Two years ago this week, Nvidia closed at $1,208.88 on its final pre-split trading day - then split 10-for-1 and reopened at $120.88. The split made the stock feel accessible without changing anything about the company. Today, the post-split share trades around $211, and the market cap sits near $5.1 trillion. The headline question circling investor forums is whether Nvidia can reach $1,000 per share again.
The answer, mathematically, is almost certainly not - and that's not the point. The real question is whether the architecture transition Nvidia is riding right now justifies keeping capital deployed here, or whether the return profile is better found elsewhere in the AI trade.
The $1,000 math tells you what the market would have to believe
Nvidia has roughly 24.3 billion shares outstanding. For the stock to reach $1,000 per share on a post-split basis, the company would need a market capitalization of approximately $24.3 trillion. That is roughly 4.7 times the current $5.1 trillion valuation.
Even with the $80 billion share buyback Nvidia announced in May - its largest ever - share count reduction alone won't bridge that gap. To see $1,000 per share without additional market cap growth would require retiring roughly 17 billion shares, or about 70% of the float. At $80 billion per year in buybacks and a $5 trillion market cap, it would take decades.
Put plainly: the $1,000 question is a valuation question in disguise. For it to happen, you need either a $24 trillion market cap, a share count collapse, or both. Neither is on the table in any timeframe an investor would realistically wait for.
But again - that's not the question I'm trying to answer.
What actually matters: Nvidia is on the right side of the inference transition
The market is shifting from AI training to AI inference - from building models to running them at scale. Training dominates the current chip spending conversation, but inference is where the revenue tailwind is heading. This is the structural transition that determines who wins the next three to five years of AI infrastructure buildout.
Nvidia's positioning for this shift is deliberate and layered. At GTC 2026 in March, Jensen Huang unveiled the Vera Rubin platform - not just a GPU upgrade, but a seven-chip system architecture comprising the Vera CPU, Rubin GPU, NVLink 6 Switch, and purpose-built inference chips. The platform is designed for agentic AI at industrial scale. Volume shipments are locked for the second half of 2026.
Huang stated he sees at least $1 trillion in visible orders for Blackwell and Vera Rubin systems combined through the pipeline. On the most recent quarter - fiscal Q1 2027, ended April 26 - Nvidia delivered $81.6 billion in revenue, up 85% year over year, and guided to $91 billion plus or minus 2% for the next quarter, well above the $86.8 billion consensus. Full fiscal 2026 revenue came in at $215.9 billion, up 65% from a year earlier.
The architecture comparison is where the competitive moat story gets tested. Vera Rubin full racks are projected at 3.3 times the performance of comparable Blackwell Ultra configurations. The all-copper scale-up network in the Rubin NVL72 form factor is an engineering differentiator that competitors haven't matched - and the Rubin Ultra variant is already in development. Beyond that, the next architecture generation - Feynman, with the new Rosa CPU - is on the board.
This product cadence matters because in the inference era, efficiency and latency matter more than raw compute horsepower. The CUDA moat that made Nvidia untouchable in training is formidable but not impenetrable in inference, where optimization paths are more open. Nvidia's answer isn't to lean solely on CUDA - it's to make the entire stack so integrated that switching costs remain prohibitive.
The supply chain signal: demand is not the problem, packaging is
Here is the part most investors miss. Nvidia's growth trajectory is not constrained by demand. It is constrained by TSMC's CoWoS packaging capacity - the process that packages Nvidia's GPUs with high-bandwidth memory into the advanced 3D structures hyperscalers are buying.
TSMC CEO C.C. Wei has said CoWoS capacity "remains sold out through 2025 and into 2026". TSMC is aggressively expanding - the company is effectively doubling CoWoS throughput through 2026 - but Nvidia already secured roughly 60% of global supply in this segment. The bottleneck is real, and it is a dual signal: it confirms demand strength, but it also creates execution risk.
What this means for investors is that Nvidia's revenue growth may be supply-capped even when orders exceed capacity. That changes the risk profile. When supply constraints ease later in 2026 as CoWoS expansion comes online, the revenue inflection could be sharp. Until then, there is an asymmetry - the upside from capacity release is large, but a sustained packaging shortfall would delay Vera Rubin revenue recognition and compress the growth timeline.
Demand is robust. The risk is in the bottleneck, not the buyer.
So where does the capital go?
Here is how I'm thinking about it. Nvidia's long-term thesis - the $15 trillion to $20 trillion market cap case by 2030 - remains structurally intact if the inference transition delivers as projected. The Vera Rubin platform, the $1 trillion order pipeline, and the product roadmap extending into Feynman all support that trajectory. I believe the company is on the right side of the current compute cycle.
However, getting from $5.1 trillion today to $20 trillion by 2030 requires roughly a 300% return over roughly four years. Much of that return curve is likely back-half weighted in 2028-2030, once Vera Rubin volume shipments accelerate and inference revenue becomes the dominant mix. The near-term return profile - the next 12 to 18 months - is narrower. The stock already reflects 85% revenue growth, dominant market share, and a $200+ billion annual revenue run rate.
The debate is not about whether Nvidia stays important. It is about whether the return profile is still as compelling as what can be found elsewhere in the AI trade at this stage. For a concentrated position that already moved from $120 to $211 since the split, the math is no longer about catching a wave - it's about whether the wave is big enough to justify the allocation.
If you're positioned here and the thesis hasn't changed, I would hold through the Vera Rubin ramp and watch CoWoS capacity as the leading indicator of when the next leg of revenue acceleration begins. If you're looking to enter, the setup is less attractive than it was two years ago at $120 - the market has digested most of the obvious thesis, and the remaining return is back-loaded. In that case, a smaller position or rotating into the inference supply chain - memory, packaging, networking - may offer a better risk/reward for the next cycle.
The stock split never mattered. The architecture transition does.