If You Missed NVIDIA, Which Shares Should You Consider? An Honest Take
The premise of the question is that NVIDIA is too late, too expensive, or too crowded, and the smart move is to find the next NVIDIA somewhere else. So I will play along with the premise, lay out the names that come up most often when investors ask this question, and then admit something at the end that the title sets up as a joke: the most honest answer to "which shares should I consider if I missed NVIDIA" usually starts with NVIDIA itself.
What follows is opinion. It reflects how I think about the AI infrastructure trade as someone who advises companies on capital structure rather than someone who manages a fund. I do not hold positions in any of these names as part of my advisory work, and I am not paid by any of the companies mentioned.
1. NVIDIA, Because The Premise Of The Question Is Often Wrong
Yes, I know. The whole point of the article is supposed to be the alternatives. But every time someone asks "which AI stock do I buy if I missed NVIDIA," the honest first answer is to look at NVIDIA again before looking past it.
Here is the picture as of early May 2026. NVIDIA crossed a USD 5 trillion market cap earlier this year, with data center revenue reaching USD 193.7 billion for fiscal year 2026. Forward earnings multiples are around 32 times, which sounds rich until you put it next to the growth rate. Fiscal 2026 revenue grew 65% year over year, and the company has projected USD 1 trillion in confirmed AI chip demand through 2027. That demand figure represents purchase orders from named hyperscalers, not consensus guesswork.
The bear case is real. Customer concentration is real, with two clients representing 36% of fiscal 2026 revenue. Custom silicon from Google, Amazon, and Meta is real. Recent share-price flatness while AMD and Micron rallied is real. But valuation arguments against NVIDIA have been wrong for three straight years, and the company keeps growing into the multiple. Goldman Sachs has a buy rating with a 12-month price target of USD 250 based on a 30 times P/E multiple, which is not a stretched assumption given the earnings trajectory.
My personal view: if the worry is "NVIDIA went up 1,000%, surely it cannot go up another 50%," that is a sentiment objection, not a fundamental one. If the worry is "I want diversification away from a single name in a single segment," that is a portfolio construction question, and there are good answers. The names below are those answers. Just do not pretend NVIDIA is uninvestable when the cash flows say otherwise.
2. Taiwan Semiconductor Manufacturing (TSM): The Multiplier
This is the name I find easiest to defend on first principles. TSMC manufactures the chips for NVIDIA, AMD, Apple, Qualcomm, and Broadcom's custom AI accelerators. When any of those companies wins, TSMC wins. When Google, Amazon, and Meta build in-house chips, those still get fabricated at TSMC. TSMC has roughly 72% share in the foundry market, and that share is highest at the leading-edge nodes where AI chips actually live.
The valuation is the punchline. TSMC trades at about 24 times forward earnings versus roughly 32 times for NVIDIA and 41 times for Broadcom. First-quarter 2026 revenue rose 41% year over year to USD 35.9 billion, with earnings up 65% to USD 3.49 per share. You are paying a meaningful discount to the chip designers for the company that has to manufacture every advanced chip the designers sell. That is an unusual setup.
The risk is geopolitical and unhedgeable. Taiwan sits where Taiwan sits. The mitigating argument is that TSMC is so embedded in the US, European, Japanese, and Korean technology supply chains that there is a strong collective interest in protecting it, and the company has been building capacity in Arizona, Japan, and Germany to diversify the manufacturing footprint. That does not eliminate the risk, but it does explain why the multiple is what it is.
My view: if you can only own one AI infrastructure name and you want something cheaper than NVIDIA without giving up exposure to the cycle, this is the obvious candidate.
3. ASML Holding (ASML): The Bottleneck Above The Bottleneck
TSMC manufactures chips. ASML manufactures the machines that let TSMC manufacture chips. ASML has a monopoly in extreme ultraviolet lithography, which is what every chip at 7nm or smaller requires. There is no second supplier, and there is no plausible second supplier on a relevant timeline. If you want true scarcity in the AI value chain, this is where you find it.
Net bookings rose 48% in 2025 to just over EUR 28 billion (about USD 33 billion), exceeding the 16% growth in annual revenue, and 2026 revenue guidance ranges from EUR 34 billion to EUR 39 billion. The stock can be more cyclical than the foundries because it sells big-ticket capital equipment that orders in lumps, but the structural position is unique.
