperformed.
• Quality last: the cash generators and mission-critical vendors usually wobble, then win. During the 2020 recession, global ad spend fell 4.2% (MAGNA, 2020), yet cloud infrastructure revenues kept compounding, AWS grew 29% in 2020 to $45.4B (company filings, 2020). Mission-critical tends to be stickier than nice-to-have.

And that’s the key with AI: the infrastructure layer, GPUs, networking, power, data center REITs, cloud platforms, MLOps/security, tends to be stickier than consumer-facing AI apps that can be turned off with a fickle budget meeting. Infrastructure looks like contracted capacity, multi-year buildouts, and SLAs. Consumer apps look like monthly churn and promotional credits. Same theme, different durability.

Important caveat, I’m not saying infra is immune. Nothing is. In 2008-2009 the S&P 500 fell 57% peak-to-trough, and even best-in-class names drew down. But the recovery path differs. Capacity you booked and racks you installed often keep earning the day after the panic fades. The app you only liked on Tuesdays? Not so much.

So what does a calm playbook for AI in a 2025 slowdown look like? I’d keep it painfully simple:

1. Define “long-term” right now: 5-10 years. If you wouldn’t be comfortable holding your core AI names to 2030, they’re not core. This is the whole ballgame.
2. Pre-commit trims and adds: set levels to shave high-multiple, story-heavy positions, and separate levels to add to durable AI infrastructure and profitable platform names. Write it down. Prices make us emotional; paper keeps us honest.
3. Prioritize resilience metrics: recurring revenue, backlog/commitments, gross margin stability, free cash flow, and customer concentration. In a pinch, I’ll pay a fair price for boring cash flow over a “maybe” at a discount.
4. Keep dry powder: even 5-10% cash can turn drawdowns into opportunity. It’s a mental edge as much as a financial one.

Now, I get excited about this part, honestly, because it’s where compounding sneaks in. The investor who trims froth and rotates into the best-long-term-ai-stocks-for-a-recession, meaning the durable, cash-generating, infrastructure-adjacent names, often exits the downturn with more shares of better businesses. Same capital, better assets. The investor who panics sells the winners, then pays up later for the same winners. It’s the same idea stated two ways: plan beats panic; planning beats panicking.

And yes, market conditions this year are noisy. Growth data is mixed, services inflation is sticky in spots, and rate-cut odds keep whipsawing week to week. AI heavyweights are still talking multi-year capacity adds announced earlier this year, even if shipment timing jiggles around quarter to quarter. You don’t need perfect timing, you need a ruleset that keeps you from doing the obviously dumb thing at the worst possible time.

One last nudge. You don’t have to guess the exact month the economy rolls over. You just have to decide, before it does, which AI exposure you’d be thrilled to own on the other side. Infrastructure that customers can’t unplug easily tends to be stickier than shiny AI apps they can live without. Write that sentence down. Then write your plan around it.

What actually holds up: business traits of recession-resilient AI companies

What actually holds up: business traits of recession‑resilient AI companies

Here’s the shortlist I use when I’m trying to separate durable AI names from the stuff that’ll get repriced the second budgets tighten. It’s intentionally boring. It’s also repeatable and you can test it right in 10‑Ks, 10‑Qs, and investor decks without squinting.

