Summary

From NVDA’s 17% one-day plunge after DeepSeek’s shock to NVIDIA’s market cap breaking $4T and the Nasdaq and S&P 500 hitting new highs, the market seemed to prove in the first half of 2025 that concerns over compute spending could be set aside for now. Agents, reinforcement learning, multimodality, and AI coding continue to consume compute, while the bearish narrative epitomized by Microsoft’s decision to cut two data centers has faded; 庄明浩’s summation was, “It seems everything is back.” But as consensus again moves to an extreme, new risks may be brewing.
Stargate was not knocked off its compute-driven logic by DeepSeek; the real bottleneck has shifted from “Should we keep piling on compute?” to “Who pays for it?” OpenAI is expected to generate only $10B–$13B in revenue for the year, yet signed a $30B-a-year data-center services agreement with Oracle; whether Microsoft, Google, Amazon, Facebook, Oracle, or another source of capital ultimately picks up the tab is more unsettled than the technology trend itself. “People are no longer discussing the trend… someone will always foot the bill.”
The projects now being placed under the Stargate umbrella no longer appear to be the clearly defined standalone entity announced in January. The original plan called for outside investors and shareholders to provide funding, Oracle to build the infrastructure, and a formal operator to participate; SoftBank is not part of that original plan, while OpenAI and Microsoft’s interests and financing arrangements, SoftBank’s financing of OpenAI, SoftBank’s own fundraising, and its Stargate commitments overlap and nest inside one another, forming “a knot that cannot be untangled.”
SoftBank is not an ordinary financial investor; it is simultaneously betting the group’s future, ARM, AI deployment in Japan, and its relationship with the Japanese government. 庄明浩 did not call failure “fatal,” saying only that it would be “quite troublesome”; Masayoshi Son’s bet is characteristically aggressive, but today’s top funds are all at the table for fundraises of hundreds of millions and valuations of billions, with “a zero added to the size of the chip,” making SoftBank’s large checks no longer unique.
Oracle remains the clearer picks-and-shovels seller in this expansion, because it does not need to pick the model that ultimately wins. As long as data centers continue to be built and operated by Oracle, “I earn this service revenue and this money”; the same logic runs down the chain to Vertiv, liquid cooling, transformers, and cables. 刘一鸣’s sense was that if Stargate truly advances to Phase 5, none of the existing technologies seems capable of delivering it; 庄明浩 did not further confirm this engineering judgment in the conversation.
Even if GPT-5 launched as expected at the time, it was more likely to reinforce the existing paradigm than recreate a once-in-a-generation shock. Foundation models, reinforcement-learning reasoning, fusion, and MoE are already highly visible paths; 庄明浩 expected it to be “very strong,” rank highly, and score well, but perhaps not particularly surprising. If Scaling Law progress has indeed slowed, training compute would face more direct pressure. By contrast, he is more focused on whether DeepSeek V4 can raise the quality of the foundation-model base and further unlock the R-series reasoning models.
OpenAI’s monetization efforts are multiplying, but still lag far behind its infrastructure-spending curve. ChatGPT’s daily message count is said to have risen from roughly 1B at the start of the year to 2.5B, and its full-year revenue target was reportedly raised from $10B to $13B; but the Oracle contract alone is worth $30B a year, and total costs go far beyond that. “When will those two curves change” remains the $600B question Sequoia posed last year—and today that figure may need to double.

Deep dive

1. The Bearish Compute Trade Has Reversed, but Extreme Consensus Still Sets Up the Next Swing

庄明浩 revisited the cycle through price action: shortly after Stargate was announced, DeepSeek triggered a 17% one-day plunge in NVDA, and Microsoft then cut two data centers. The view that data-center CapEx was excessive became almost uniformly bearish.
Six months later, NVIDIA’s market cap had broken $4T, while the Nasdaq and S&P 500 hit new highs again; Agents, reinforcement learning, multimodality, and AI coding pushed compute demand higher. “The thing we were worried about seems not to have happened—and it is still moving forward at full speed.”
His caveat was equally clear: the first half of 2025 proved only that there was “no need to worry for now,” not that the risk had disappeared permanently. Once everyone forms an extreme consensus again, “a new threat may emerge,” and markets will continue swinging between enthusiasm and pessimism.

2. Stargate Has Shifted from a Standalone Company into a Catch-All Label for a Convoluted Financing Structure

刘一鸣’s key question was whether OpenAI’s newly signed data-center projects fell outside the scope finalized in January, implying that it was moving away from SoftBank. 庄明浩 noted that although OpenAI classifies the Oracle project under Stargate, the two “do not appear to have much to do with each other”; SoftBank was also not part of the original plan.
The original structure was supposed to be an independent company: outside investors and shareholders would provide the funding, Oracle would handle the infrastructure, and a formal operator would participate. Six months of stalled execution suggests that too many stakeholders are involved and the boundaries between entities have become blurred.
More troubling is the nested funding chain: OpenAI and Microsoft are still negotiating their respective interests and financing arrangements; SoftBank is both the largest investor in OpenAI’s new tens-of-billions financing round and the party that committed $100B to Stargate’s first phase, with both sums requiring SoftBank to keep raising capital.
庄明浩 described it as “a knot that cannot be untangled”: equity, investment, the new entity, and the buyer-seller relationship in the Oracle contract are all tied together. The concern is no longer whether DeepSeek proved that less compute can still succeed, but “who is paying for this.”