The honest counterpoint is what one analyst flagged recently: when foundries are in heavy expansion mode, ASML does great; when foundries shift to utilizing the capacity they have already built, equipment orders slow. That is a real cycle risk, and it is why ASML can underperform the foundries at certain points in the buildout. For a multi-year holding period, though, the monopoly economics are hard to argue against.
My view: this is a sleep-well-at-night holding for someone who wants AI exposure with a moat that no amount of competitor capex can dissolve.
4. Broadcom (AVGO): The Custom Silicon Story
The bear case on NVIDIA leans heavily on hyperscalers building their own chips. Google has TPUs. Amazon has Trainium. Meta is reportedly designing its own. The interesting question is who designs those chips alongside the hyperscalers. The answer, for several of them, is Broadcom.
Broadcom has positioned itself as the custom AI accelerator partner for hyperscalers that want chips tailored to their workloads but do not want to build a chip-design organization from scratch. Broadcom has guided to USD 100 billion in AI chip revenue in fiscal 2027, up from USD 20 billion in fiscal 2025. That is a serious ramp, and it is partly insulated from the NVIDIA-versus-AMD GPU debate because it sits one layer below.
The valuation is the reason I cannot get fully enthusiastic. Broadcom trades at roughly 41 times forward earnings, the most expensive of the three large infrastructure names. The software business from the VMware acquisition gives it some defensive characteristics, but the price assumes the custom silicon ramp comes through cleanly, and any slip would compress the multiple meaningfully.
My view: this is the right name if you specifically believe custom silicon takes more share from merchant GPUs over the next three years. It is the wrong name if you think NVIDIA's CUDA moat holds and hyperscalers buy more GPUs than they design.
5. AMD: The Credible Number Two
AMD spent years being the company that almost caught up. The MI400 series and the recent Meta deal suggest "almost" is doing too much work in that sentence. AMD's MI400 series chips feature 432 GB of HBM4 memory and 19.6 TB/s bandwidth, with a reported USD 60 billion deal with Meta for MI400 deployment. That is the first major hyperscaler win at scale that did not come from NVIDIA.
Q1 2026 revenue came in at USD 10.25 billion, up 38% from last year, with the data center segment delivering USD 5.78 billion, up 57%. The growth curve has steepened, and the customer roster now includes Meta, AWS, Google Cloud, Microsoft Azure, OpenAI, and Tencent. CEO Lisa Su has been the most consistent operator in semiconductors over the past decade, and she has positioned the company well.
The two cautions I keep in mind. First, trailing P/E sits near 134, with a forward multiple closer to 54, which means the market has already priced in serious share gains. Second, NVIDIA's CUDA software ecosystem is a real moat, and AMD's ROCm has narrowed the gap but not closed it. Buying AMD is essentially a bet that the gap closes further and that hyperscalers want a credible second source enough to absorb the switching costs.
My view: this is a more aggressive way to play the same theme as NVIDIA. The upside scenario works. The downside scenario, where execution slips or pricing pressure shows up, has more compression risk than NVIDIA does at current multiples.
6. The Power And Cooling Adjacency: Vertiv, Constellation Energy, GE Vernova
This is the segment that has surprised me most. Once you understand that data centers now make up 40% of electricity demand growth and Goldman Sachs has flagged that electricity prices jumped 6.9% in 2025, the AI trade stops being purely a chip story and becomes a power story too.
Vertiv (VRT)
Vertiv sells the unglamorous equipment that keeps AI data centers running, namely power management and thermal management. Shares are up 270% over the past year, full-year 2026 EPS guidance was raised to USD 6.35, and revenue guidance is now USD 13.75 billion, a 34% increase over the prior year. Trailing twelve-month organic orders grew 81%. The cooling thesis alone is interesting because AI chips run hotter and denser than anything that came before them, and Vertiv sits in a small group of companies with the scale to address liquid cooling.
Constellation Energy (CEG)
Constellation operates the largest US nuclear fleet and has been signing direct power purchase agreements with hyperscalers. It closed the Calpine acquisition in January 2026, becoming the largest private power producer in the US at 55 gigawatts combined. Nuclear delivers 24/7 baseload at the carbon profile that Microsoft, Google, and Meta need to keep their net-zero claims credible. The thesis here is essentially that AI demand turns dispatchable clean power into a long-duration, premium-priced asset.