1. Recurring revenue with multi‑year contracts and low churn beats usage‑only models. Read the revenue note and the KPIs. Security and EDA vendors have been saying this for years. CrowdStrike reported dollar‑based net retention of 119% in FY2024 and gross retention around 98% (company filings), which basically means customers expand more than they shrink even when macro gets weird. In chip design, Synopsys has consistently disclosed that the bulk of revenue is time‑based, multi‑year, and recurring (over 85% in FY2024 per its 10‑K). Cadence said roughly 95% of revenue was recurring in 2024 (company filings). Usage‑only AI tools can work, but they wobble when customers improve spend. Subscriptions with term commitments don’t wobble as much.
2. Net cash (or modest use) + positive free cash flow = flexibility. You’ll see it in the balance sheet and cash flow statement. CrowdStrike ended FY2024 with a net cash position and generated roughly 30%+ free cash flow margin for the year (filings). Zscaler also carried net cash and posted strong FCF conversion in FY2024. That flexibility matters when credit spreads gap out mid‑cycle. They can keep hiring engineers, not just issue press releases. Simple rule I use: if FCF is positive through a soft patch and share count isn’t ballooning, the model is probably real.
3. Mission‑critical spend is harder to cut mid‑cycle. Security, core cloud, and chip design tools live higher in the priority stack than shiny AI add‑ons. CFOs don’t cancel endpoint security or the EDA software that tapes out next year’s chips. That shows up in backlog too: Synopsys reported double‑digit billions of remaining performance obligations (RPO) in 2024, locked in over multi‑year terms (10‑K). Datadog, while more usage‑linked, still maintained 80%+ gross margins from 2023 through 2024 and positive FCF, which helped customers’ consolidation trend work in their favor when budgets were tight.
4. Diversified end markets reduce single‑customer or single‑sector risk. Check the 10‑K concentration footnote. Microsoft and Adobe benefit from sell‑to‑everyone models. On the infra side, hyperscalers are more concentrated by category but diversified across industries. Contrast that with vendors leaning on one mega customer or one vertical, those break when that vertical sneezes. If any customer is >10% of revenue, I mentally haircut the multiple.
5. Pricing power shows up in gross margin stability during slowdowns. Don’t overthink it, plot gross margin by quarter. Adobe kept gross margin in the high‑80% range across 2020-2024 (filings), despite currency swings and enterprise deal pushes. Datadog’s gross margin held roughly 81-83% during 2023-2024 while growth decelerated, which says value > price cuts. If an AI vendor’s GM compresses 400-600 bps the minute growth cools, that’s not pricing power; that’s discounting.
6. Sane dilution policies (SBC under control) and buyback capacity help per‑share compounding. This one gets ignored in hype cycles and hurts later. You can literally compute SBC as a % of revenue and see if the share count is creeping. Adobe kept SBC to single‑digit % of revenue in 2023-2024 and repurchased shares (filings). Synopsys’ share count has been stable to down with buybacks funded by FCF. On the flip side, several AI‑adjacent SaaS names ran SBC at high‑teens to ~20% of revenue in 2024; the business can be fine, but per‑share math lags when the stock is the currency.

If you want a working filter you can run this afternoon, use this:

• Revenue mix: Recurring ≥80%, term ≥24 months, NRR ≥110% (source: revenue recognition note, KPI section).
• Balance sheet/cash: Net cash or ND/EBITDA ≤1x; FCF margin ≥15% last 12 months (cash flow statement).
• Durability: RPO growth ≥ revenue growth, contract terms multi‑year (RPO disclosure).
• Diversification: No single customer ≥10% of revenue; top‑10 customers ≤30% combined (concentration footnote).
• Margins: Gross margin down ≤150 bps y/y in a slower growth quarter (P&L by quarter).
• Dilution: SBC ≤10% of revenue and basic share count CAGR ≤2%; active buyback optionality (equity comp note + repurchase authorization).

Quick sanity check against filings: CrowdStrike (FY2024) hits recurring + NRR + net cash + FCF. Synopsys/Cadence hit recurring + RPO + multi‑year. Adobe hits GM stability + SBC sanity + buybacks. That’s the pattern.

And yeah, 2025 has been choppy, rate‑cut odds flipping, IT budgets scrutinized, AI capex still green‑lit but with shipment timing slipping quarter to quarter. That’s exactly why this filter helps. When the music slows, the names with termed recurring, cash, pricing power, and reasonable dilution keep compounding. Everyone else does vibes.

Choose your layer: the durable AI stack (and where it can break)

Here’s how I bucket the AI stack into investable layers, with the specific recession tripwires I watch. And just to keep it real: I’m not handing you stock picks, these are examples investors know, not nudges.

