AI investment enters the second phase: Why has NVIDIA become the core of market repricing?

Eastern Time on June 29, 2026, the NASDAQ Composite Index surged 522.52 points, up 2.07%, closing at 25,820.14 points. NVIDIA rose 1.27% on the day, closing at $194.97. Its market value is approximately $4.72 trillion. But just a few trading days earlier, the world’s most valuable semiconductor company had just gone through five consecutive days of declines.

Behind the short-term volatility in the stock price, a deeper structural change is underway: AI investment is moving from the “storytelling” stage into the “accounting” stage. The market is no longer only asking “who is participating in AI,” but has started asking “who can truly make money from AI.” This shift is redefining the valuation logic across the entire AI industry chain—from chips to cloud services—and NVIDIA sits at the center of this repricing storm.

From “Compute Scarcity” to “Return Verification”: The Deep Shift in AI Investment Logic

Over the past three years, the AI industry has followed a clear and powerful logic line: the scarcer the compute, the more rational capital expenditures are; the larger the capital expenditures, the higher the valuation. This self-reinforcing cycle has been challenged by almost no one. However, entering 2026, every link in this logical chain is being stress-tested.

The most core change comes from the demand side. Financial reports show that in 2026, the combined capital expenditures of four hyperscale cloud giants—Google, Amazon, Microsoft, and Meta—rose to $725 billion, up 77% year over year from $410 billion in 2025. Goldman Sachs tracking data is even more granular: in just about the past six months, market expectations for 2026 cloud providers’ capital expenditures were revised up by nearly 80%, from approximately $520 billion to $772 billion. Barclays expects that major cloud providers’ capital expenditures will reach $919 billion in 2027, and further rise to about $1.16 trillion in 2028.

But the scale of spending is no longer the only focus of the market. In a research note published in June, Goldman Sachs explicitly pointed out that the core contradiction in the AI market is intensifying—fundamentals remain strong, but the market has already priced in too much of future earnings. The share of U.S. tech investment in GDP has risen to about 4.9%, surpassing the peak around the internet bubble era in the early 2000s. The market’s pricing pace for future AI returns is clearly faster than the pace at which productivity dividends are truly realized.

It is in this context that the AI industry has reached a critical threshold. According to the Exponential View report, as of the first quarter of 2026, the global generative AI industry (excluding China) saw quarterly revenue for the first time exceed the depreciation expenses of AI infrastructure in the same period. Annual AI infrastructure depreciation expenses in 2026 are expected to approach $111 billion. In other words, the cash flow generated by AI businesses can already cover the accounting depreciation costs formed by servers, GPUs, and data centers—the industry has crossed the first gate of being “able to support itself.”

However, there is still a considerable gap before proving that the entire capital cycle can deliver reasonable returns. The report projects that by the end of 2026, cumulative AI-related capital expenditures from global hyperscale cloud providers and emerging AI cloud platforms will reach approximately $2 trillion. The market is shifting from “faith in compute scarcity” to a systematic examination of return on investment.

NVIDIA’s Industry Coordinates: From Market Share to an Ecosystem Moat

In this round of valuation reshaping, the core question facing NVIDIA is: as the market shifts from chasing “AI themes” to validating “AI performance,” is its industry position strong enough to support its current valuation level?

From the perspective of market share, NVIDIA’s dominance in the AI accelerator market remains solid. As of early 2026, NVIDIA controlled about 81% to 90% of the share in the AI accelerator and data center chip market. In the core area of AI training, its share is even higher, about 85% to 90%. Although with AMD scaling up and hyperscale cloud service providers deploying custom chips (ASICs), the overall market share is expected to fall to about 75% by 2026, absolute revenue figures are still growing—because the pace of expansion of the total addressable market is far faster than any single competitor’s ability to capture it.

Financial data provides even stronger support. NVIDIA’s fiscal year 2026 (as of January 2026) full-year revenue was $215.94 billion, up 65% year over year. Of that, data center revenue reached $194 billion, up 68% year over year. In the first quarter of fiscal year 2027 (as of April 2026), the company’s revenue climbed further to $81.6 billion, up 85% year over year, with data center revenue reaching $75.2 billion. The company’s gross margin has been maintained at around 75%.

