Top 5 AI Stocks at Bargain Prices
Introduction
AI infrastructure spans five essential markets: processors, memory, chip manufacturing, custom chips and networking, and servers. We have selected NVIDIA, Micron, TSMC, Broadcom and Dell, one from each category. This report explains the value case for each stock, compares the closest alternatives and identifies what could weaken the investment case.
NVIDIA and Broadcom offer lower forward earnings multiples than the selected direct competitors. Micron and Dell are more nuanced choices: both have cheaper rivals, but other business and ownership factors support our preference. TSMC offers a more focused foundry investment than its broader competitors.
The five stocks at a glance
| Selected stock | AI infrastructure role | Forward P/E | Forecast fiscal year / ending |
|---|---|---|---|
| NVIDIA Corporation (NVDA) | AI processors | 10.3× | FY2028 / Jan 2028 |
| Micron Technology (MU) | DRAM and HBM memory | 4.4× | FY2027 / Sep 2027 |
| Taiwan Semiconductor Manufacturing Company (TSM) | Advanced chip manufacturing | 17.5× | FY2027 / Dec 2027 |
| Broadcom (AVGO) | Custom chips and networking | 17.9× | FY2027 / Oct 2027 |
| Dell Technologies (DELL) | AI servers | 11.3× | FY2028 / Jan 2028 |
Data snapshot: November 13, 2026 local-market closes, reviewed November 14. Forward P/E uses consensus adjusted earnings for the fiscal year shown; the forecast periods differ. Category numbers organize the article and do not rank expected returns.
Full assessment criteria, weights and scorecards appear after the stock analysis.
Investor disclaimer
This article is educational, informational research—not financial advice.
We strive to keep data accurate, but unavoidable errors or omissions may occur. Verify important information with primary sources.
1. NVIDIA (NVDA): our AI processor pick
The value case. NVIDIA trades at 10.3× estimated fiscal 2028 earnings, compared with AMD at 30.2× for fiscal 2027. The fiscal years end only one month apart, making this one of the closer-period comparisons in the report. Our preference combines that lower multiple with NVIDIA’s established accelerator platform.
| Company | Forward P/E | Forecast fiscal year / ending |
|---|---|---|
| NVIDIA | 10.3× | FY2028 / Jan 2028 |
| Advanced Micro Devices | 30.2× | FY2027 / Dec 2027 |
The closest alternative. AMD is the direct accelerator challenger. It has stronger reported cash conversion in this snapshot, but its substantially higher forward P/E demands more confidence in future execution. Both companies have net-cash positions.
What could undermine the bargain. NVIDIA’s November leadership transition adds uncertainty to the business outlook. Export restrictions, customer spending and earnings revisions could also weaken the case. Reported free cash flow was about 66% of net income, so collections and investment commitments deserve attention.
What to watch next: whether earnings expectations hold through the leadership transition and whether cash generation keeps pace with profits.
2. Micron (MU): our memory pick, with a cheaper challenger
The value case. Micron trades at 4.4× estimated fiscal 2027 earnings and supplies DRAM and high-bandwidth memory used in AI systems. Its net-cash position and direct common-stock ownership strengthen its overall case. The major assumption is that forecast memory profits remain sustainable.
| Company | Forward P/E | Forecast fiscal year / ending |
|---|---|---|
| Micron | 4.4× | FY2027 / Sep 2027 |
| SK hynix | 2.3× | FY2027 / Dec 2027 |
| Sandisk | 3.9× | FY2027 / Jun–Jul 2027 |
| Samsung Electronics | 2.7× | FY2027 / Dec 2027 |
The closest alternative. SK hynix is cheaper at 2.3× and has an established HBM business. Micron’s narrow preference reflects its reporting and ownership structure, not a lower price for forecast earnings. SK hynix also has a U.S. depositary-share listing, so it should not be dismissed on access alone.
Sandisk provides NAND/storage exposure rather than a comparable DRAM/HBM mix. Samsung offers memory exposure within a broader group. Both belong on the comparison list, but neither is an interchangeable substitute for Micron.
