AI stocks, the 2026 guide to investing in AI, a retro screen-print of a human head whose brain is a circuit board beside a rising stock chart

AI Stocks in 2026: How to Invest in Artificial Intelligence (Updated Weekly)

Published January 18, 2025 · Updated September 10, 2026 · 21 min read · Prices refresh weekly
// What this covers
  • The five layers of the AI stack, and why the layer you buy matters more than the ticker
  • Live prices on fifteen names, refreshed weekly
  • The capex number the entire trade rests on, given as a range with the basis named, because every published tally counts it differently
  • The circular financing question: NVIDIA's disclosed $279 billion of supply commitments, its $42.8 billion of equity stakes in customers, and Broadcom arranging $35 billion of outside money so a customer could buy its chips
  • How to reach OpenAI, Anthropic and xAI from a brokerage account, and the one that has reportedly filed to go public

"AI stocks" is the most searched investment phrase in the market right now, and the lists that answer it are close to useless. They rank by market cap, name the same eight companies, and never tell you the thing that actually determines your outcome: where in the chain you are standing, and who is paying the person standing next to you.

Because the striking feature of this cycle is not the demand. It is the financing. NVIDIA discloses $279 billion of supply and capacity commitments, up from $119 billion a quarter earlier, while holding $42.8 billion of marketable equity securities that include stakes in its own customers. Broadcom arranged a $35 billion financing vehicle with Apollo and Blackstone so that a customer could take delivery of its chips. Amazon booked $53.4 billion of pre-tax other income in a single quarter from revaluing its Anthropic stake, which is a mark, not a sale.

None of that means the demand is fake. It means the money is moving in a circle, and where you sit in that circle is your actual risk.

// The map

The Five Layers of the AI Stack

LAYER 01

Silicon

Designs and manufactures the chips. The purest exposure and the most cyclical. Revenue is enormous and already growing triple digits.

NVDA · AVGO · AMD · MU · TSM

LAYER 02

Infrastructure

Racks, networking, cooling and rented compute. Arms dealers to the buildout. One of them is a levered landlord.

VRT · ANET · CRWV

LAYER 03

Power

Electricity is the binding constraint. These are not AI companies, they are utilities and fuel cell makers whose customer list changed.

CEG · VST · BE

LAYER 04

Platforms

The hyperscalers doing the spending. AI is one segment inside a much larger business, and they are the demand the other layers depend on.

MSFT · GOOGL · AMZN · SPCX

LAYER 05

Applications

Sells AI as the product rather than the input, whether that is software or a car that drives itself. The richest layer by multiple.

PLTR · TSLA

Read the layer before the ticker. Layer 1 and 2 sell into Layer 4's capex budget, which means their revenue is Layer 4's expense. Layer 3 sells into all of it and is the only layer with a physical constraint that money cannot compress. Layer 5 is the only one whose revenue comes from customers outside this diagram. Those are four different businesses, and a single AI headline moves them in different directions.
// The board

AI Stocks: The Seventeen Tickers That Carry the Category

Prices refresh automatically every week. Every other figure comes from company filings and is dated.

  • NVDA NVIDIA logo
    NVDA
    NVIDIA
    $218.36
    ▼ -2.37% today
    MKT CAP $5.27TP/E 27.652W $164-237
    ◎ BEST PLAY: Defined-risk debit spread · 30-45 DTE

    The centre of gravity. Q2 FY2027 revenue $96.2B (+106%) with a $108B Q3 guide. Total supply and capacity commitments of $279B, up from $119B a quarter earlier, plus $42.8B of equity securities including stakes in customers. On Sept 2 2026 it signed a definitive agreement to buy Hugging Face for about $11.9B plus up to $1.0B retention, expected to close in H1 2027 subject to regulatory approval.

  • AVGO Broadcom logo
    AVGO
    Broadcom
    $360.83
    ▼ -0.97% today
    MKT CAP $1.72TP/E 46.152W $290-495
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    The only merchant supplier of custom AI accelerators at scale. Q3 FY2026 revenue $29.591B (+86%), of which AI semiconductors were $16.7B, up 221% YoY and 56% of the company. Q4 guided to ~$34.8B. Management targets roughly $58B of AI revenue in FY2026 rising to about $115B in FY2027.