More importantly is demand visibility. Institutions such as Wedbush, Citi, and BofA point out that NVIDIA’s order backlog for the Blackwell and Rubin architectures is approximately $500 billion, with visible demand exceeding $1 trillion before 2027. Recently, the company raised its revenue opportunity forecast for 2027 from $500 billion to $1 trillion.

But the competitive landscape is evolving. Inference is becoming the main battleground for incremental share—ASICs and XPUs are growing far faster than GPUs. In 2026, the expected growth rate for ASIC server shipments is 44.6%, while GPU server shipments are expected to grow only 16.1%. Broadcom is expected to capture about 60% of the custom AI chip market share, and its 2026 full-year AI revenue guidance has already reached $56 billion. The share of inference in total AI compute demand has risen from one-third in 2023 to two-thirds in 2026.

NVIDIA’s response strategy is twofold. On the one hand, the company continues to increase investment in research and development, planning R&D spending of $45 billion for fiscal year 2027. On the other hand, it strengthens its inference-dedicated chip capabilities by acquiring Groq ($17 billion). TrendForce data shows that in 2026, NVIDIA’s expected year-over-year growth in shipments of high-end GPUs is nearly 26%, and the shipment mix share of the Blackwell series is set to jump from 61% to 71%.

Reconstructing Valuation Logic: From Price-to-Sales Dreams to Price-to-Earnings

In 2024 and 2025, NVIDIA’s forward P/E ratio was as high as 35 to 40 times. But by June 2026, this multiple had fallen significantly. Based on a share price of $192.53 and consensus earnings per share of about $9.34 for fiscal year 2027, NVIDIA’s forward P/E ratio is approximately 21 times. The TTM P/E ratio is about 29.8 times.

This valuation compression itself is the most direct reflection of the market’s shift from “thematic investing” to “earnings verification.” Morgan Stanley analyst Joseph Moore outlined three scenarios: a base case of $250; an upside to $330 if NVIDIA executes its roadmap; and a downside to $150 if AI infrastructure spending slows faster than expected. Over a 12-month horizon, the range from $150 to $330 is driven almost entirely by P/E multiples rather than disagreements about near-term revenue.

According to analyst consensus, among 38 analysts, NVDA receives a “Strong Buy” rating, with an average 12-month target price of about $300. In May, UBS raised its target price from $245 to $275, based on a 2027 EPS estimate of $14.35 and a 19x P/E ratio. Lyon maintains a “High Conviction Outperform” rating, based on a fiscal year 2028 forecast P/E ratio of 32x and a target price of $300. CICC raised its target price to $268.30.

But valuations are not without downside risks. Goldman Sachs notes that the core assumption supporting the current valuation of the compute supply chain is the “perpetual growth of CAPEX” by tech giants. Once that assumption loosens, valuation adjustments are inevitable even if demand fundamentals remain fine. Current AI valuations are already expensive, and the more optimistic assumptions are stacked, the higher the risk—any crack in the narrative could trigger market volatility.

High Interest Rates and Capital Constraints: The Ceiling on Valuations

The macro environment is creating new constraints on AI investment. Goldman Sachs data shows that in 2026, hyperscale cloud providers’ capital expenditures as a proportion of operating cash flow will rise to about 100%—meaning these companies are reinvesting almost all of their internal cash flows back into AI infrastructure. Barclays’ calculations are more detailed: the proportion of operating cash flow allocated to capital expenditures by major cloud providers will rise from 61% in 2025 to 91% in 2026 and 92% in 2027.

This means that even with strong fundamentals, further expansion in capital expenditures faces increasingly hard constraints. In a high interest rate environment, the market requires higher discounting of future cash flows and has less tolerance for longer payback periods. Morgan Stanley expects the overall total addressable market size for AI infrastructure to reach $3 to $4 trillion per year by 2030—but achieving this number depends on sustained and verifiable returns.