What could undermine the bargain. Memory profits are cyclical, and exceptionally strong pricing can make forward multiples look unusually low. Our assessment has not estimated mid-cycle earnings. The scoring bands also put Micron and SK hynix in the same valuation bucket, making this a particularly method-sensitive choice.
What to watch next: memory pricing, HBM demand and cash generation after capacity investment. SK hynix remains a credible alternative for readers who prioritize the larger valuation discount.
3. TSMC (TSM): our advanced manufacturing pick
The value case. TSMC trades at 17.5× estimated fiscal 2027 earnings and offers dedicated foundry exposure. It manufactures chips rather than selling the finished computing systems. Its established manufacturing business provides a clearer investment case than valuing a foundry inside a diversified group.
| Company | Forward P/E | Forecast fiscal year / ending |
|---|---|---|
| TSMC | 17.5× | FY2027 / Dec 2027 |
| Samsung Electronics | 2.7× | FY2027 / Dec 2027 |
| Intel | 33.3× | FY2027 / Dec 2027 |
The closest alternatives. Samsung has a much lower group forward P/E, but that figure includes memory, devices and other operations. It does not reveal what investors are paying for Samsung Foundry alone. Intel’s higher forward P/E and foundry turnaround requirements make it a different proposition.
Neither Samsung Foundry nor Intel Foundry is separately listed. Their parent-company multiples remain useful context, but cannot establish which foundry business is cheapest. TSMC is our preferred business exposure, not the lowest group multiple in the table.
What could undermine the bargain. Taiwan exposure, overseas manufacturing costs and the capital required for new capacity can reduce returns. Strong demand must produce durable profits after those investments. A successful competitor turnaround could also change the comparison.
What to watch next: earnings growth and returns on new manufacturing capacity, particularly as overseas production expands.
4. Broadcom (AVGO): our custom-chip and networking pick
The value case. Broadcom trades at 17.9× estimated fiscal 2027 earnings, below Marvell and Arista in this comparison. It offers custom-chip and connectivity exposure alongside an infrastructure software business. Reported free cash flow slightly exceeds net income, supporting the cash-quality assessment.
| Company | Forward P/E | Forecast fiscal year / ending |
|---|---|---|
| Broadcom | 17.9× | FY2027 / Oct 2027 |
| Marvell Technology | 21.3× | FY2028 / Jan 2028 |
| Arista Networks | 32.7× | FY2027 / Dec 2027 |
The closest alternatives. Marvell is the more direct custom-chip comparison, at 21.3× forward earnings. Arista supplies networking systems and has a stronger net-cash position, but trades at 32.7×. Arista ranks second in our combined-category scorecard; Marvell remains the closer product rival.
Broadcom’s software operations contribute to consolidated earnings and cash flow. Its group multiple therefore does not value AI chips alone. The attraction is the combined business at a lower forecast earnings multiple than these selected alternatives.
What could undermine the bargain. Custom-chip programs can depend on a small number of large customers and their deployment schedules. Delays or weaker margins could reduce forecast earnings. Adjustments excluded from consensus EPS also need scrutiny.
What to watch next: program delivery, customer concentration and whether cash generation continues to support reported profits.
5. Dell (DELL): our AI server pick
The value case. Dell trades at 11.3× estimated fiscal 2028 earnings and has separately disclosed substantial AI-server sales. Its latest cited quarter included $16.4 billion of AI-server revenue. We prefer the combination of that commercial evidence and its financial position over simply choosing the lowest multiple.
| Company | Forward P/E | Forecast fiscal year / ending |
|---|---|---|
| Dell | 11.3× | FY2028 / Jan 2028 |
| HPE | 10.2× | FY2027 / Oct 2027 |
| Supermicro | 4.1× | FY2027 / Jun 2027 |
The closest alternative. HPE is cheaper at 10.2× and has stronger reported cash conversion. Dell has lower consolidated debt/EBITDA—1.95× versus 2.82×—and higher interest coverage, at 9.60× versus 3.67×. Those financial differences and the business assessment support Dell’s selection.
Supermicro is cheaper still, but its negative twelve-month free cash flow raises concerns about converting shipments into cash. HPE remains the closest overall alternative. Both Dell and HPE have financing operations, which make consolidated leverage an imperfect comparison.