• AI accelerators and systems (examples: NVDA, AMD): Scale leaders can stay sold‑out in tight cycles, but orders are lumpy and hyperscaler‑heavy. What breaks? Double ordering and channel inventory. If one cloud slows a deployment, shipments can slip a quarter, or three. 2025 capex commentary across MSFT/AMZN/GOOGL still implies $150B+ combined data center/AI outlays this year, but that money isn’t linear by quarter. Watch: backlog convert vs shipments, distributor inventory days, customer concentration by the top 3 clouds.
• Picks‑and‑shovels semicap & tools (examples: ASML, AMAT, KLAC, LRCX): Unavoidable for capacity builds, yet cyclical by definition. SEMI reported wafer fab equipment spending of about $100B in 2023, with forecasts for ~$124B in 2024 and ~$128B in 2025 (SEMI industry outlook). That up‑and‑to‑the‑right headline hides quarterly air pockets. In a slowdown, I prioritize high moat subsystems and service mix, not just tool shipments. Watch: installed base service growth, EUV/High‑NA mix, China exposure gates.
• EDA / chip design software (examples: SNPS, CDNS): Mission‑critical, high recurring, and an oligopoly. Renewals are multi‑year and sticky. Going off memory, the top two players are roughly two‑thirds of the EDA market; don’t quote me to the decimal, but the structure is the point. Recession risk is lower here; design cycles don’t pause easily. Watch: RPO growth vs billings, price lift on next‑gen nodes, and any wobble in start‑up ASIC demand.
• Cloud platforms with AI workloads (examples: MSFT Azure, AMZN AWS, GOOGL Cloud): Enterprise AI spend is sticky once models are in prod, but usage can slow in a budget squeeze. Track committed spend, not just on‑demand. If reserved instances and prepayments are rising while reported usage decelerates, customers are improve. 2024‑2025 reports showed steady AI‑related GPU capacity constraints alongside cloud optimization headwinds, both can be true at once. Watch: disclosed remaining performance obligations (RPO), AI service attach, and capex to revenue lag.
• AI‑driven cybersecurity (examples: CRWD, PANW): Breach risk doesn’t recess. IBM’s 2024 Cost of a Data Breach report pegged the average breach at about $4.88M, up from prior years, boards don’t ignore that in a downturn. The durable names tend to post net revenue retention around or above 120% in filings and keep margins expanding. Watch: NRR, large‑deal mix, and whether AI features are driving real module adoption or just marketing fluff.
• Networking & interconnect for AI data centers (examples: ANET, MRVL): This layer is tied directly to AI build‑outs, optics, switches, custom interconnect. Backlog visibility matters because orders can bunch ahead of new cluster turns. In a recession, customers may sequence by rack and delay leaf‑spine upgrades. Watch: book‑to‑bill, AI vs traditional networking mix, and customer shipment schedules (weeks matter here).
• Data infrastructure software (examples: MDB, SNOW): Durable if embedded in workflows and priced on consumption that scales with AI apps. But when macro tightens, consumption growth can drift lower for a couple quarters. Valuation also bites here, great tech, sure, but cash flows need to catch up. Watch: consumption per customer, multi‑cloud portability, backlog of committed capacity, and gross margin stability as AI features roll out.

Quick mental model for 2025’s choppy tape: the closer you are to AI capacity build (accelerators, semicap, networking), the more you live and die by shipment timing and a few massive buyers. The closer you are to design tools and security, the steadier the renewals, with less whipsaw. And yes, there are exceptions, I’ve sat through quarters where a single customer push‑out wrecked a beautiful setup. It happens.

Right now context: rates, capex, and budgets that matter in 2025

Quick grounding first: the U.S. policy backdrop hasn’t shifted as much as equity charts suggest. The federal funds target range sat at 5.25%-5.50% through much of 2024 per Federal Reserve communications, which kept discount rates elevated heading into 2025. That’s the opposite of the 2020-2021 setup when the target range was 0.00%-0.25%. Why it matters is simple, long-duration cash flows get hit harder. AI software with heavy back-end loaded payoffs trades more like a long bond; cash‑rich hardware incumbents with near-term earnings have a thicker cushion. I’m oversimplifying a bit, but you get the point.