  • MU Micron logo
    MU
    Micron
    $977.41
    ▼ -4.90% today
    MKT CAP $1.10TP/E 22.152W $148-1255
    ◎ BEST PLAY: Defined-risk debit spread · 30-45 DTE

    The most extraordinary quarter in the group: fiscal Q3 2026 revenue of $41.46B against $9.30B a year earlier, up about 346%, at roughly 86% gross margin, with Q4 guided to $50B. High bandwidth memory became the scarcest input in the buildout. Treat that margin as a shortage artifact, not a structural feature.

  • AMD Advanced Micro Devices logo
    AMD
    Advanced Micro Devices
    $503.60
    ▼ -3.36% today
    MKT CAP $822.11BP/E 128.852W $150-585
    ◎ BEST PLAY: Defined-risk debit spread · 30-45 DTE

    Q2 2026 revenue $11.536B (+50%) with Data Center at $6.718B, up 107% and now 58% of the company. Q3 guided to about $13B at ~56% non-GAAP gross margin. Carries $30.276B of future purchase commitments. Priced near 133x trailing earnings, so the execution bar is set high.

  • TSM Taiwan Semiconductor logo
    TSM
    Taiwan Semiconductor
    $428.03
    ▼ -1.68% today
    MKT CAP $2.22TP/E 31.652W $258-479
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    Manufactures essentially every leading-edge AI accelerator. Q2 2026 revenue US$40.20B (+33.7%) at 67.7% gross margin and 60.3% operating margin. The tell is capex: NT$496B in a single quarter against NT$287B of free cash flow. The foundry is now spending faster than it earns.

  • VRT Vertiv logo
    VRT
    Vertiv
    $248.13
    ▼ -5.61% today
    MKT CAP $95.53BP/E 56.152W $133-380
    ◎ BEST PLAY: Debit spread · 30-45 DTE

    Sells the power and cooling a rack cannot run without. Q2 2026 net sales $3.274B (+24.1%), full-year guidance raised to $13.8B-$14.2B, about 31% organic growth at the midpoint, with adjusted free cash flow guided to $2.4B-$2.6B. Trades near 8.8x trailing sales.

  • ANET Arista Networks logo
    ANET
    Arista Networks
    $188.99
    ▼ -2.04% today
    MKT CAP $238.36BP/E 59.852W $115-215
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    The Ethernet fabric of AI datacenters. Q2 2026 revenue $3.036B (+37.7%), its first quarter above $3B, with roughly $8.4B of remaining performance obligations and a debt-free balance sheet holding $13.3B of cash. Its disclosed 2026 AI fabric target is at least $3.5B.

  • CRWV CoreWeave logo
    CRWV
    CoreWeave
    $89.12
    ▼ -6.13% today
    MKT CAP $49.15BP/E N/A52W $61-153
    ◎ BEST PLAY: Defined risk, size for volatility · 30-45 DTE

    A levered landlord for someone else’s compute. Q2 2026 revenue $2.575B (+112%) and remaining performance obligations of $103.7B, plus $25B signed after quarter end. Against that: $31.4B recourse debt and $3.7B non-recourse against $5.5B of cash. The backlog is the bull case and the counterparty risk in one number.

  • CEG Constellation Energy logo
    CEG
    Constellation Energy
    $285.97
    ▼ -2.70% today
    MKT CAP $101.32BP/E 28.052W $229-413
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    Not an AI company, an independent power producer whose customer list changed. Sells firm around-the-clock carbon-free power from a ~55 GW fleet and signed another 920 MW of long-term PPAs in Q2 2026. FY2026 adjusted operating EPS guidance raised to $11.50-$12.50.

  • VST Vistra logo
    VST
    Vistra
    $147.05
    ▼ -2.68% today
    MKT CAP $49.36BP/E 24.852W $133-220
    ◎ BEST PLAY: Debit spread / covered call · 30-45 DTE

    Also not an AI company. Signed 20-year PPAs with Meta in January 2026 covering 2,609 MW from its PJM nuclear plants. Reaffirmed 2026 adjusted EBITDA guidance of $6.8B-$7.6B. Share count is shrinking with no equity raise in the last 12 months, against about $19.6B of debt.

  • BE Bloom Energy logo
    BE
    Bloom Energy
    $258.49
    ▼ -4.01% today
    MKT CAP $76.13BP/E 331.452W $61-351
    ◎ BEST PLAY: Defined risk, size for volatility · 21-30 DTE

    The impatient power option: solid oxide fuel cells installed onsite that deliver grid-independent power in months instead of an interconnection queue. Q2 2026 revenue $1.065B (+165.5%), FY guidance raised to $3.9B-$4.2B. The risk is concentration: one customer was about 73% of first-half revenue against $442.4M of GAAP backlog.