What could undermine the bargain. AI-server shipments must produce profitable revenue and timely collections. The selection also depends on our higher business rating for Dell; reducing that rating would allow HPE to win the tiebreaker. Dell is not the cheapest server stock in the table.
What to watch next: margins, customer collections and the funding required to deliver AI-server orders.
Our verdict: the trade-off behind each pick
| Pick | Why we selected it | Main trade-off |
|---|---|---|
| NVIDIA | Lower forward multiple than AMD; established accelerator platform | Leadership transition and earnings execution |
| Micron | DRAM/HBM exposure, net cash and simpler common-stock ownership | SK hynix is cheaper; the scoring margin is narrow |
| TSMC | Dedicated, established foundry business | Samsung has a cheaper group multiple; Taiwan and capacity-investment risks |
| Broadcom | Lower forward multiple than Marvell and Arista; strong cash conversion | Customer concentration and a mixed chip/software business |
| Dell | Disclosed AI-server scale and stronger consolidated leverage metrics than HPE | HPE is cheaper and converts more reported profit into cash |
NVIDIA, Micron, TSMC, Broadcom and Dell are our selections under the stated assessment. NVIDIA and Broadcom have clearer forward-multiple advantages over the selected direct peers. Micron, TSMC and Dell require readers to accept specific trade-offs against cheaper alternatives.
The common risk is a slowdown in AI infrastructure spending. Different categories still depend on overlapping customer budgets, while memory pricing can amplify the cycle. The research notes below explain the weights, scoring limitations and downside tests behind these choices.
Explore the research: weights, full scorecards and calculation inputs
Research notes: how we selected the stocks
Valuation receives the largest weight because this is a bargain-stock screen. AI business strength follows because the company must supply a relevant, competitive product. Cash conversion and financial strength test whether the business can support its earnings and investment needs.
| Criterion | Weight | What it measures |
|---|---|---|
| Valuation | 35% | Forward P/E for an explicitly stated fiscal year |
| AI business and competitive position | 30% | Established commercial products, competitive position and fit with the category |
| Cash conversion | 15% | Trailing free cash flow relative to reported net income |
| Balance-sheet strength | 15% | Net cash, debt/EBITDA, interest coverage and loss or cash-burn constraints |
| Ownership and reporting | 5% | Listing route, reporting framework and ownership complexity |
| Total | 100% | The same weights apply to every company and category |
Each criterion receives a score from 1 to 5. Total score = Σ(criterion score ÷ 5 × weight). For example, a valuation score of 4 contributes 28 of the possible 35 points.
These are editorial screening weights, not a backtested investment strategy. A score of 80 means 80 points under this model, not an 80% chance of success. Business ratings involve judgment; the financial scoring rules below make the other inputs reproducible.
The highest unrounded score determines the winner within each category. Exact ties go first to the higher business score, then the higher cash score. These selections are screening judgments, not measured discounts to intrinsic value.
Full company scorecards
The following tables preserve the complete assessment behind the five selections. Component scores run from 1 to 5; the final column applies the weights above. Scores are useful for tracing our judgments, not predicting investment returns.