Enterprise spending: what CIOs actually said last year matters more than the hot take on X this morning. Across 2024 CIO surveys, priorities clustered around security, cloud migration/modernization, and AI initiatives. In 2025, the filter is tighter: check guidance on (1) committed cloud spend, reserved instances, savings plans, and prepayments, and (2) AI pilot-to-production conversion rates. If management is still talking “pilots” without naming production workloads and budget owners by Q4, assume a slower revenue ramp. When macro tightens, consumption curves flatten for a couple of quarters; we saw that dynamic, yes, even in otherwise great platforms, earlier this cycle.

Hyperscaler capex: management teams flagged higher AI-related capex across 2023-2024 earnings transcripts. The mistake I see is treating capex dollars as the end of the story. In 2025, the question is who’s converting that spend into utility‑scale revenue, GPU hours, vector DB transactions, managed model serving, private networking bandwidth attached to AI clusters. Watch the mix: training-heavy spend is lumpy; inference tied to customer workloads is stickier. Also, tie back to our earlier point on shipment timing, weeks really matter here. A single delivery push can shift revenue recognition across quarters and whipsaw sentiment.

Supply chain tells: durable AI demand into 2026 will show up in lead times and commentary before it shows up in lagging KPIs. Track what companies print in 10‑Qs and say on earnings calls about advanced-node capacity, HBM memory availability, and networking gear (800G/1.6T). If lead times stabilize at elevated levels and allocations persist, the buildout has legs. If you hear “lead times normalizing” alongside higher channel inventories, that’s a caution flag for the next two quarters, even if headline capex sounds big.

Valuation sensitivity: with rates still elevated versus 2020-2021, the market’s paying more attention to duration. Long-duration AI software multiples are more rate‑sensitive; cash‑generative hardware and semicap names feel the floor from near-term cash flows. Don’t overthink it, just map each name’s cash flow timing to your cost of capital. If your model needs a 2027-2029 margin ramp to make it work, the fed funds reality from 2024 (5.25%-5.50%) still hangs over it in 2025. I wish that weren’t true, but it is.

What you can actually check this year: the Fed’s dot plot and statements (2025 path), company 10‑Qs for capex and backlog, hyperscaler disclosures on AI data center spend and expected depreciation lives, CIO survey updates for H2 budget priorities, and lead-time commentary across suppliers. If something doesn’t line up, capex up, but no capacity revenue, assume delay first, not conspiracy. And yeah, I’ve been burned before by assuming “imminent” meant this quarter. It didn’t.

Build the watchlist: tiers, guardrails, and entries you can actually execute

I’m a big believer in writing it down. In a recession window, fuzzy plans get expensive. Here’s how I turn the framework into something I can actually click, size, and survive with. Not perfect, just disciplined.

Tiering approach (for an AI sleeve aimed at the best long-term AI stocks for a recession):

• Core: durable moats, positive FCF today, clear pricing power. Think semicap with backlog visibility or infra software with high net retention. Target 60-70% of total AI exposure.
• Satellite: faster growth, some cyclicality. Positive unit economics, but more beta to cycles or budgets.
• Experimental: small, thesis-in-progress names. Strict risk caps, assume you’re early or wrong (because sometimes we are).

Valuation guardrails, yes, boring, but they save you:

• FCF yield vs. cost of capital: prefer FCF yield > your WACC. If your equity hurdle is ~10% and the name prints 12-15% forward FCF yield, green. If it’s 4-5%, that’s a rate bet, not a cash-flow bet.
• Software specifics: Rule-of-40 ≥ 40% on a forward basis; net dollar retention ≥ 115% for true AI tailwind, ≥ 120% is elite. Watch SBC < 20% of revenue if you can; high-SBC names belong in tax-advantaged accounts when possible.
• Hardware/semicap: track backlog growth vs. shipments, and gross margin and operating margin through the cycle. A simple band: GM should stay within ±200 bps across down quarters; op margin shouldn’t collapse below mid-teens if the moat is real.

Balance sheet check in a high-rate tape: prioritize net cash or interest coverage well above 5x (EBIT / interest). With the policy rate at 5.25-5.50% during 2024 (and still restrictive this year), refinancing risk isn’t theoretical. I’ve seen too many “great stories” forced to issue equity at the worst time because interest coverage slipped to 2-3x. Hard pass.

