meath

Throw out a number first: OpenAI is losing $12.3 billion, yet the CFO internally commits to taking it public in 2027. This isn’t a financial signal—it’s a preview of a fundraising strategy. High-growth AI companies will choose to expand using equity rather than debt; the premium on day one of listing is low-cost ammunition. But compared to that, the valuation logic of AI-native tokens like WLD will be forced to shift from the “narrative” to the “unit-economics model benchmarked against OpenAI.” Previously, AI tokens could support their prices with vague roadmaps and private-market valuations. However, after listing, all AI assets’ valuations will be tested against each other in the same public market. Specifically, look at WLD’s situation: Worldcoin’s biometric data and distributed-AI narrative still rely on building independent value outside the technical paths already taken by OpenAI. If, when OpenAI goes public in 2027, WLD’s on-chain activity and revenue model haven’t caught up, its capital premium will face a double squeeze—both to resist OpenAI’s gravitational pull and to prove its own independent cash flows.
Signal: OpenAI’s IPO is a turning point for AI valuation, and WLD’s window is only one and a half years left.

Swarms just rolled out 200+ updates this time, and everyone in the community is watching the competition prize and the Screener. But I actually think the publicly available API with 26 endpoints is the real headline. The competition ranks agents by on-chain trading volume, with rewards going to the top three. This isn’t simple gamification—it directly equates "agent quality" with "real gold-and-silver trading activity," filtering out gimmick projects. The Tokenized Agent Screener adds a data layer to this chaotic market—sorted by volume, trends, and newly launched listings. But opening the API means these data can be called directly by external market makers or DeFi strategies, turning "AI agent tokens" from a storytelling phase into a measurable liquidity game. Combined with @gmgnai’s buy-in integration, both frontend distribution and backend data are locked in place by Swarms. Another detail is the MCP Servers directory: it provides a standardized entry point for the tooling layer, so developers can discover and call agent capabilities without leaving the ecosystem. Once standards for the tooling layer are established, the cost of binding to the ecosystem afterward will be very high.
Signal: What Swarms wants isn’t yet another agent platform—it’s the "data infrastructure" for tokenized AI. $SWARMS

Markets are all watching whether a “Zcash ETF can get approved,” but what’s really worth breaking down is the details of Grayscale’s revision to the S-1. In essence, it’s a dry run for how “privacy coins can satisfy the SEC’s AML/KYC framework.” The filing makes specific amendments to third-party arrangements for custody, liquidity, and on-chain monitoring. This isn’t a simple compliance patch—it’s transforming Zcash from a “freely tradable privacy asset” into a “regulated securitized product.” If this path holds, it effectively draws a replicable route for all Layer 1 privacy projects (e.g., Monero). But note: under the new framework, privacy protection will inevitably involve trade-offs. On-chain auditing and regulatory interfaces are unavoidable hard costs.
The key here isn’t the probability of ETF approval—it’s that, for the first time, U.S. regulators have defined the boundary of “privacy” within a compliance framework. Once that boundary is clear, the compliance cost for subsequent privacy coins becomes calculable, which in turn determines their long-term valuation models.
Signal: In the short term, the approval narrative; in the medium term, how much “compliance premium” privacy coins can capture; in the long term, whether this kind of packaging will cause Zcash to lose its most core privacy moat. $ZEC

The first batch of production-grade Vera Rubin has been deployed in Microsoft Azure data centers, with the initial rollout going to Microsoft. But behind this move is a bigger chess game by Nvidia: turning the concept of synthetic hyperscalers into reality. Last week, Brad Gerstner described Nvidia as a “synthetic hyperscaler”—a cloud giant that doesn’t own data centers yet controls computing capacity supply. Nvidia has followed a clear path: lending money to OpenAI for its Ohio AI compute center, co-building an AI data center with Cloverleaf, and today directly supplying cards to Microsoft. Nvidia’s quarterly revenue hit $81.6 billion, up 85% year over year, and its data center networking business is growing faster than the GPU business itself—suggesting that when customers expand capacity, they inevitably have to pay a “toll” to InfiniBand. On Microsoft’s side, Azure just broke $100 billion in annualized revenue, and Copilot paid seats reached 30 million. But its own order with TSMC for 300,000 chips of the Maia 300 instead became a shackle that effectively handed production capacity over to Nvidia. Microsoft wants to develop its own chips, but Vera Rubin’s first launch still went to it, indicating that Azure’s customer demand simply can’t wait for Maia mass production. Google’s TPU 8 still hasn’t secured clear external customers, while Nvidia has already locked in the deployment schedule of top cloud players with production-grade delivery. This timing gap will directly translate into next year’s market share. Next, watch whether Nvidia’s pricing for renting out its Ohio compute power to external customers is lower than Azure’s price for the same specification—if it is, the synthetic hyperscaler model will be confirmed. Then keep an eye on Microsoft’s next earnings report to see whether Azure AI gross margins and Maia orders get delayed—because that would be the most direct evidence of a shift in supply-chain bargaining power.