  • SPCX SpaceX logo
    SPCX
    SpaceX
    $148.18
    ▲ +0.43% today
    MKT CAP $1.95TP/E N/A52W $105-226
    ◎ BEST PLAY: Defined risk, size for volatility · 30-45 DTE

    The AI name most lists miss. SpaceX acquired X.AI Holdings on February 2 2026, so xAI (which already owned X Corp) is a wholly owned subsidiary, and SPCX reports AI as one of three segments alongside Space and Connectivity. Q2 2026 AI segment revenue was $2,561M against a $1,257M operating loss on $15,828M of segment capex. Musk holds about 82.4% of the voting power, so it is a controlled company.

  • MSFT Microsoft logo
    MSFT
    Microsoft
    $492.44
    ▲ +0.16% today
    MKT CAP $3.66TP/E 27.452W $349-554
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    A platform writing the cheques the other layers cash. FY26 Q4 revenue $90.0B (+18%) with commercial remaining performance obligations of $678B, up 84%, and Azure past $100B in annual revenue. Also holds the largest corporate stake in OpenAI, reported around 27%.

  • GOOGL Alphabet logo
    GOOGL
    Alphabet
    $332.60
    ▲ +0.59% today
    MKT CAP $4.07TP/E 16.752W $236-409
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    Q2 2026 revenue $119.8B (+24%) with Cloud up 82% and a cited $514B backlog. Holds a reported ~14% of Anthropic, hard capped at 15%, while also selling it compute. Share count is rising rather than shrinking, with buybacks at zero in the first half of 2026.

  • AMZN Amazon logo
    AMZN
    Amazon
    $251.89
    ▼ -0.20% today
    MKT CAP $2.72TP/E 20.352W $196-287
    ◎ BEST PLAY: Covered call / debit spread · 30-45 DTE

    Q2 2026 net sales $200.6B (+20%), AWS $42.2B (+37%) at a 39.4% segment margin, performance obligations near $496B with a 6.4-year weighted average life. Read the earnings line carefully: $53.4B of Q2 pre-tax other income came from revaluing its Anthropic stake, which is a mark rather than cash from operations.

  • PLTR Palantir logo
    PLTR
    Palantir
    $165.86
    ▼ -2.16% today
    MKT CAP $398.57BP/E 141.852W $106-208
    ◎ BEST PLAY: Defined-risk debit spread · 30-45 DTE

    The application layer, and the only name here whose customers sit outside the AI capex loop. Q2 2026 revenue $1.935B (+93%) with US commercial up 149% and $4.9B of noncancelable contracted revenue. It is also the most expensive: roughly 49x guided 2026 revenue and a trailing P/E near 149.

  • TSLA Tesla logo
    TSLA
    Tesla
    $363.56
    ▼ -1.16% today
    MKT CAP $1.44TP/E 339.852W $297-499
    ◎ BEST PLAY: Defined-risk debit spread · 30-45 DTE

    Designs its own inference silicon (AI4 shipping, AI5 taped out at TSMC and Samsung) and is building a fab in Austin, but sells none of it to anyone, which is why it sits in applications not silicon. Reports exactly two segments, automotive and energy: no AI segment and no disclosed AI revenue. Discloses 1.48M active FSD subscriptions (+56%). Q2 2026 operating margin 1.4%, free cash flow negative $1.092B.

// Live quotes as of Sep 10, 2026 · delayed snapshot, not real-time
// Free list

See What We Are Watching

Get on our free list and the watchlist and alerts we send land in your inbox. AI and semiconductor names turn up on them regularly.

Free. No spam. Education only, nothing here is financial advice. Unsubscribe anytime.

// Layer 01

Silicon: Where the Revenue Actually Landed

NVIDIA (NVDA) is the center of gravity. Fiscal Q2 2027 revenue was $96.2 billion, up 106% year over year, with a Q3 guide of $108 billion and gross margin around 75%. The number that matters more than any of those is the balance sheet disclosure: total supply and capacity commitments of $279 billion, up from $119 billion the prior quarter. The filing itself notes certain vendor agreements are cancellable or reschedulable, so not all of it is irrevocable, but it is the largest forward datapoint in the cycle.