Position sizing that won’t blow up the portfolio:

• Core bucket = 60-70% of AI exposure. If your total portfolio is 60% equities and 15% of that is AI, then 9-10.5% sits in Core.
• Single-name caps: any high-beta or pre-profit name ≤ 2-3% of total portfolio. Non-negotiable when recession chatter gets loud.

Entry tactics, pick one before the tape gets emotional:

• Dollar-cost averaging: weekly or biweekly lots; widen the spacing if volatility spikes (ATR up, size down).
• Buy earnings dislocations: pre-set levels where you add on −10% to −20% gaps if the thesis items (retention, backlog, margin bands) are intact.
• Cash-secured puts: write at strike levels you’re happy to own; target annualized yields ≥ your equity hurdle (e.g., >10%). Use front-months around events; be ready to take assignment.

Tax placement that actually helps after-fee, after-tax returns:

• High-turnover, high-SBC software → tax-advantaged accounts when you have the space.
• In taxable, harvest losses on broken entries while avoiding the wash-sale rule (30-day window; switch to a close substitute, not a substantially identical name).

Checklist, not heroics: tier every name; confirm FCF yield vs. WACC; insist on Rule-of-40/retention or backlog/margin bands; net cash or >5x coverage; size Core at 60-70%; cap any high-beta at 2-3%; pre-commit entries (DCA, earnings gaps, or cash puts). When the recession tape gets choppy, the list does the thinking.

I’ll repeat myself because it matters: if your model needs a 2027-2029 margin ramp to work, assume a higher discount rate than your heart wants. I hate it too, but that discipline kept me from chasing “imminent” ramps that never arrived.

Okay, so what if you wait? The real cost of doing nothing

Short answer: you’ll likely end up buying strength and selling fear, again. I’ve watched this movie enough times. In 2020-2021, investors who skipped the cloud infrastructure build phase wound up paying nosebleed multiples for the same cash flows, BVP’s Cloud Index peaked around ~35x next‑twelve‑month revenue in late 2021, then reset near ~6-7x in 2022. If you pass on the build phase now and only show up when the earnings are “de‑risked,” you’re volunteering to pay peak multiples later for stuff you could’ve owned at reasonable forward FCF yields during the capex slog.

There’s also the cash trap. Sitting in T‑bills at ~5% feels smart while the Fed funds rate sits at 5.25-5.50% (as of September 2025). But across full cycles, it rarely wins. Since 2000, the S&P 500’s total return has compounded roughly 6-7% annually while 3‑month T‑bills earned closer to ~1.5-2% (Ibbotson/FRED long‑run data). And inflation doesn’t take a vacation just because you’re in cash. Even with headline CPI easing from the 2022 spike, the last couple years have bounced around that 3% neighborhood. Cash at 5% today sounds great; cash reinvested at 3% tomorrow with 2-3% inflation is just treading water. Quality compounders with pricing power and rising free cash flow generally beat that, over a full cycle. Not every quarter. But over the whole lap.

Meanwhile, competitors aren’t waiting for your comfort level. Enterprise AI that started as pilots in 2023-2024 is moving into production budgets in 2025. You can see it in spending footprints: the big platforms are telling you with their wallets. Based on public guidance and run‑rates this year, the hyperscaler/consumer internet group (Alphabet, Microsoft, Amazon, Meta) is pacing toward $200B+ of combined capex in 2025, with AI infrastructure a major line item. When your suppliers and your rivals scale at the same time, the window for great entries usually doesn’t stay open.

And yeah, I get the hesitation. Earnings wobbles will happen. We’re still in a rate regime where a hot payroll print can yank 10‑year yields 15-20 bps in a day and take 3-5% off the high‑beta AI cohort by lunch. But that’s the point: if you don’t set rules now, you default to chasing green candles when the headlines are clean and puking shares into a guidance reset. A simple playbook, tiers, sizing caps, pre‑committed add points, keeps you out of your own way. It’s boring. It works.

Let me say this more like we’re chatting over coffee: you don’t need to be a hero. Start small if you must, but start with a plan. Buy your Core names on a DCA, sell cash‑secured puts where you actually want the stock, and keep dry powder for earnings gap‑downs. Future‑you will thank present‑you the next time a CFO says “prudent pacing” and the stock trades down 12% on nothing structural.