AI processors
| Company | Value | Business | Cash | Balance | Reporting | Total /100 |
|---|---|---|---|---|---|---|
| NVIDIA | 4 | 5 | 3 | 5 | 5 | 87.0 |
| Advanced Micro Devices | 2 | 4 | 4 | 5 | 5 | 70.0 |
Memory
| Company | Value | Business | Cash | Balance | Reporting | Total /100 |
|---|---|---|---|---|---|---|
| Micron | 5 | 5 | 3 | 5 | 5 | 94.0 |
| SK hynix | 5 | 5 | 3 | 5 | 4 | 93.0 |
| Sandisk | 5 | 3 | 4 | 5 | 5 | 85.0 |
| Samsung Electronics | 5 | 3 | 4 | 5 | 3 | 83.0 |
Advanced manufacturing
| Company | Value | Business | Cash | Balance | Reporting | Total /100 |
|---|---|---|---|---|---|---|
| TSMC | 4 | 5 | 3 | 5 | 4 | 86.0 |
| Samsung Electronics | 5 | 2 | 4 | 5 | 3 | 77.0 |
| Intel | 2 | 2 | 2 | 2 | 5 | 43.0 |
Custom chips and networking
| Company | Value | Business | Cash | Balance | Reporting | Total /100 |
|---|---|---|---|---|---|---|
| Broadcom | 4 | 5 | 5 | 4 | 5 | 90.0 |
| Arista Networks | 2 | 5 | 5 | 5 | 5 | 79.0 |
| Marvell Technology | 3 | 4 | 3 | 3 | 5 | 68.0 |
AI servers
| Company | Value | Business | Cash | Balance | Reporting | Total /100 |
|---|---|---|---|---|---|---|
| Dell | 4 | 5 | 4 | 3 | 5 | 84.0 |
| HPE | 4 | 4 | 5 | 2 | 5 | 78.0 |
| Supermicro | 5 | 4 | 1 | 2 | 5 | 73.0 |
How the financial scores are calculated
Forward P/E determines the valuation score and receives the full 35% valuation weight. Cash generation remains a separate business-quality assessment, with a 15% weight. The bands below are editorial screening conventions, not industry standards or estimates of fair value.
| Score | Positive forward P/E | Cash conversion: trailing FCF ÷ net income |
|---|---|---|
| 5 | 10× or less | At least 100% |
| 4 | Above 10× through 20× | 75% to below 100% |
| 3 | Above 20× through 30× | 50% to below 75% |
| 2 | Above 30× through 50× | 25% to below 50% |
| 1 | Above 50× | Below 25%, or negative FCF |
Nonpositive forecast EPS receives 1 for valuation. Positive FCF with a reported net loss receives 2 for cash conversion. Genuinely missing forecast EPS would leave the total unscored; reported cash flow does not enter the valuation score.
Balance-sheet scoring starts at 5 for net cash. Otherwise, 4 requires debt/EBITDA of at most 1.5× and interest coverage of at least 10×. A score of 3 requires at most 2.5× and at least 5×; 2 requires at most 4× and at least 2×.
Other net-debt cases receive 1. A reported net loss or negative FCF caps the balance score at 2, including net-cash companies. These consolidated ratios do not separately value the financing operations at Dell or HPE.
How we rate the business and ownership structure
The business score evaluates the company’s role in the specific category. It combines commercial evidence with an editorial judgment about competitive strength. It is not an estimate of the percentage of revenue attributable to AI.
| Business score | Assessment |
|---|---|
| 5 | Established category leader or specialist with strong commercial evidence |
| 4 | Established direct supplier with competitive, execution or disclosure limitations |
| 3 | Relevant but adjacent product exposure or a substantially broader business mix |
| 2 | Turnaround economics or weak fit between group valuation and the selected business |
| 1 | Little demonstrated commercial exposure to the category |
For ownership/reporting, domestic SEC reporting with common shares receives 5; a U.S. ADS route with foreign-issuer reporting receives 4. The Korean ordinary-share route used for Samsung receives 3. These scores measure structural simplicity, not national regulatory quality or management integrity.
Samsung has the same financial scores in both categories. Its business score changes from 3 in memory to 2 in manufacturing because its group earnings do not isolate foundry economics. Every other criterion and weight remains unchanged.
Forward P/E: the earnings period behind each comparison
The snapshot uses November 13, 2026 regular-session closes, reviewed November 14. U.S. closes are at 4 p.m. Eastern Time; Korean closes are at 3:30 p.m. Korean Standard Time. These are the same local date, not simultaneous prices.
Forward P/E = closing price ÷ consensus adjusted EPS for the labeled fiscal year. Every valuation table uses this calculation. These annual estimates are not next-twelve-months estimates, and we do not label them as such.
The tables consistently use that explicit fiscal-year forward calculation. A quote service’s headline forward P/E may use another estimate window and therefore differ. Forecast earnings are consensus estimates, not company guidance, and the adjustments can differ between issuers.
January 2028 fiscal years are used for NVIDIA, Dell and Marvell to approach late-2027 peer periods. Remaining differences are labeled, including Supermicro’s June 2027 and Sandisk’s June–July 2027 years. We do not invent calendarized earnings to remove those differences.