NVIDIA is also buying its way up the stack. On September 2, 2026 it signed a definitive agreement to acquire Hugging Face, the repository where open-weight models are actually distributed, for approximately $11.9 billion to stockholders plus up to $1.0 billion in equity retention. Hugging Face hosts more than 3 million models and 500,000 datasets for over 18 million developers. If it closes, NVIDIA would own the discovery and distribution layer for open models on top of the silicon, systems, networking and CUDA software it already owns.

Signed is not closed. NVIDIA's own 8-K says the Hugging Face deal is expected to close in the first half of 2027, subject to customary conditions including required regulatory approvals. A vertically integrated chip vendor buying the main neutral distribution point for open models is exactly the shape antitrust reviewers look at, so treat it as pending rather than done.

Broadcom (AVGO) is the only merchant supplier of custom AI accelerators at scale. Q3 fiscal 2026 revenue was $29.591 billion, up 86%, of which AI semiconductors were $16.7 billion, up 221% year over year and 56% of the company. Q4 is guided to about $34.8 billion. Management has put dated targets on the AI line: roughly $58 billion in fiscal 2026, about $115 billion in fiscal 2027 and about $230 billion in fiscal 2028.

Micron (MU) had the most extraordinary quarter of any of them. Fiscal Q3 2026 revenue was $41.46 billion against $9.30 billion a year earlier, up about 346%, at roughly 86% gross margin, with Q4 guided to $50 billion. High bandwidth memory turned into the scarcest input in the buildout. Treat that margin as a pricing artifact of a shortage rather than a structural feature, because memory has always been a commodity cycle and this is what a cycle peak looks like.

AMD reported Q2 2026 revenue of $11.536 billion, up 50%, with Data Center at $6.718 billion, up 107% and now 58% of the company. Q3 is guided to about $13 billion.

TSMC (TSM) manufactures essentially all of it. Q2 2026 revenue was US$40.20 billion, up 33.7%, at 67.7% gross margin. The tell is capex: NT$496 billion in a single quarter against NT$287 billion of free cash flow. The foundry is now spending faster than it earns to keep up.

// Layer 02

Infrastructure: Two Arms Dealers and One Landlord

Vertiv (VRT) sells the power and cooling that a rack cannot run without. Q2 2026 net sales were $3.274 billion, up 24.1%, with full-year guidance raised to $13.8 to $14.2 billion, about 31% organic growth at the midpoint. Arista (ANET) sells the Ethernet fabric: Q2 2026 revenue of $3.036 billion, up 37.7%, its first quarter above $3 billion, with roughly $8.4 billion of remaining performance obligations and a debt-free balance sheet holding $13.3 billion of cash.

CoreWeave (CRWV) is a different animal and deserves reading carefully. Q2 2026 revenue was $2.575 billion, up 112%, and remaining performance obligations were $103.7 billion, with more than $25 billion of further commitments signed after quarter end. Against that sits $31.4 billion of recourse debt plus $3.7 billion non-recourse, against $5.5 billion of cash.

What CoreWeave actually is. It borrows to buy GPUs, then rents them under long take-or-pay contracts. The $103.7 billion backlog is the bull case and the counterparty risk in the same number: it converts to revenue only if the customers pay and CoreWeave can deliver the capacity on schedule. That is a levered, single-purpose landlord for someone else's compute, which is a legitimate business and an entirely different risk from owning the chipmaker.
// Layer 03

Power: The Constraint Money Cannot Compress

One more source of capacity is worth knowing about, because it did not exist as an AI story a year ago. The large bitcoin miners spent 2026 signing multi year leases with AI companies for the power and buildings they already owned, and several now carry billions of debt against a handful of millions in quarterly revenue while they build. We cover that side of it, and what it did to those balance sheets, in the crypto and bitcoin stocks guide.

You can order more chips. You cannot order more grid. That is why utilities ended up on AI stock lists, and they belong there for a reason that has nothing to do with technology.

Constellation (CEG) sells firm, around the clock, carbon free generation from a roughly 55 gigawatt fleet, and signed an additional 920 megawatts of long term power purchase agreements in Q2 2026. Full-year adjusted operating earnings guidance was raised to $11.50 to $12.50 per share. Vistra (VST) signed 20 year agreements with Meta in January 2026 covering 2,609 megawatts from its PJM nuclear plants, and reaffirmed 2026 adjusted EBITDA guidance of $6.8 to $7.6 billion.

Bloom Energy (BE) is the impatient option: solid oxide fuel cells installed onsite that deliver grid independent power in months rather than the years an interconnection queue takes. Q2 2026 revenue was $1.065 billion, up 165.5%, with full-year guidance raised to $3.9 to $4.2 billion.