The opportunity cost checklist if you sit it out:

• Paying up later: Missing the infrastructure phase usually means paying higher EV/Revenue or lower FCF yields once AI revenue is obvious.
• Cash drag: Today’s ~5% T‑bill yield can reset fast; over cycles, equities (6-7% since 2000) beat cash (~1.5-2%).
• Competitive gap: 2025 is when pilots become production; incumbents that adopt earlier lock in data moats and workflow stickiness.
• Behavior gap: No rules = buy strength, sell fear. Pre‑committing entries flips that.

One last nudge: if your thesis only works with a heroic 2027-2029 margin ramp, assume a tougher discount rate and size it smaller. But don’t confuse disciplined entry with doing nothing. In this AI build‑out cycle, inaction is a decision, and it has a price.

Frequently Asked Questions

Q: Should I worry about timing a recession before buying AI stocks?

A: Short answer: no. Time the process, not the point. Set a 5-10 year window, pick durable AI names (infrastructure first), and automate buys with monthly DCA. Keep 10-20% of your equity sleeve as dry powder for -15% or -30% dips. Rebalance with bands (say 5%). And cash for life? Keep 6-12 months expenses separate, untouched.

Q: How do I build a long-term AI portfolio for a slowdown?

A: Keep it boring-smart. Aim 60-70% in infrastructure: semis and tooling (think NVDA, TSMC, ASML, AVGO) plus cloud platforms (MSFT, AMZN, GOOGL). Put 20-30% in mission‑critical software/data (service management, security, data platforms). Cap “shiny apps” at 10-20%, they’re fun, but budget gets cut there first. Use ETFs (SMH/SOXX for semis; IGM/QQQ for broad tech) if you don’t want single‑stock risk. DCA monthly, set buy bands at -15%/-30%, position size 2-5% per name, and rebalance semiannually. No margin, funding dries up when you least want it.

Q: What’s the difference between AI infrastructure and AI applications in a recession?

A: Infrastructure is the picks‑and‑shovels layer: GPUs, networking, foundry capacity, and cloud compute. It’s tied to multi‑year capex plans and mission‑critical workloads. It wobbles with cycles, but customers rarely turn it off. Applications are what end‑users touch, new copilots, niche models, pilots. Those budgets are easier to pause. In 2020, AWS still grew 29% to $45.4B (company filings, 2020) while global ad spend fell 4.2% (MAGNA, 2020). That’s the pattern: infra and mission‑critical bend; nice‑to‑have snaps.

Q: Is it better to keep extra cash or stay fully invested in AI names during a recession scare?

A: Do both, on purpose. Separate buckets. 1) Safety cash: 6-12 months of expenses in high‑yield savings or T‑bills, this is non‑negotiable and not “investing cash.” 2) Portfolio cash: hold 10-20% of your equity allocation as dry powder when volatility is high. Then run rules. Example plan I use with clients: DCA 50-60% of intended AI exposure over 6-9 months. Pre‑set buy bands for the rest: add 25% of your dry powder at -15% from a 3‑month high, 35% at -25%, 40% at -35%. Rebalance when any position drifts +/-5% from target to avoid concentration creep (yes, even with the semis). Prefer infrastructure core (semis, foundry, cloud) over speculative apps until the economy re‑accelerates; history says quality wins last. Tax side: place higher‑volatility semis in tax‑advantaged accounts if you can, and be ready to harvest losses in taxable if 30‑day wash rules can be managed with close substitutes (ETF pairs work). Optional, not required: cash‑secured puts 5-10% OTM on names you want to own, sized small, expiry 30-60 days, paid patience. What not to do: use, concentrated options bets, or selling winners just because headlines are loud. I’ve done that mistake. Regretted it. The written plan beats the panic every time.

@article{best-long-term-ai-stocks-for-a-recession-top-picks,
    title   = {Best Long-Term AI Stocks for a Recession: Top Picks},
    author  = {Beeri Sparks},
    year    = {2025},
    journal = {Bankpointe},
    url     = {https://bankpointe.com/articles/best-ai-stocks-recession/}
}