The Trump administration’s “maximum-scale coordinated economic isolation” of Iran hinges on the keywords “coordination” and “secondary sanctions.” “Coordination” means allies take synchronized action; “secondary sanctions” means long-arm jurisdiction is directly aimed at foreign companies. China buying about 90% of Iran’s oil exports is the central test scenario for this operation.

Industrial chain breakdown: 1. Crude oil chain: Iran’s oil accounts for about 10–12% of China’s crude imports. In the short term, any shortfall can be offset with incremental supply from Saudi Arabia and Russia, but extending the transport radius will first move VLCC freight rates. 2. Settlement chain: If Chinese banks are hit by secondary sanctions, dedicated channels such as Kunlun Bank would be overloaded, and the daily average throughput of China’s cross-border payments system (CIPS) would become a key observation metric. 3. Energy security

Logic: The release schedule of China’s strategic petroleum reserves will change—this is a hidden variable when looking at futures spreads.

Financial market transmission path: First, near-month crude contracts rise; second, gold breaks above the prior high; third, as BTC serves as a settlement tool not constrained by sanctions, the liquidity premium increases.

This is not risk decreasing—it’s a transfer of control: a crack in the transition from the USD settlement system to a period where multiple-track systems run in parallel. Need to confirm: the specific sanctions list and the effective dates have not yet been released, and the wording “maximum-scale” depends on the text of the actual executive order.

Rumors about Anthropic’s IPO: the focus isn’t the listing itself, but the fund-raising size directly matching SpaceX’s $8.62 billion record. A few details: the filing has been submitted, and the earliest it could be made public is before the end of this month. If this amount is confirmed, Anthropic will become the largest public-market financing event in the AI sector. Industry chain
Signals: 1. Valuation anchors shift. Previously, AI unicorns were judged by revenue-growth multiples; now it’s about "who can raise the most money." This isn’t a healthy signal, but it’s the pricing logic in today’s market. 2. Liquidity allocation in the secondary market. If Anthropic takes down financing in the $80 billion range, other tech IPOs at the same time will be drained. Changes in where capital flows will transmit throughout the tech sector’s valuations. 3. Competitor response. If OpenAI and xAI are forced to follow with IPOs, it will accelerate the pace of securitizing AI assets.
Information that needs confirmation: the offering size is still market rumor and has not been confirmed; SpaceX’s $8.62 billion is just a reference point for the financing record and doesn’t mean Anthropic is necessarily going to reach that level. Betting-style
Conclusion: If the filings are publicly confirmed at the end of this month as a scale of more than $8 billion, then pricing power in the AI sector will shift completely from the primary market to the secondary market.

NVIDIA’s licensing of a $6B model—treating this as just a normal investment is what the market is doing; I see it as an inflection point for the industry chain. A few details: Poolside is a startup building code generation models, valued at $12B. NVIDIA first pays $6B to obtain the license, then invests another $1B to take an equity stake. The structure of this deal shows that what NVIDIA wants is usage rights, not financial returns. Direct implications for the industry chain: 1. Model weights now have a clearly defined market price, and they are measured in billions of dollars. Previously, H100 was priced as hardware; now it’s the software layer that’s being priced. 2. Code generation is the area closest to the commercial closed loop of AGI. By choosing this track, NVIDIA is effectively acknowledging that “data + models” is closer to customer budgets than “chips + networking.” 3. Data compliance is no longer a cost center—it’s a strategic asset. The license chain for Poolside’s training data will become a template for future due diligence. Next validation points: If the “model licensing revenue” or a breakdown line item for “IP costs” appears as a specific account in the Q3 earnings report, that would indicate NVIDIA is reshaping its profit structure. $NVDA

Cui Taiyuan personally went to Japan’s Miyagi Prefecture to check the plant site. SK hynix plans to build a memory fab in Japan, with an investment amount in the tens of trillions of won. This is neither abandoning the U.S. for Japan nor replacing domestic investment—RenYi Y2 and Cheongju M17’s domestic fab plans totaling 54 trillion won will proceed as scheduled. The Miyagi plant is incremental capacity: an overseas foothold and a one-stone-three-birds strategy.

