How AI and Intelligent Robots Will Reshape Employment and Wealth Patterns in the Next Decade: A Full Analysis of Affected Asset Classes
2026年7月9日China Allows Alibaba, ByteDance to Buy H200: Why Did Nvidia Only Rise 1%? What Conditions Are Needed for the Stock to Double Again?
⚠️ Disclaimer: This content is for informational purposes only and does not constitute investment advice or financial guidance. All analysis and opinions are the author’s personal judgment and do not represent any institutional position. Investing involves risk; readers should assess risks independently and consult professionals when necessary.
The Chinese government plans to approve limited purchases of Nvidia H200 chips by Alibaba, ByteDance, and DeepSeek. Following the news, Nvidia’s stock rose only about 1%. The market’s muted reaction is not a failure but a precise pricing of five core constraints. For Nvidia’s stock to double from its current market cap of approximately $5 trillion, the financial conditions required far exceed the scope of this 200,000 H200 chips.
Is the H200 Nvidia’s latest chip? How big is the gap with the most advanced?
The H200 belongs to the Hopper architecture (previous generation), same generation as the H100, with the main upgrade being the switch from HBM2e to HBM3e memory, significantly increasing capacity and bandwidth, but the compute architecture remains unchanged. Nvidia’s current most advanced product line is the Blackwell architecture (B100, B200, GB200 NVL72 rack system), with AI training and inference performance several times higher than the H200.
Chinese companies previously could only purchase Nvidia’s downgraded versions (H20, L20). Although the H200 is a previous generation, its performance far exceeds these special chips, and it is fully compatible with the CUDA ecosystem, ready for large model training. This is the core reason why China’s top AI companies still desperately want the H200.
Why is the Chinese government only now approving the H200? Is Alibaba’s inclusion on the U.S. blacklist relevant?
The two events are contextually related but driven by different logics:
The U.S. had already approved in principle in May 2026: The U.S. Department of Commerce approved about 10 Chinese companies to purchase H200 chips during the U.S.-China summit, with each company limited to about 75,000 chips. The U.S. restriction had already been set.
The bottleneck was within the Chinese government: Due to supply chain security reviews and policy support for domestic chips (Huawei Ascend), China had been holding back approvals for domestic companies to place orders. The current approval is because the computing power gap for cutting-edge large model training has reached a critical point, and relying solely on domestic chips has clear bottlenecks.
Alibaba was added to the U.S. Pentagon’s 1260H “Chinese military company” blacklist in June 2026, which restricts business dealings with U.S. entities but does not directly block chip procurement channels. The bottleneck for the H200 has always been Chinese government approval, not U.S. sanctions.
Why did Nvidia only rise 1%? Five pricing reasons
① Expectations already priced in: The news of U.S. approval for H200 exports was already reflected in the stock price two months ago. This is just confirmation of China’s approval, the tail end of “buy the rumor, sell the fact.”
② Limited revenue contribution from total purchases: No more than 200,000 H200 chips, at an average price of about $30,000 each, total value approximately $6 billion. Nvidia’s fiscal year 2026 revenue is $215.9 billion, so this accounts for about 2.8%, and deliveries will be phased.
③ Inventory reduction, not a new surge in orders: The H200 is a previous generation; Wall Street focuses on Blackwell architecture’s global production capacity and gross margins. H200 shipments improve inventory turnover but do not raise Nvidia’s future technology ceiling or valuation flexibility.
④ Strict usage restrictions weaken long-term value: The chips can only be used for model training; inference must still use domestic chips like Huawei Ascend. This signals to the market that China has not abandoned its chip self-sufficiency path, and Nvidia’s long-term market share in China cannot be completely reversed by this special approval.
⑤ Technical resistance: Nvidia’s stock faces strong resistance near $195–200, in a high-range consolidation zone. A single geopolitical news event is unlikely to break through technical resistance.
[Bullish Forecast] What conditions are needed for Nvidia’s stock to double? What is Wall Street’s math?
Nvidia’s current market cap is about $5 trillion, with a forward P/E ratio of about 30–35x. Doubling the stock price would mean a market cap exceeding $9–10 trillion. Assuming the valuation multiple remains unchanged, net profit and revenue must double simultaneously, requiring annual revenue to increase from $215.9 billion to about $400–430 billion, an additional $200 billion in revenue.
At an average price of $35,000 per Blackwell B200 chip:
$200 billion ÷ $35,000 ≈ 5.71 million B200 chips
This means the world must stably deliver 5–6 million top-tier GPUs annually, with corresponding supply chain requirements:
| Supply Chain Segment | Conditions Needed for Doubling |
|---|---|
| TSMC CoWoS Packaging | Monthly capacity must increase to 80,000–100,000 wafers, approaching the limit of advanced packaging globally |
| HBM3e Memory (SK Hynix/Micron/Samsung) | Nearly 50% of the highest-end HBM capacity must be dedicated to Nvidia |
| CapEx of Four Hyperscalers | Microsoft/Meta/Amazon/Google must collectively maintain over $500 billion in AI CapEx annually |
| Gross Margin | Must remain in the 73–76% range, not declining due to rising HBM costs or competitive pressure |
Wall Street’s core question: Can the four hyperscalers, spending $50 billion annually on chips, generate enough AI software monetization (subscriptions, API calls) to cover the return on investment (ROI) of these capital expenditures? If monetization falls short, orders may be cut, and Nvidia’s valuation bubble would be the first to burst.