3. SoftBank Is All-In, but the Capital Battlefield Is More Crowded than in the Vision Fund Era

Asked whether a project failure would deal SoftBank a fatal blow, 庄明浩 said only that it would be “quite troublesome.” SoftBank is simultaneously an investment institution and, in some sense, a quasi-sovereign fund for Japan; it owns ARM and is involved in bringing OpenAI to Japan. “Every chip that could be pushed onto the table has been pushed into this one bet.”
Son Masayoshi’s playbook remains “make one super-large bet.” Before its own funding was fully in place, SoftBank raised the flag around the new plan discussed with Trump and Stargate, using the latest tens-of-billions financing round to put the future of SoftBank Group—not merely the Vision Fund—behind AI.
The difference is that the table has expanded. In the past, raising tens of millions of dollars at a $1B valuation was considered enormous; today, companies valued at tens of billions and raising hundreds of millions in a single round appear almost daily. “The numbers may need one or two zeros added,” with every top late-stage fund now participating in what was once the “SoftBank round” battlefield.

4. Oracle and the Infrastructure Chain Have More Direct Paths to Revenue

庄明浩 stood by his January view: Oracle remains the protagonist because it is not on the front line of the model war. Regardless of how OpenAI redefines Stargate, as long as Oracle builds and operates the data centers, “I get to earn that money.”
The plan committed $100B in its first year, 2025. 庄明浩 said that in January, only $10B–$20B appeared genuinely executable; even including Oracle’s roughly $25B expected CapEx for the year and OpenAI’s $30B annual contract, the scale still fell short of the first-year commitment. He did not confirm which phase the project had reached, instead using funding scale to answer 刘一鸣’s question about the delay to Phase 3.
Capital constraints also explain Meta’s talent war: hiring nearly everyone it can poach costs at most a few billion dollars, while data centers are measured in “hundreds of billions a year.” OpenAI is not a public company; “all of its assets would not be enough to sit at this table.” The $7T plan Sam Altman once discussed may sound fantastical, but it reveals the true scale involved.
刘一鸣 cited Vertiv and asked who would benefit on the engineering side, expanding the discussion to liquid cooling, transformers, cables, and power supply. 刘一鸣’s sense was that if Stargate truly reaches Phase 5, existing technologies may be unable to deliver it; that was his engineering judgment, not a conclusion 庄明浩 confirmed in the conversation.

5. Model Competition Is Splitting into Three Fronts: Capability, Capital, and Sovereign Positioning

xAI does not have an easy hand either: xAI combined with Twitter/X is valued at only a few tens of billions of dollars, yet burns roughly $1B a month and must keep buying chips, building data centers, and raising capital. The line reposted by Elon Musk was, “Gold used to be the source of confidence; today, data centers are the source of information.”
Among the four U.S. leaders identified by 庄明浩—OpenAI, Anthropic, xAI, and Google—the biggest problem for all 3 startups may be money. Model capability still matters, but capital scale has become “the most nakedly obvious standard.”
Mistral’s value does not come entirely from benchmarks. 刘一鸣 believes that although it adopted MoE relatively early, its products and technology still appear weaker than DeepSeek’s; 庄明浩 also acknowledged that it has not broken out in any particular area, while stressing that “it is there, and its position matters.”
This is the logic of “sovereign AI”: the interests of the U.S. players are already too complicated for companies such as Apple, while China is off-limits, leaving Europe as the only place to look for a partner. Earlier reports of Apple’s proximity to Mistral, along with 刘一鸣’s mention of Lenovo Capital’s investment in Mistral, suggest that Mistral’s position could produce some form of cooperation or outcome. 庄明浩’s view was that “there will always be a way,” not that Mistral had a confirmed technical edge.

6. The Surprise Threshold for GPT-5 Has Fallen; the Real Question Is When Revenue Covers Compute

庄明浩 speculated that GPT-5’s delay could be related to a change in product definition: the scope expanded from combining foundation and reasoning models to product forms such as Deep Research and an Agent platform. Industry expectations are no longer as “crazy and unrealistic” as they were a year ago, reducing the significance of any single model launch.
His base case is that GPT-5 will follow the paths of foundation models, reinforcement-learning reasoning, fusion, and MoE to achieve a very high ranking and strong scores, but will “find it very, very difficult to be particularly surprising.” In China, the more important watchpoint is DeepSeek V4: V4 is the foundation model, while the R-series is the reasoning model, whose capability ceiling rests on the former. Kimi K2 and new Qwen versions provide contemporaneous reference points.
刘一鸣 also asked about a female CEO reportedly due to join OpenAI in mid-August to oversee product and commercial operations. 庄明浩 did not confirm the details, saying only that the matter would be important for some time to come. He also noted that Alphabet has rallied sharply this year, with its estimated forward P/E rising from roughly 20–30 at the start of the year to 40, though it is difficult to tell whether that reflects monetization efforts or simply being carried higher by the wave.
Monetization efforts are increasing: OpenAI is planning for $10B in full-year revenue, while internally it is reportedly aiming for $13B; ChatGPT’s daily message count has risen from roughly 1B to 2.5B. But against the $30B-a-year Oracle contract, revenue remains small relative to costs. “When will those two curves change” remains the $600B question Sequoia raised last year, and 庄明浩 said the figure may need to double today.

a16z GP, Martin Casado: Anthropic vs OpenAI & Why Open Source is a National Security Risk with China 2025 Silicon Valley Tech Mid-Year Review with Fusion Fund’s 张璐