GE Vernova (GEV)
This is the grid-side play. GE Vernova booked USD 2.4 billion in Electrification equipment orders for data centers in Q1 2026 alone, more than all of 2025, with Q1 orders of USD 18.3 billion (up 71% organically) and an Electrification book-to-bill near 2.5. If the bottleneck on AI is no longer chips but the grid that delivers electrons, GE Vernova is a primary beneficiary.
My view on this segment: I find it more interesting than people realize. The AI trade is increasingly bottlenecked by power, not silicon, and the names that solve power problems trade at multiples that are still not as stretched as the chip designers. If you do not want to chase semiconductors at current valuations, this is a defensible pivot.
How These Names Compare At A Glance
None of these multiples are cheap on an absolute basis. They reflect a market that has decided AI infrastructure is the central theme of the cycle. The relative comparison is what matters.
| Company | Role In The AI Stack | What You Are Paying For | Main Risk |
|---|---|---|---|
| NVIDIA | AI accelerators and CUDA software platform | Dominant share, the strongest software moat in the stack, USD 1 trillion of forward demand visibility | Customer concentration, custom silicon competition, valuation already prices in continued execution |
| TSMC | Foundry that manufactures every leading-edge AI chip | 72% foundry share, multiplier effect across the whole ecosystem, lower multiple than the designers | Geopolitical exposure to Taiwan, capital intensity |
| ASML | Sole supplier of EUV lithography equipment | Genuine monopoly economics, irreplaceable position in advanced node manufacturing | Cyclical order book, slowdown when foundries digest existing capacity |
| Broadcom | Custom AI accelerator partner for hyperscalers | Direct beneficiary of in-house silicon trend, large software business from VMware | Highest forward multiple in the group, ramp is large and not yet fully delivered |
| AMD | Number-two GPU designer with credible hyperscaler wins | MI400 momentum, Meta deal, broad cloud customer roster | Software ecosystem still narrower than CUDA, valuation prices in continued share gains |
| Vertiv | Power and thermal management for AI data centers | The only large player covering both power and cooling at scale, accelerating order book | Stock has run hard, sensitivity to data center capex pacing |
| Constellation Energy | Largest US nuclear and private power producer | Direct hyperscaler power purchase agreements, scarce dispatchable clean power | Regulatory and nuclear-specific operating risk, capital-intensive uprate program |
| GE Vernova | Grid equipment and electrification infrastructure | Record electrification backlog, structural undersupply of grid equipment | Execution risk on a complex industrial business, exposure to broader power capex cycles |
The Honest Summary
If you missed NVIDIA, the good news is that you did not miss the AI infrastructure cycle. The cycle is still playing out across foundry, lithography, custom silicon, GPU competition, power generation, cooling, and grid build-out. There are at least seven defensible entry points that are not NVIDIA.
The honest news is that "I missed NVIDIA" is usually the wrong frame. NVIDIA's earnings have grown into the multiple every year, and the people loudest about valuation in 2023 and 2024 mostly missed two more years of compounding. If your real concern is concentration in a single name, build a basket. If your real concern is that the thesis is over, look at hyperscaler capex projections and ask yourself whether USD 700 billion of annual spending really stops in 2027.
My personal preference, if I were structuring exposure today rather than advising clients on capital, would be a barbell: TSMC for the multiplier effect at a reasonable multiple, NVIDIA for the platform position, and one of Vertiv or Constellation for the power adjacency. That is three names instead of one, and it spreads the risk across the parts of the AI stack that have to keep working together.
Take this as opinion. Take it as one person's framing. Do not take it as advice you can act on without doing your own work.
Full Disclosure: This article reflects personal opinion and is provided for general informational purposes only. It is not investment advice, financial advice, tax advice, or legal advice. It is not a recommendation to buy, sell, hold, or refrain from transacting in any security. FG Capital Advisors is a capital markets advisory firm focused on debt advisory and capital introduction; it is not a registered investment adviser, broker-dealer, or licensed financial planner. The author does not hold positions in the securities discussed as part of FG Capital Advisors business activities and is not compensated by any of the companies named. Equity investments involve risk, including the risk of total loss. Past performance does not predict future results. Forward-looking statements quoted from third-party sources reflect those sources' views, not the author's. Before making any investment decision, consult a licensed financial professional who can evaluate your specific circumstances, objectives, and risk tolerance.