The first layer of logic is capacity. During the global memory shortage cycle, HBM capacity is bottlenecked at TSV packaging and yield. Japan’s accumulated strengths in materials and precision equipment can directly fill the gaps. Miyagi is one of Japan’s three major semiconductor industry clusters officially designated by the government, and subsidies and supporting measures are already in place.

The second layer of logic is geopolitical hedging. The U.S. is putting pressure on Korean companies to build fabs in America. The Seoul Economic Daily pointed out that building a wafer fab costs 150 to 200 trillion won, and it’s simply not feasible without government support. By placing its bet first in Japan, SK hynix gains a negotiating chip with the U.S.: if you want me to go to the U.S., you’d have to offer better terms.

The third layer of logic is political maneuvering. In South Korea, there is sensitive sentiment at home about investing in semiconductors in Japan, but SK hynix is not the first company, and it won’t be the last. If the Miyagi plant is completed, it will become the third foreign semiconductor company to set up a facility in Japan after Micron and TSMC. Samsung currently only has a packaging and R&D center in Yokohama.

Who is harmed by this move? Samsung. Japan’s government resources are limited. By supporting SK hynix, future Japanese investment by Samsung will face even stricter scrutiny. Who benefits? Japan’s semiconductor equipment and materials suppliers—SK hynix’s plant will mean multi-year procurement contracts. Next, watch for three signals: SK hynix’s official confirmation of the investment amount and the start of construction; reactions from the U.S. Department of Commerce or the Congress—whether they apply pressure or offer incentives; and whether HBM supply data in the second half of 2026 will accelerate ramp-up due to synergy with the Japanese plant.

Broadcom is currently in talks with a syndicate for a debt financing deal of over $60 billion, with a possible total size of up to $100 billion. The money will flow to AI companies such as Anthropic through an SPV structure—essentially turning chip companies into a compute-bank. Broadcom only guarantees the senior secured portion ($60–70 billion); the remaining $30 billion is unsecured subordinated debt. The risk has not disappeared—it has been transferred. Subordinated debt holders will be the first to bear the model-company default risk. Pure-play leasing intermediaries like CoreWeave are likely to be hit hardest, because they have neither chip design capabilities nor capital-operations capabilities.
Watch the signals: the pricing of subordinated debt, Anthropic’s cash flow, and whether Nvidia GPU pricing shows any easing. This is not a debate over whether an AI bubble exists; it’s a question of how long this round of leverage can hold up.

On the same day that Trump announced the harshest economic sanctions against Iran in history, the LME copper registered warrants recorded their largest single-day increase since 2024 (+35,650 tonnes). Read together, these two pieces of information carry far more industrial-chain signal value than if you were to break them down separately. The usual explanation for the sudden surge in copper inventories is that traders are making deliveries into the system, but in light of the sanctions timing, a more plausible guess is: some international bulk commodity traders hold copper concentrate or cathode positions linked to Iran and, ahead of the sanctions taking effect, accelerate the transfer of title into the exchange clearing system to avoid the risk of secondary sanctions. Such actions would show up in the data as a sharp jump in registered warrants on the day. The impact on copper prices needs to be viewed in phases. In the short term, an increase in inventory levels will weigh on spot premiums and put pressure on the spread between nearby and forward months; in the medium term, if sanctions lead to a substantial contraction in Iran-related copper exports (including transshipment via third countries), the buildup of LME inventories may instead become the final source of selling pressure. Another indicator worth tracking in parallel is the production guidance for Chile’s Caserones copper mine—due to the second winter snowstorm, its annual output target was cut from 130,000–140,000 tonnes to 120,000–130,000 tonnes. When a contraction in South American supply coincides with the accumulation of LME inventories, it suggests the copper supply-demand balance sheet may be experiencing a mismatch: fewer physical barrels are available, but the circulating layer is becoming more clogged. The Federal Reserve meeting minutes show that some participants supported rate hikes; combined with the rebound in inflation expectations triggered by the sanctions, copper’s financial attributes may react ahead of its commodity fundamentals. Keep an eye on subsequent changes in LME cancelled warrants—if the cancellation ratio rises quickly, it indicates that the inventory buildup is only a hedging move by traders rather than evidence of genuine weakness in demand.