Two cautions on this layer. Neither CEG nor VST is an AI company; they are independent power producers whose customer mix changed, and they will trade on power prices, capacity auctions and rate cases as much as on AI headlines. And on Bloom, one customer was about 73% of first-half 2026 revenue, against GAAP backlog of only $442.4 million. Concentration that heavy is the risk, not the growth rate.
// Layer 04

Platforms: The Ones Writing the Cheques

The hyperscalers are not an AI trade in the way the other layers are. AI is a segment inside a much larger business, and their capex is the revenue line for everyone in Layers 1, 2 and 3.

Amazon (AMZN) reported Q2 2026 net sales of $200.6 billion, up 20%, with AWS at $42.2 billion, up 37%, and performance obligations of roughly $496 billion with a 6.4 year weighted average life. AWS said its AI business passed a $25 billion annual run rate. One line deserves flagging: $53.4 billion of Q2 pre-tax other income came from revaluing its Anthropic stake. That is an accounting mark on a private holding, not cash from operations, and it makes the headline earnings number a poor guide to the business.

SpaceX (SPCX) belongs here for a reason most AI lists have not caught up with. SpaceX completed its acquisition of X.AI Holdings on February 2, 2026, so xAI, which already owned X Corp, is now a wholly owned subsidiary of the entity that trades as SPCX. It is not an equity stake or a shared shareholder: SPCX reports AI as one of its three segments, alongside Space and Connectivity. In Q2 2026 that AI segment produced $2,561 million of revenue and a $1,257 million loss from operations, on $15,828 million of segment capital expenditure.

Microsoft (MSFT) reported fiscal Q4 revenue of $90.0 billion, up 18%, with commercial remaining performance obligations of $678 billion, up 84%. Alphabet (GOOGL) reported Q2 2026 revenue of $119.8 billion, up 24%, with Cloud up 82% and a cited $514 billion backlog.

// Layer 05

Applications: Selling AI as the Product

Palantir (PLTR) is the cleanest example on the board, and the only name here whose customers sit outside the AI capex loop entirely. Q2 2026 revenue was $1.935 billion, up 93%, with US commercial up 149% and $4.9 billion of noncancelable contracted revenue. It is also the most expensive thing here, at roughly 49 times guided 2026 revenue.

Tesla (TSLA) earns its place, with a caveat worth stating plainly. Tesla designs its own inference silicon: AI4 ships in cars today, AI5 taped out at TSMC in April 2026 and at Samsung on a 2nm class node reported in July, and it is building its own semiconductor fab in Austin. Dojo was shut down in August 2025 and restarted in January 2026 once the AI5 design settled. In Q2 2026 it also acquired an AI hardware company for $1.95 billion in stock.

What it does not do is sell any of that to anyone else, which is why it sits in applications rather than silicon. Tesla reports exactly two segments, automotive and energy generation and storage. There is no AI segment and no disclosed AI or robotics revenue. The one AI demand metric it does publish is FSD uptake: 1.48 million active subscriptions in Q2 2026, up 56%, with more than 55% of new North American deliveries attaching FSD. The revenue from that is bundled, not broken out. The spending is visible on the balance sheet instead, where an AI infrastructure line in property and equipment reached $10,823 million at June 30, 2026, up from $6,816 million.

Buy Tesla for AI and you buy a car company. Q2 2026 revenue was $28.236 billion, up 26%, but operating margin fell to 1.4% and free cash flow was negative $1.092 billion. Every AI dollar is booked inside vehicle sales and services. There is no way to own the AI program separately, because it is not a segment, a subsidiary or a security.
// The number

The Capex Figure the Whole Trade Rests On

Every thesis on this page reduces to one question: how much will the hyperscalers spend, and for how long? Adding up the four companies' own stated 2026 figures gives roughly $735 to $770 billion, against something near $410 billion the prior year.

Do not treat that as a fact, treat it as a range. Every published tally counts it differently. Meta's own guidance of $130 to $145 billion includes principal payments on finance leases; Amazon guides "capital investments" on a cash flow statement basis, which is not the same thing; and Microsoft reports on a June fiscal year, so any calendar 2026 Microsoft figure is a construction rather than a company statement. Mixing those bases is the most common error in AI capex headlines. Projections above a trillion dollars for 2027 are forecasts, not guidance.