How do the four giants pay for $50 billion in annual GPU purchases? What are the payment terms?
Nvidia holds a monopoly in global AI computing power, and payment terms are dictated by Nvidia. The main models:
- Prepayment to lock capacity: Buyers must pay billions of dollars in deposits several quarters in advance. Nvidia uses this money to secure CoWoS packaging capacity and HBM memory quotas from TSMC.
- Very short payment terms: Cash settlement within 30–45 days after delivery; no 120-day terms. Default means losing queue position.
- No equity payments: Issuing shares dilutes EPS, a major taboo in U.S. stocks. Shareholding customers create conflicts of interest and high compliance risks.
OpenAI obtains computing power through Microsoft’s “proxy purchase” model (Microsoft buys chips to build Azure, converting them into computing power investments for OpenAI); xAI directly purchases with full cash payments through independent fundraising (single rounds of $6 billion).
Won’t the four tech giants’ $50 billion annual spending drag down their financials?
Profits can support it, but free cash flow is already showing pressure:
| Company | Q1 2026 Net Profit | Trailing 12-Month Net Profit |
|---|---|---|
| Microsoft | $31.8 billion | ~$120 billion |
| Google (Alphabet) | ~$30 billion | ~$113 billion |
| Meta | $26.77 billion | ~$70 billion |
| Amazon | ~$18 billion | ~$67.6 billion |
Combined annual net profit of the four is close to $370 billion. Chip purchases are recorded as capital expenditures (CapEx), not directly deducted from profits. With depreciation over 4–5 years, the annual income statement only bears depreciation expenses.
The real pressure is on free cash flow (FCF): Amazon’s full-year 2026 CapEx plan is about $200 billion, and TTM free cash flow has plummeted from $25.9 billion to about $1.2 billion. Google urgently issued $31.1 billion in bonds to supplement liquidity. Meta, after raising its CapEx cap to $145 billion, saw its stock drop about 7% after hours following its earnings call.
The giants choose debt over equity issuance: They have high credit ratings (Microsoft AAA), low bond yields, tax-deductible interest, and debt maturities can be extended to 10–30 years to spread repayment pressure. Issuing shares dilutes EPS, a major capital market taboo.
Conclusion
China’s approval of the H200 is a policy easing signal, but its substantive contribution to Nvidia’s revenue is limited. The core driver for the stock to double is whether Blackwell’s annual production capacity can reach 5–6 million units and whether the four hyperscalers’ AI subscriptions and enterprise services can generate ROI sufficient to cover trillion-dollar CapEx within 2–3 years. Drying free cash flow and rising debt are the most critical risk indicators to monitor in this AI computing arms race.
Frequently Asked Questions
Q: How big is the performance gap between H200 and Blackwell?
A: Blackwell (B200) is several times more powerful than H200 in AI training and inference, representing a completely different generation. H200 is a memory upgrade of the Hopper architecture, while B200 uses the new Blackwell architecture with updated process and interconnect systems.
Q: How much revenue can 200,000 H200 chips bring to Nvidia?
A: At an average price of about $30,000 per chip, the total value is about $6 billion, accounting for about 2.8% of Nvidia’s fiscal 2026 revenue of $215.9 billion. Deliveries are phased, so the marginal contribution to quarterly revenue is limited.
Q: What is the most critical condition for Nvidia’s stock to double?
A: Annual revenue must increase from $215.9 billion to over $400 billion, requiring an additional $200 billion, which means selling about 5.71 million B200 chips. The core bottlenecks are TSMC’s CoWoS packaging capacity and HBM3e memory supply, as well as whether the four hyperscalers’ AI monetization can sustain massive CapEx.
Q: Why do tech giants choose to issue bonds rather than stocks to buy GPUs?
A: Issuing shares dilutes earnings per share (EPS), a major taboo in U.S. stock valuation, as the market interprets it as a negative signal of insufficient core business cash flow. Bonds have low yields, tax-deductible interest, and long maturities, without affecting EPS, making them a standard financial tool for mature large companies.
Q: What catalysts should be monitored for Nvidia after this news?
A: Quarterly earnings of the four hyperscalers (especially AI CapEx and cloud revenue growth), TSMC’s CoWoS expansion timeline, the production progress of the next-generation Blackwell Vera Rubin architecture, and dynamic adjustments to U.S. Department of Commerce regulations on H200 exports.
Original Article
News: China Plans to Allow Alibaba, ByteDance, DeepSeek to Limitedly Purchase Nvidia H200 Chips
U.S. media cited sources saying that Beijing plans to allow several top Chinese AI companies to purchase a limited number of Nvidia H200 chips, indicating that China is easing restrictions on the highly sought-after products of this U.S. chip giant.
U.S. tech media “The Information” reported the news on Wednesday evening (July 8). Reuters and Bloomberg cited reports that Chinese officials told tech giants Alibaba, ByteDance, and AI star company DeepSeek in recent weeks that they may soon receive permission to purchase some H200 chips. These companies need to specify the quantity and purpose of the chips required to obtain approval.
After the report, Nvidia’s stock rose 1%.
Source: Lianhe Zaobao (https://www.zaobao.com.sg/finance/china/story20260708-9335122)
⚠️ Disclaimer: This content is for informational purposes only and does not constitute investment advice or financial guidance. All analysis and opinions are the author’s personal judgment and do not represent any institutional position. Investing involves risk; readers should assess risks independently and consult professionals when necessary.