The shipping lane quietly established by the U.S. military in the Strait of Hormuz, in essence, pushes geopolitical rivalry down to the logistics execution layer. The value of this move lies in the fact that it allows the U.S. to selectively permit or intercept oil tankers passing through the strait without imposing a full blockade. Iranian tankers are effectively excluded, but exports from other Gulf oil-producing countries are hardly affected. This selective enforcement brings two immediate consequences: First, Iran’s oil revenues are precisely reduced, but the total global supply of crude oil does not experience a sudden cliff-like drop. The pricing logic of the Brent crude price’s geopolitical premium shifts from the previous panic-driven “comprehensive blockade” expectations to a “structural sanctions premium.” As a result, the spread structure between WTI and Brent will undergo subtle changes. Second, Iran is forced into an asymmetric response window. If Iran concludes that the U.S. can sustain this selective blockade over the long term while other producers continue supplying, Iran may escalate threats against the Strait of Hormuz itself to raise everyone’s transit costs. This turns oil-price risk from a one-way sanctions logic into a two-way game. For traders, directly tracking subsequent U.S.-Iran moves is less important than watching the daily changes in tanker insurance premiums and the frequency of exercises by Iran’s Islamic Revolutionary Guard Corps near the strait. In addition, today the LME copper registered warrants rose by the largest single-day increase since 2024; while copper prices face near-term pressure, it also suggests that physical cargo may be moving into exchange warehouses—possibly meaning that some traders are hedging logistics risks from potential geopolitical escalation in advance. BTC’s performance in this round of geopolitical events looks more like a shadow of dollar liquidity than a safe-haven tool. If crude oil breaks through key moving averages due to escalation in the standoff, risk appetite in the crypto market would truly come under pressure.

Trump announced an “Economic D-Day” against Iran, saying the wording was “the most crushing economic operation ever.” But what is truly worth watching for the industry supply chain isn’t oil prices—it’s the six targets explicitly named in the statement: oil smuggling, swap lines, cash transfers, exchange houses, ship registries, and front companies. This is a sanctions checklist, meaning the United States has already dismantled Iran’s grey financial channels item by item. The next step is to target all third parties that provide support to this system, including but not limited to banks, shipowners, and foreign exchange bureaus.

For the crypto market, there’s a connection that has been overlooked: when Iran’s USD channels are completely shut down, crypto assets could, in theory, become an alternative for trade settlement. But that’s not good news. If the U.S. Treasury believes crypto exchanges are becoming Iran’s clearing channels, the scope of secondary sanctions will extend to offshore platforms with lax KYC. Bessent hinted last week that more would be added, and Trump is simply putting the framework into practice this time. The only signal to watch next is this: in OFAC’s next sanctions list, whether a crypto exchange or stablecoin issuer appears. If it does, it means the second battlefield of D-Day is about to begin.

Trump announced launching an “economic war” against Iran. The keyword isn’t war, it’s economics. Iran’s weak point has never been missiles—it’s the rial. The gap between the official exchange rate and the black-market rate has already exceeded 50%. Usable foreign-exchange reserves may be short of $30 billion, while the annual import bill is about $60 billion. In Bessent’s list, there are oil smuggling routes, exchange houses, ship-flag registrations, and even airports—meaning the channel for people to carry cash across borders would also be cut off.

The hardest hit is Dubai: the gold market, parallel importers, and exchange houses all rely on Iranian customers’ USD business. If sanctions force the UAE to tighten account verification, Iran would have to shift to Iraq or Turkey, losing several thousand basis points per $1 million transaction.

Potential beneficiaries may include BTC. Once SWIFT and exchange services are locked out, BTC is one of the few tools that can transfer value across borders without intermediaries. USDT issuers may freeze Iranian addresses, but BTC has no issuer that can be subpoenaed. Note the liquidity issue: Iranian merchants want large, immediate settlement, and on-chain depth still doesn’t match.