On the demand side, the International Energy Agency's Electricity 2026 report projects global data center electricity demand rising toward roughly 945 terawatt hours by 2030 from 415 terawatt hours in 2024, and attributes about half of US electricity demand growth through 2030 to data centers. Power, not silicon, is where this eventually binds.

// The bear case

The Circular Financing Question, Stated Fairly

The strongest bear case on record is not that AI demand is fake. It is that the demand is being partly financed by the sellers, which makes it look larger and more independent than it is.

The specific mechanisms are disclosed, not alleged. NVIDIA holds $42.8 billion of marketable equity securities including stakes in customers. Broadcom closed a first $35 billion tranche on a financing platform with Apollo and Blackstone tied to Anthropic capacity, so a customer could fund the purchase of Broadcom silicon. Amazon and Alphabet both hold large stakes in Anthropic while also selling it compute. CoreWeave borrows to buy the GPUs it rents back to the same ecosystem.

Michael Burry made the comparison explicit in an August 21, 2026 post, writing: "The shenanigans are apparent today for those that care to look." His argument is the Lucent and Nortel dynamic of 1999 to 2001, where vendor equity stakes and vendor financing inflated apparent end demand.

A second leg of the bear case concerns depreciation: the argument that AI hardware has a two to three year economic life while it is depreciated over five to six, flattering reported earnings across the group. Estimates of the gap circulating publicly run to roughly $176 billion across 2026 to 2028. Those are third party estimates rather than disclosed figures, but the input is checkable: useful life assumptions are stated in every 10-K, and you can read them yourself.

The honest framing. Nobody credible on record is calling 2026 demand fake. The bear case is about timing and financing structure, and most of its proponents describe it as a 2028 question rather than a 2026 one. Treat it as a reason to know which layer you own and how it is funded, not as a reason to avoid the sector.
// The privates

OpenAI, Anthropic and xAI: What You Can Actually Buy

Most people asking about AI exposure mean one of these labs. Two are private. One is not, and that surprises people.

CompanyReported valuationHow to get exposure
OpenAI$852B post-moneyMicrosoft holds the largest corporate stake, reported around 27% of OpenAI Group, plus IP and Azure rights
Anthropic$965B post-moneyAlphabet reported near 14%, hard capped at 15%. Amazon has invested $13B plus more. Reportedly filed confidentially for a US IPO on June 1, 2026
xAINo longer standalonePublic via SPCX. SpaceX acquired xAI on February 2, 2026 and reports it as its AI segment. Buying SPCX is pro rata ownership of 100% of xAI, X and Cursor
Databricks$190B post-moneyThin. NVIDIA is a disclosed strategic investor; Alphabet's CapitalG and Microsoft's M12 hold immaterial positions
xAI is the exception, and it is already public. SpaceX closed its acquisition of X.AI Holdings on February 2, 2026, and xAI had itself acquired X Corp in March 2025. SPCX listed on Nasdaq on June 12, 2026 at $135 per share and now reports AI as a full segment. There is no minority stub and no tracking stock: an SPCX holder owns xAI pro rata. That segment kept buying: SpaceX closed an all stock acquisition of Anysphere, the maker of the AI coding tool Cursor, on August 14, 2026, issuing roughly 389.3 million Class A shares at an implied Cursor equity value of $60.0 billion, and Cursor now sits inside the same AI segment as a wholly owned subsidiary. Two things to weigh before treating that as clean exposure. The AI segment lost $1,257 million in Q2 2026 on $2,561 million of revenue, and Elon Musk holds roughly 82.4% of the voting power, so SPCX is a controlled company under Nasdaq rules.
The one to watch. Anthropic is reported to have filed confidentially with the SEC on June 1, 2026. Confidential filings are not public documents, so the reporting is second hand until an S-1 is made public, but if it lists it would be the first frontier AI lab available directly rather than through a parent's balance sheet. Valuations above are reported figures from funding rounds, not audited marks.
// The funds

If You Would Rather Own the Basket

ETFNet assetsWhat you are actually buyingYTD
AIQ$10.15BPassive, Indxx AI & Big Data index. Top holding Palantir at 3.76%+26.05%
CHAT$1.87BActively managed. Top ten is about 38% of the fund, led by NVIDIA, Alphabet, Broadcom+53.82%
BOTZ$3.38BRobotics tilted and heavily Japanese: Keyence 10.4%, NVIDIA 9.9%, ABB 9.2%, Fanuc 7.8%-2.85%
QQQ$485.91BNasdaq-100, not an AI fund, but the top ten is roughly 46% and led by NVIDIA at 8.7%+16.88%

The spread between CHAT at plus 53.82% and BOTZ at minus 2.85% year to date is the point. Both are sold as AI exposure. They own very different companies, and one of them is mostly Japanese industrial automation.