TSMC’s forecast ADS earnings use five ordinary shares per ADS and NT$32.8938 per dollar. Korean ratios use local prices and local earnings. The calculation inputs for every company appear below.
Price and earnings inputs
These are the inputs used to calculate each forward multiple. Korean ordinary-share rows use won; other rows use dollars. TSMC’s displayed EPS is rounded, while its multiple uses the unrounded ADS calculation.
| Company | Closing price | Forecast adjusted EPS | Forecast fiscal year / ending |
|---|---|---|---|
| NVIDIA | $162.85 | $15.83 | FY2028 / Jan 2028 |
| Advanced Micro Devices | $534.06 | $17.69 | FY2027 / Dec 2027 |
| Micron | $778.31 | $176.50 | FY2027 / Sep 2027 |
| SK hynix | ₩1,216,000 | ₩535,730 | FY2027 / Dec 2027 |
| Samsung Electronics | ₩206,000 | ₩77,320 | FY2027 / Dec 2027 |
| Sandisk | $885.01 | $227.48 | FY2027 / Jun–Jul 2027 |
| TSMC | $398.73 | $22.73 | FY2027 / Dec 2027 |
| Intel | $59.66 | $1.79 | FY2027 / Dec 2027 |
| Broadcom | $347.01 | $19.41 | FY2027 / Oct 2027 |
| Marvell Technology | $156.76 | $7.35 | FY2028 / Jan 2028 |
| Arista Networks | $197.29 | $6.04 | FY2027 / Dec 2027 |
| Dell | $344.53 | $30.41 | FY2028 / Jan 2028 |
| HPE | $48.29 | $4.74 | FY2027 / Oct 2027 |
| Supermicro | $26.87 | $6.52 | FY2027 / Jun 2027 |
What the forecast earnings include
The forward P/E tables use consensus adjusted EPS. Adjusted earnings can exclude amortization, stock compensation, acquisition costs and other items. The nature and size of those exclusions can differ between companies.
A lower multiple therefore needs a credible earnings forecast behind it. The separate cash-quality assessment checks whether reported profits translate into cash after investment. That measure evaluates the business; it is not another price-based valuation ratio.
Cash-flow evidence behind the scores
These figures show the cash-conversion inputs behind the ratings. Amounts are in U.S. dollars except the Korean rows, which use won. Each ratio compares the same company’s latest available twelve-month FCF and reported net income.
| Company | Net income | Free cash flow | FCF ÷ net income | Cash score /5 |
|---|---|---|---|---|
| NVIDIA | 192.88 bn | 127.01 bn | 66% | 3 |
| AMD | 7.62 bn | 7.06 bn | 93% | 4 |
| Micron | 84.97 bn | 58.96 bn | 69% | 3 |
| SK hynix | 189.26 tn won | 124.52 tn won | 66% | 3 |
| Samsung Electronics | 212.96 tn won | 203.08 tn won | 95% | 4 |
| Sandisk | 19.42 bn | 17.37 bn | 89% | 4 |
| TSMC | 79.31 bn | 45.02 bn | 57% | 3 |
| Intel | -12.03 bn | 3.77 bn | NM: net loss | 2 |
| Broadcom | 38.27 bn | 39.40 bn | 103% | 5 |
| Marvell | 2.64 bn | 1.73 bn | 66% | 3 |
| Arista Networks | 4.75 bn | 5.02 bn | 106% | 5 |
| Dell | 11.38 bn | 8.56 bn | 75% | 4 |
| HPE | 2.68 bn | 4.16 bn | 155% | 5 |
| Supermicro | 3.29 bn | -6.50 bn | -198% | 1 |
A high conversion ratio is not automatically recurring cash quality. Noncash charges can reduce net income, while working-capital releases can temporarily increase cash flow. Conversely, investment spending can lower conversion at an expanding manufacturer without implying a failing business.
Supermicro’s negative trailing FCF is a stronger caution than one weak quarter alone. Intel has positive FCF but a reported net loss, so its conversion ratio is not meaningful. Neither receives a favorable score simply because its denominator is unusual.
How sensitive are the selections?