The shipping side of the CL crude oil futures is the focus. If the U.S. starts scrutinizing insurance and ship flags for the shadow fleet, insurance rates would soar, directly raising the delivered-at-port costs for Asian buyers. China is Iran’s largest buyer of crude oil. If the banking system is singled out, the shockwave could spread to the RMB exchange rate.

Next signal: the frequency of the Iranian central bank’s gold-reserve disclosures. If it starts reporting increased holdings more frequently, it indicates the rial is switching anchors—from petrodollars to physical assets. That would be the turning point of the second phase.

Coinbase’s statement appears to welcome the SEC’s progress on token taxonomy, but it is effectively urging two actions: passing the Clarity Act and issuing an Innovation Exemption. The former addresses the classification of older tokens (commodity vs. security), while the latter determines the trading venues for future tokenized assets. Impact on trading structure: if token taxonomy is formally implemented, the tradability of Layer 1 tokens like HYPE on U.S. compliant exchanges would increase significantly. This would allow U.S. market makers and custodians to participate directly—opening institutional capital channels that had previously been forced to shut down due to uncertainty under the Howey test. Innovation Exemption is the real marginal variable. If it allows broker-dealers to operate ATS markets for tokenized securities within a specific sandbox, RWA tokens like ONDO would receive exchange-level liquidity support—not just OTC quotes and on-chain liquidity pools. As of now, the exemption details have not been released, and the timeline remains in a “proposed” status. The true validation point: watch whether, before 8/31, SRO members or major market makers submit ATS licensing applications based on the Innovation Exemption. If they do, the first U.S.-compliant tokenized stock on a secondary market will emerge before Q1 2027—that will be the breakthrough. Token taxonomy is merely a setup.

Cerebras 发布新推理芯片，媒体焦点都放在服务器系统性能上，但产业链的真实信号在部署形态。如果 Cerebras 这次主推交付物理服务器而非云端 API，那么它的毛利率会被硬件成本拖累，无法形成 Groq 那样的软件订阅收入结构。 更关键的是推理市场分层
逻辑：训练集群的成本极度敏感，由 NVIDIA 的主导地位压制；而推理市场正在分为两个子赛道——高吞吐批处理（由 HBM 容量决定）和低延迟单请求（由芯片间互联带宽决定）。Cerebras 和 Groq 赌的都是后者。
用实际场景量化：AI 编程助手在自动补全时用户能感知的延迟上限约 1 秒，Agent 调度系统的服务质量取决于每次工具调用的响应时间。这两类场景正处在应用爆发前夜，低延迟推理的算力缺口开始影响用户体验和留存，推动一波新的采购周期。
我预计 2026 Q4 能看到 Groq 或 Cerebras 公布单一客户超过 $50M 的推理采购量，届时这会是算力租赁市场的定价锚。

Hyperliquid's open interest has broken through $12B—the first time since 10/10. But when you break down the structure, the signal with real value isn’t the total amount, but the source of the incremental increase. In the 10/10 move, BTC simultaneously broke above 80K, and market sentiment overall was euphoric. Today, BTC is stable in the 65K–70K range, ETH/SOL are both consolidating sideways, and only HYPE perpetual contracts are continuing to see heavy volume. This kind of independence suggests that the capital isn’t coming from rotation among mainstream coins—it's high-risk appetite capital from the Solana ecosystem, directly entering to short or hedge the new asset.

The more important transmission to watch: Hyperliquid's open interest is now effectively a leading indicator for new-asset issuance. After new projects list perpetual contracts on HL, market makers must establish hedged positions at the same time—this directly drives up open interest. If, in the next two weeks—specifically the window from 8/21 to 8/25—top-tier projects announce issuance of perps on HL, then this new high in open interest is only the beginning. Another angle is the liquidation map: if HYPE’s price suddenly spikes upward and triggers short squeezes and liquidations, it could replicate another 10/10-style short-term “needle” at the ~20% level.