// The options layer

What the Options Market Will Let You Do

TickerChain depthWhat that permits
NVDA, MSFT, GOOGL, AMZNAmong the deepest listed anywhereAny structure, any expiry, spreads a few cents wide
AVGO, AMD, MU, TSMVery deep, weeklies plus LEAPSAny structure. Pronounced earnings volatility bumps
PLTR, ANET, VRTDeep, heavy retail and event interestDefined-risk spreads around earnings
CRWV, BEListed and active, wider marketsDefined risk, size for the volatility
CEG, VSTSolid, utility-style flowCovered calls and spreads at monthlies
The structural feature of this sector's options. Upside call skew stays persistently bid across the AI complex, which means long calls are expensive in a specific way: you are paying for the tail everyone else also wants. Defined-risk spreads that sell some of that premium back are usually the better expression, and on the richest names they are the only sane one.
// Practical

Three Questions Before Any AI Position

1. Which layer am I buying, and who pays it? If you own Layer 1 or 2, your revenue is Layer 4's capex budget, which is a decision made by four companies. If you own Layer 3, you own a physical constraint. If you own Layer 5, you own the only layer whose customers sit outside this diagram.

2. Is this company's demand independently financed? Vendor equity stakes, take or pay contracts with cash consuming counterparties, and debt funded hardware purchases all make demand look sturdier than it is. That is not a reason to avoid a name. It is a reason to know the answer before you size the position.

3. What is the customer concentration? Bloom at roughly 73% from one customer, CoreWeave's contracted backlog, Broadcom's handful of XPU counterparties, Arista's hyperscaler mix. In this sector concentration is the norm, and the filings tell you the number.

Two related reads on this site: AI memory stocks and how Micron and SanDisk outran the chip giants for a closer look at Layer 1's scarcest input, and Is AI a bubble, what 400 years of manias reveal for the historical version of the argument above.

// Free

Get On Our Free List

Our watchlist and our alerts go out to the free list, and AI and semiconductor names turn up on them regularly. It costs nothing, and it is the simplest way to see which names we are studying and what we are watching on them.

Free. No spam. Education only, nothing here is financial advice. Unsubscribe anytime.

// FAQ

Frequently Asked Questions

Is Tesla an AI stock?

Partly, and it belongs in the applications layer rather than silicon. Tesla designs its own inference chips, with AI4 shipping in cars today and AI5 taped out at TSMC in April 2026 and at Samsung in July, and it is building a semiconductor fab in Austin. But it sells none of that to anyone else. Tesla reports exactly two segments, automotive and energy generation and storage, with no AI segment and no disclosed AI or robotics revenue. The one AI demand metric it publishes is 1.48 million active FSD subscriptions in Q2 2026, up 56 percent, and even that revenue is bundled rather than broken out. Q2 2026 operating margin was 1.4 percent and free cash flow was negative 1.092 billion dollars.

How can I invest in xAI?

Buy SPCX. SpaceX completed its acquisition of X.AI Holdings on February 2 2026, and xAI had already acquired X Corp in March 2025, so both are wholly owned subsidiaries of the entity listed on Nasdaq since June 12 2026. SPCX reports AI as one of three segments: in Q2 2026 that segment produced 2,561 million dollars of revenue against a 1,257 million dollar operating loss, on 15,828 million dollars of segment capital expenditure. There is no minority stub and no tracking stock, so an SPCX holder owns xAI pro rata. Note that Elon Musk holds roughly 82.4 percent of the voting power, making SPCX a controlled company under Nasdaq rules.

What are the best AI stocks to buy right now?

We do not publish buy calls. The more useful question is which layer of the AI stack you want to own, because they behave differently. Silicon is NVIDIA, Broadcom, Micron, AMD and TSMC. Infrastructure is Vertiv, Arista and CoreWeave. Power is Constellation, Vistra and Bloom Energy. Platforms are Microsoft, Alphabet and Amazon. Applications, where the customers sit outside the AI capex loop, is Palantir.

How much are the hyperscalers actually spending on AI in 2026?

Adding up the four companies' own stated 2026 figures gives roughly 735 to 770 billion dollars, against something near 410 billion the prior year. Treat that as a range rather than a fact, because every tally counts it differently: Meta's guidance of 130 to 145 billion includes principal payments on finance leases, Amazon guides capital investments on a cash flow statement basis, and Microsoft reports on a June fiscal year, so any calendar 2026 Microsoft figure is a construction rather than a company statement.