We tested four alternatives, moving five percentage points between criteria each time. No company rating changes in this exercise. This tests weight sensitivity; it does not resolve uncertainty in the underlying estimates or editorial ratings.
| Weight scenario | Processors | Memory | Manufacturing | Chips/networking | Servers |
|---|---|---|---|---|---|
| Baseline: 35 / 30 / 15 / 15 / 5 | NVIDIA | Micron | TSMC | Broadcom | Dell |
| More valuation: 40 / 25 / 15 / 15 / 5 | NVIDIA | Micron | TSMC | Broadcom | Dell |
| More business: 30 / 35 / 15 / 15 / 5 | NVIDIA | Micron | TSMC | Broadcom | Dell |
| More cash: 35 / 25 / 20 / 15 / 5 | NVIDIA | Micron | TSMC | Broadcom | Dell |
| More reporting: 30 / 30 / 15 / 15 / 10 | NVIDIA | Micron | TSMC | Broadcom | Dell |
Weights are listed in the order valuation, business, cash, balance and reporting. All five selections survive these particular weight changes. This does not test different valuation-band boundaries, which could materially alter the memory comparison.
A separate one-point reduction in each winner’s business rating reverses the memory choice and brings Dell level with HPE. HPE then wins the cash-score tiebreaker. NVIDIA, TSMC and Broadcom retain their leads in that test.
What happens if forecast earnings fall?
The next table holds prices fixed and reduces the selected annual earnings estimates by 20%. It shows the resulting forward multiple, not a projected share-price return. A memory downturn could exceed this illustrative haircut by a wide margin.
| Selected stock | Baseline forward P/E | P/E with 20% lower forecast EPS | Forecast period |
|---|---|---|---|
| NVIDIA | 10.3× | 12.9× | FY2028 / Jan 2028 |
| Micron | 4.4× | 5.5× | FY2027 / Sep 2027 |
| TSMC | 17.5× | 21.9× | FY2027 / Dec 2027 |
| Broadcom | 17.9× | 22.3× | FY2027 / Oct 2027 |
| Dell | 11.3× | 14.2× | FY2028 / Jan 2028 |
All five companies remain exposed to AI infrastructure spending. Different categories do not eliminate the shared risk of weaker customer budgets. Company-specific political, leadership and customer-concentration risks also remain outside this simple numerical stress test.
Primary Sources
- NVIDIA leadership-transition filing
- SK hynix third-quarter 2026 results
- TSMC third-quarter 2026 financial materials
- Broadcom third-quarter fiscal 2026 results
- Dell second-quarter fiscal 2027 results and guidance
- HPE third-quarter fiscal 2026 results and outlook
FAQ
What weights determine the five stock selections?
Valuation receives 35%, AI business and competitive position 30%, cash conversion 15%, balance-sheet strength 15%, and ownership/reporting 5%. The same weights apply to every candidate. Business and reporting ratings remain disclosed editorial judgments.
Why does Dell rank above HPE?
Dell’s business and balance-sheet scores outweigh HPE’s stronger cash conversion. HPE has the lower forward P/E, but both receive the same valuation-band score. A one-point lower business rating for Dell creates a tie, which HPE wins on cash conversion.
Why does Micron rank above the cheaper SK hynix?
Both fall in the highest forward-P/E scoring band and receive equal business, cash and balance scores. Micron’s reporting score provides a one-point lead. SK hynix is still cheaper; a method that rewards the full size of that discount could change the result.
How is manufacturing valued?
TSMC, Samsung and Intel all have forward P/E figures calculated from their stated fiscal-year estimates. Samsung and Intel ratios describe their entire groups, not separately listed foundry businesses. The research notes include the price and EPS inputs for each calculation.
Why does the article’s forward P/E differ from a quote page?
The article divides the dated closing price by consensus adjusted EPS for an explicitly labeled fiscal year. A quote page may use a different forward window or estimate set. Those figures should not be substituted without checking their definitions.
Does the score prove a stock trades below fair value?
No. This is a transparent screening model, not a discounted-cash-flow valuation or a prediction of returns. Memory cycles, non-recurring gains and changes in customer spending can alter the conclusions substantially.