Circle officially announced that EURC’s circulating supply has surpassed 400 million euros, with a year-over-year growth rate of over 100%. At the August 2026 point in time, this data has clear industrial significance: euro stablecoins have moved from being experimental fringe products to becoming institutional settlement tools. Structurally, EURC’s growth source has already shifted from Crypto trading pairs to Payments and Institutional Workflows. This means European corporate finance teams are starting to use EURC for cross-border settlement and liquidity management—traditional euro-denominated assets can now be traded and settled on-chain 24/7. Implications for the market: first, the market for non-USD stablecoins is not fake demand—compliance plus local fiat-currency pegging is the real track; second, as the issuer, Circle’s expansion of EURC will also feed back into the USDC ecosystem, because the issuance and redemption rails for both are shared; third, in the short term, meme coins like DOGE and EURC have no direct connection, but as euro stablecoin liquidity deepens, friction costs for European retail users entering Crypto will decrease—creating an indirect incremental-funding narrative. 400 million euros isn’t huge, but there’s no need to doubt the direction of the trend.

TechCrunch reports that Anthropic’s annualized revenue has reached $6.5 billion. Looking at this figure against the August 2026 milestone, it suggests that AI application layers have moved from the cash-burning phase into the revenue validation phase. The penetration of the Claude series in the enterprise market is improving noticeably, especially in code generation and data analysis scenarios. For the crypto market, the most direct mapping is the AI Agent track—not the kind of MEME coins, but Agent protocols with real compute costs and user-paid demand. Anthropic’s $6.5 billion revenue proves that enterprises are willing to pay for AI labor; if an on-chain Agent can provide comparable services, it doesn’t need to be perfect—having a pricing advantage alone can give it room to survive. At the same time, pay attention to compute costs: Anthropic’s scale economies will lower the per-inference unit cost. This is a double-edged sword for subnetwork competition models like Bittensor (TAO)—lower costs help drive demand growth, but a price war over centralized APIs will squeeze the gross margin of decentralized networks. In the short term, the narrative in the AI sector will shift from “concept” to “revenue proof,” and protocols with real usage are likely to outperform projects driven purely by storytelling.

NVIDIA teams up with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish an AI computing power infrastructure financing platform, aiming to unlock more than $500 billion in third-party capital. This is the financialization of GPU clusters: like a power plant, computing power can generate steady cash flow, enabling them to issue bonds and engage in asset securitization. For the crypto market, the key is that the valuation anchor for the DePIN track has changed. Previously, Render and Akash priced services based on token demand; now Wall Street sets fixed-income benchmarks for computing power. Institutional capital can then compare them side by side. If the ROI of AI computing power in traditional finance is 6–8%, DePIN projects must prove they can deliver similar revenue streams and governance capabilities—otherwise, funds will flow into Wall Street products. Another perspective: if this $500 billion materializes, AI computing supply will grow exponentially and training costs will drop, which would also reduce the development cost of Crypto AI Agents. That’s a mid-term positive for the application layer, but in the short term, the financialization of centralized computing power puts pressure on the differentiation value proposition of decentralized computing networks. More details will likely emerge at industry conferences in late August, so it’s worth tracking.

Nvidia has provided a $105 billion lease guarantee for OpenAI’s data center in Ohio, and invested $1.5 billion for an equity stake in SB Energy. This deal makes Nvidia the sole chip supplier for the park, securing 4.25GW of capacity (expandable to 8GW), with the first 800MW expected to come online in 2028. OpenAI receives a 20-year lease, but pays only for the capacity that is actually delivered. This is not a typical procurement transaction. By guaranteeing future chip revenues to underwrite today’s capital expenditures, Nvidia effectively folds the customer’s credit risk directly into its own balance sheet. The previously discussed $250 billion guarantee has shrunk to $105 billion, suggesting Nvidia has also exercised restraint in its off-balance-sheet exposure. Beneficiaries: OpenAI locks in long-term compute power at extremely low cost of capital; Nvidia secures exclusive supply status and moves upstream into energy; SB Energy receives both equity injection and long-term lease commitments. Damaged parties: Nvidia’s contingent liabilities expand, and market doubts about its revolving financing will intensify; smaller and mid-sized AI labs face even greater pressure in securing compute. The transmission path to the crypto market runs through the energy side. 8GW is roughly the electricity consumption of 6 million U.S. households. AI data centers, by grabbing grid capacity at the GW level, will create structural squeeze on electricity procurement by Bitcoin miners over the next three to five years. This is not a short-term sentiment issue, but long-term competition at the infrastructure level. Next things to watch: Nvidia’s disclosure framework for contingent liabilities in its quarterly report; SB Energy’s financing structure; whether OpenAI uses this lease as underlying collateral for another round of financing.