Is AI a bubble?

The strongest bear case on record is not that demand is fake, it is that some of the demand is financed by the sellers. NVIDIA holds 42.8 billion dollars of equity securities including stakes in customers, Broadcom arranged a 35 billion dollar financing vehicle with Apollo and Blackstone so a customer could buy its chips, and Amazon and Alphabet both hold large Anthropic stakes while selling it compute. Michael Burry compared it to the Lucent and Nortel vendor financing of 1999 to 2001 in an August 21 2026 post. Most proponents of that view frame it as a 2028 question rather than a 2026 one.

Can I invest in OpenAI or Anthropic?

Two of the three are private; xAI is not. Both OpenAI and Anthropic remain private. Microsoft holds the largest corporate stake in OpenAI, reported around 27 percent of OpenAI Group, at a reported 852 billion dollar post-money valuation. Anthropic, at a reported 965 billion dollars, is roughly 14 percent owned by Alphabet with a hard cap at 15 percent, and Amazon has invested 13 billion dollars plus more. Anthropic is reported to have filed confidentially for a US IPO with the SEC on June 1 2026. xAI is the exception and is already public: SpaceX completed its acquisition of X.AI Holdings on February 2 2026, so buying SPCX on Nasdaq is pro rata ownership of xAI and X, which SPCX reports as its AI segment.

What are the best AI energy or AI power stocks?

Electricity is the constraint that money cannot compress, which is why utilities appear on AI lists. Constellation sells firm carbon-free power from a roughly 55 gigawatt fleet and signed another 920 megawatts of long-term agreements in Q2 2026. Vistra signed 20-year agreements with Meta covering 2,609 megawatts from its PJM nuclear plants. Bloom Energy sells onsite fuel cells that deliver power in months rather than years, though one customer was about 73 percent of its first-half 2026 revenue. Note that none of these is an AI company; they are power businesses whose customer mix changed.

Which AI stock has the most revenue growth?

Micron, by a wide margin, and it is a memory cycle rather than a software story. Fiscal Q3 2026 revenue was 41.46 billion dollars against 9.30 billion a year earlier, roughly 346 percent growth, at about 86 percent gross margin, with the next quarter guided to 50 billion. Broadcom's AI semiconductor line grew 221 percent year over year to 16.7 billion, and Palantir grew 93 percent to 1.935 billion.

What is CoreWeave and is it risky?

CoreWeave borrows money to buy GPUs and rents them back to the AI ecosystem under long take-or-pay contracts. Q2 2026 revenue was 2.575 billion dollars, up 112 percent, with 103.7 billion dollars of remaining performance obligations. Against that it carries 31.4 billion dollars of recourse debt plus 3.7 billion non-recourse, against 5.5 billion of cash. The backlog is both the bull case and the counterparty risk, because it becomes revenue only if customers pay and capacity is delivered on schedule.

Which AI ETF should I look at?

They are far less alike than the labels suggest. Year to date, CHAT is up 53.82 percent and BOTZ is down 2.85 percent, and both are sold as AI exposure. AIQ holds 10.15 billion dollars and tracks a passive AI and big data index led by Palantir. CHAT is actively managed with a top ten of about 38 percent. BOTZ is robotics tilted and heavily Japanese, with Keyence at 10.4 percent and Fanuc at 7.8 percent. QQQ is not an AI fund at all, but its top ten is roughly 46 percent and led by NVIDIA.

What Members Get When an AI Name Sets Up

Everything above is homework you can do yourself. Members get the other half: three to five options alerts a week, each with the full contract and a written plan, sent by text, email and Discord. Every alert we send goes into the public tracker, the ones that worked and the ones that did not.

See What Members Get
PPP Team
PPP Team
Premium Options Trading Education

The PPP Team brings decades of combined experience from some of the most well-known companies in the trading industry. Founded in 2020, Pure Power Picks delivers options trading education, platform reviews, and trade alerts to help everyday traders develop real skills. Our content is strictly educational.

// DISCLAIMER: Pure Power Picks publishes educational content only. We are not financial advisors and nothing on this page is financial, investment, tax or legal advice. Options carry substantial risk and are not suitable for every investor. Figures are drawn from company filings on the dates shown and change without notice. Do your own research.

Leave a Reply

Your email address will not be published. Required fields are marked *