At NVIDIA’s June 2026 shareholder meeting, Jensen Huang said that the question of AI investment returns “already has an answer.” He cited financial data: fiscal year 2026 revenue grew 65% to $216 billion, and operating cash flow reached $103 billion. Nearly 40 countries and regions, representing a combined $50 trillion GDP, are building AI factories driven by NVIDIA’s infrastructure. But the market needs more than past performance; it needs continuous validation of the future.

The Second Phase of AI Investment: Differentiation and Screening

Goldman Sachs strategists believe that Wall Street’s AI trades are entering a more complex phase: the market still believes in the AI investment cycle, but it no longer places all AI companies within the same valuation framework. AI trades have moved from thematic investing to a return-on-investment verification phase.

In this stage, sharp differentiation is occurring within the industry chain. The upstream compute layer faces more rigid demand, so earnings realization is relatively faster, partially offsetting valuation pressure; the midstream and downstream application layers generally face the difficulty of long commercialization cycles and slow revenue recognition, meaning valuation digestion requires more time. The upstream of the AI industry chain mainly includes core areas such as AI chips, storage, and optical modules—large-model training and inference demand continues to grow, global cloud providers’ capital expenditures continue to increase, and this drives high-speed earnings growth in the upstream sector.

The upstream compute layer where NVIDIA operates is precisely the segment with the fastest earnings realization. But its challenges are equally clear: the ceiling on market share, erosion from custom chips, and the marginal slowdown in the growth rate of capital expenditures. Together, these factors determine that NVIDIA’s valuation logic is shifting from “a market share premium” to the validation of “earnings sustainability and cash flow quality.”

On June 29, 2026, NVIDIA closed at $194.97. This price is already quite a distance from its 52-week high of $236.54. But more important than the specific price is the more fundamental question the market is asking with it: when AI moves from “a story” to “a business,” who can continuously prove themselves in real profit statements?

The answer to this question will determine the valuation anchor for NVIDIA and the entire AI industry chain over the next few years.

FAQ

Q1: What is the essential difference between current AI investment and the internet bubble era?

Goldman Sachs analysis indicates that the share of corporate profits in the United States relative to GDP remains near historical highs (about 14%), and the growth rate of wages and unit labor costs is slower than in the late 1990s. Although cloud providers are pouring large amounts of cash flow into AI infrastructure, the issue has not yet spread to the entire corporate sector; the financial balance of the overall non-financial corporate sector has not shown any obvious deterioration. The rapid growth in corporate earnings currently provides some real support for high valuations.

Q2: Does NVIDIA’s competitive position in the AI chip market face a substantive threat?

NVIDIA still holds about 85% to 90% share in the AI training market, but on the inference side it faces competition from ASICs and XPUs. In 2026, the expected growth rate for ASIC server shipments is 44.6%, far exceeding the 16.1% growth rate for GPU server shipments. However, NVIDIA strengthens its inference capabilities through R&D investment (planned $45 billion in fiscal year 2027) and the acquisition of Groq; while long-term market share may decline, absolute revenue will still grow.

Q3: Is NVIDIA’s current valuation level reasonable?

Based on NVIDIA’s June 2026 stock price, its forward P/E ratio is about 21x, and its TTM P/E ratio is about 29.8x. This level is far below the 35–40x multiples seen in 2024–2025. Analysts’ average 12-month target price is about $300. But the downside risk comes from the slowdown in capital expenditure growth—Morgan Stanley sets a $150 target price in its most pessimistic scenario.

Q4: How long is the return cycle for AI infrastructure investment?

As of the first quarter of 2026, quarterly revenue in the AI industry exceeded depreciation expenses for the first time. But by the end of 2026, cumulative AI-related capital expenditures by global hyperscale cloud providers are expected to reach approximately $2 trillion. The industry has crossed the threshold of being able to “support itself,” but there is still a gap before proving that the entire capital cycle can deliver reasonable returns. AI revenue remains on track with a year-over-year growth rate of about 200%.