In November 2025, SoftBank Group’s decision to liquidate all its Nvidia shares for $5.83 billion sent shockwaves through the global technology capital market. As a seasoned investor who has entered and exited Nvidia twice and witnessed its market capitalization soar from hundreds of billions to trillions, SoftBank’s high-level cash-out was no coincidence but a precise prediction of the latent risks in the AI hardware sector. This capital withdrawal reflects multiple concerns regarding the AI hardware industry in terms of policy, competition, valuation, and ecosystem, sounding the alarm for the fervent AI investment boom. SoftBank’s liquidation decision starkly contrasts with Nvidia’s market value fluctuations. By the end of 2025, this AI chip giant experienced a trillion-dollar market value shrinkage twice within a month, with a cumulative loss comparable to the total market value of three Kweichow Moutai, as the once high-growth myth began to fade. SoftBank’s exit is not an isolated case; legendary short-seller Michael Burry held $200 million in put options to short Nvidia, compounded by the executive team’s cumulative $700 million reduction in holdings, signaling a collective indication of risk accumulation in the AI hardware industry. As the core hub of the AI industry chain, Nvidia’s volatility is essentially a concentrated outbreak of industry contradictions, and SoftBank’s precise exit provides the market with the best lens to examine the risks of AI hardware.
Policy Uncertainty: The Sudden “Supply Cut” Risk in Core Markets
The global layout of AI hardware has always faced dual pressures from geopolitical and policy regulation, which has become the most uncontrollable systemic risk in the industry. In April 2025, the U.S. intensified export restrictions on high-end AI chips, directly severing Nvidia’s supply channel to China, which had contributed nearly 30% of its data center revenue. The policy blockade led to $2.5 billion in unfulfilled orders for Nvidia, with revenue from the Chinese market plummeting 24% year-on-year, forcing a $4.5 billion inventory impairment. The once “cash cow” instantly became a performance burden. This policy risk has a strong transmission effect, impacting not only individual companies but also triggering chain reactions throughout the industry chain. For AI hardware companies that heavily rely on a single market, encountering trade barriers could completely collapse existing capacity planning, customer cooperation, and revenue expectations. SoftBank clearly recognized the irreversibility of this risk—against the backdrop of global cooperation facing obstacles, the market expansion of AI hardware companies could be stalled at any time due to policy changes, which was a crucial consideration for its decisive liquidation.
Intensifying Competition: The Impact of Challengers in a Monopolistic Landscape
The era of dominance in the AI hardware industry is coming to an end, as competition from both inside and outside is dismantling the traditional giants’ monopolistic advantages, constituting the most direct survival risk in the industry. Nvidia once monopolized the high-end AI chip market with a 94% market share, but is now facing a dual-line siege: externally, AMD is leveraging the high cost-performance of its MI300X chip, achieving 80% of H100 performance at 30% lower prices, with a threefold increase in shipments in the third quarter, capturing 18% of the market share; internally, cloud giants are undermining Nvidia’s position with self-developed alternatives, such as Google’s TPUv5 and Amazon’s Trainium2, which are cheaper and more adaptable, expected to capture 15%-20% of the high-end market within three years. More critically, the “de-Nvidia” trend is impacting the ecological barriers. Chinese companies are being forced to accelerate the replacement process, with Huawei’s Ascend 910C achieving 80% of H100 performance at one-tenth the inference cost, attracting giants like ByteDance and Tencent to adapt, causing Nvidia’s market share in the Chinese AI chip market to drop from 70% to 54%. The download volume of algorithm libraries compatible with AMD’s ROCm architecture has also increased by 178% quarter-on-quarter, indicating that Nvidia’s CUDA ecosystem’s monopolistic advantage has weakened from 90% dependency to 75%. As the performance gap in hardware continues to narrow and ecological dependency gradually decreases, the competitive barriers for AI hardware companies will sharply weaken, which is the core concern of investors like SoftBank.
Valuation Bubble: The Rational Demystification of High Growth Expectations
The valuation logic of the AI hardware industry is facing reconstruction as growth slows, with excessive valuation bubbles hanging over the industry like the “Sword of Damocles.” Nvidia’s dynamic price-to-earnings ratio once reached 52.85 times, far exceeding the semiconductor industry’s average of 25 times, and this high valuation was built on the optimistic expectation of “perpetual growth in AI spending.” However, real data has begun to falsify this assumption. Goldman Sachs warned that global AI capital expenditure growth is expected to halve from 60% to 30% in 2026, while Nvidia’s second-quarter data center revenue growth has slowed from 73% to 56%, with year-on-year revenue growth plummeting from a peak of 279% to 101%. The contradiction between slowing growth and high valuations becomes more pronounced during macroeconomic shifts. In September 2025, the unexpected rise in U.S. PPI raised concerns about interest rate hikes, with the 10-year U.S. Treasury yield breaking through 4.7%, a 16-year high, causing a significant flow of funds from overvalued tech stocks to defensive assets. For AI hardware companies, their performance is highly dependent on capital expenditure, and when market liquidity tightens and investor risk appetite declines, excessive valuations will be difficult to maintain. SoftBank’s cash-out while Nvidia’s stock price was still close to its historical high was a precise grasp of the valuation correction risk after the “high growth myth” faded, avoiding losses from a potential bubble burst.
Ecological Dependency: Systemic Risks of a Single Architecture
The core competitiveness of AI hardware lies not only in the hardware itself but also in the software ecosystem it builds. However, this ecological dependency can also become a “fatal flaw” for companies. Nvidia’s rise is supported by the CUDA ecosystem, but as “de-Nvidia” becomes a trend, the ecological advantage may turn into a burden for transformation. Companies need to invest substantial resources to maintain existing ecosystems while also facing the ecological breakthroughs of competitors, which will significantly increase operational costs. More alarmingly, AI hardware is deeply bound to downstream application scenarios; once downstream demand cools, it will directly trigger inventory accumulation and overcapacity in the industry. Nvidia has already faced inventory accumulation of older H100 chips, and the inability to timely mass-produce the next-generation chips due to TSMC’s capacity constraints is a direct signal of cooling demand. Concerns about AI infrastructure construction are also beginning to surface, as a conference call from Nvidia’s subsidiary CoreWeave triggered market panic, leading to an almost 8% drop in its stock price. This “hardware-ecosystem-demand” chain dependency makes the risks in the AI hardware industry highly transmissible and amplifying.
SoftBank’s liquidation of Nvidia is not a denial of the AI track but a rational avoidance of the phase risks in the AI hardware industry—Masayoshi Son is simultaneously planning to invest $30 billion in OpenAI and promote the construction of a trillion-scale AI manufacturing center, which precisely indicates that capital’s long-term confidence in AI remains. However, this capital action undoubtedly provides important insights for the market: the AI hardware industry has bid farewell to the wild growth dividend period, with policy fluctuations, intensified competition, valuation corrections, and ecological transformations becoming the norm in industry development. For investors, it is essential to abandon the mindset of blindly chasing highs and focus on companies with core technological barriers, diversified market layouts, and stable cash flows; for AI hardware companies, continuous improvement in technological innovation, ecological openness, and risk resistance capabilities is necessary to stand firm amid industry reshuffling. The development prospects of AI hardware remain broad, but the growth of any emerging industry inevitably comes with risks and challenges. SoftBank’s liquidation action is merely a microcosm of the industry’s return from frenzy to rationality, and only industries that have undergone the baptism of risk can usher in healthier and more sustainable development.