The ledgers of AI companies, chipmakers, and tech giants are all undergoing a transformation: profits from unrealized gains, orders from investments, and EV cash flow now earmarked for AI consumption.
1｜Alphabet Earns $112.1 Billion in Q1, Strongest Business Shifts Toward AI Investment
Alphabet's Q2 revenue reached $119.8 billion, with a net profit of $112.1 billion, nearly tripling year-on-year. The most eye-catching aspect was not advertising or cloud services, but a $97.98 billion investment portfolio unrealized gain, primarily from revaluations of stakes in SpaceX, Anthropic, and others. Excluding this paper gain, Alphabet's normalized EPS is around $2.85, below market expectations.
Google Cloud continues to grow rapidly, with revenue up over 80% year-on-year and a backlog of orders worth $514 billion. However, the progress in the cloud business has been completely overshadowed by investment gains. The return on investment for large companies developing AI internally has not yet been proven, as investments in AI and space unicorns have already reshaped the profit statement.
The core business of tech giants has not disappeared but is increasingly resembling a cash flow generator for a venture capital ledger.
(Source: Alphabet Earnings Report / Fortune / Bloomberg / NYT)
2｜AMD Invests $5 Billion in Anthropic, Chipmaker Starts Spending to Acquire Customers
AMD has invested $5 billion in Anthropic and obtained a minority stake. In exchange, Anthropic will deploy AMD's MI450 chips in a new data center, and both parties have committed to building a 2GW-scale computing infrastructure, with the first 1GW cluster expected to be operational in the first half of 2027.
Anthropic's May annualized revenue has already reached $47 billion, with only $9 billion for the entire year of 2025, showing growth of more than four times in six months. This scale qualifies it to charge a "membership fee" to chip suppliers. While NVIDIA still dominates the AI GPU market, AMD aims to enter the top-tier training workload segment. Merely relying on performance narratives is insufficient; capital must also be integrated into orders.
Chip companies paying to secure supply qualifications indicate that the price signal of compute scarcity is, in turn, raising the bargaining power of large model companies.
(Source: WSJ / TechCrunch / Tom's Hardware / Reuters)
3｜Huang Renxun Voices Support for China's Open-Source Model, Washington Pushes Distillation into a Sanction Topic
Hong Renxun praised China's AI open-source contributions at the Shanghai World Artificial Intelligence Conference, stating that the U.S. should not ban the use of Chinese models. In the same week, White House Chief Technology Officer Michael Kratsios accused Dark Side of the Moon's Kimi K3 of stealing Anthropic Fable technology through distillation, with Treasury Secretary Bessent stating they are evaluating sanctions against relevant companies.
These two signals constitute a direct clash. Dark Side of the Moon denies the distillation accusation, while the U.S. claims to have evidence. On the surface, this is a dispute over model sources, but at its core is whether the U.S. should incorporate open-source training behavior into tech controls.
The large-model industry has historically relied on mutual learning and rapid replication. Once distillation is defined as cross-border IP theft, the open-source ecosystem will be rewritten under a national security framework.
(Source: Fortune / Tom's Hardware / TechCrunch / U.S. Treasury Department)
4｜OpenAI $750 Billion Infrastructure Plan Turns Data Centers into Power Politics
OpenAI's initial investment in Project Camellia in Georgia exceeds $30 billion, with a power demand of 3.2GW, part of its $750 billion AI infrastructure plan. Musk's SpaceXAI is simultaneously building a massive data center in Texas. BNEF's latest forecast predicts that U.S. data centers will consume 20% of the national electricity supply by 2035, reaching 194GW, an 83% increase from seven months ago.
Georgia Power has planned grid expansions for the new load. AI companies are becoming the largest group of power consumers in the U.S., with a single company's capital spending on par with mid-sized power companies.
OpenAI's CFO has expressed concerns about the expenditure rate. The $750 billion is a commitment amount, and actual execution depends on whether revenue can keep pace with capital expenditure. If commercialization lags behind infrastructure investment, data centers will shift from moats to liabilities.
(Source: Bloomberg / TechCrunch / Tom's Hardware / Georgia Power)
5｜Tesla and IBM Financial Reports Highlight: AI Budgets Squeezing Core Business
Tesla's Q2 earnings report fell below expectations, with earnings per share of $0.33, lower than analysts' expected $0.51. Free cash flow was -$1.1 billion, with capital expenditures rising to $5.8 billion, mostly directed towards Robotaxi, Optimus, Terafab, and computational power expansion.
IBM lowered its full-year revenue guidance on the same day, citing customers shifting budgets from traditional IT to AI infrastructure. While the two companies are in different positions, the signal is the same: AI spending is not an additional budgetary pool out of thin air; it will be drawn from existing business.
Tesla is betting on robots with its electric vehicle cash flow, while IBM's customers are using their traditional IT budget to buy AI infrastructure. The capital markets are still willing to give a premium to the AI narrative, but enterprise balance sheets have begun to show the cost.
(Source: Tesla Earnings Report / IBM Earnings Report / Reuters)
Also Worth Knowing ↓
Four former DOGE employees have founded the military cybersecurity company Cathedral, with a16z and Sequoia leading the investment, valuing the company at $1.4 billion. The company is still in stealth mode, aiming to enhance the U.S. military's network defense capabilities using AI and planning to acquire a dedicated data center. Leaving government efficiency departments to start a venture in national defense, this path is taking shape. (Source: Reuters / Gizmodo)
The U.S. Army exhausted its entire annual AI dialogue budget before the end of the fiscal year. The AI API tokens procured by the military were depleted in the third quarter, and the actual usage rate of military AI applications far exceeded procurement estimates. The speed at which AI is being consumed is much faster than the budget approval rate. (Source: Ars Technica)
Amazon has laid off some positions in the AGI team, folding them into the cloud services division. The independent AGI research path has given way to pragmatism, with resources focused on the cloud AI product line that can directly serve customers. (Source: Reuters)
Trump signed a nuclear energy cooperation agreement with Saudi Arabia, allowing uranium enrichment terms for the first time. Previously, the U.S. had consistently banned enrichment in civil nuclear cooperation. Saudi Arabia's insistence on retaining enrichment rights has been a major reason for the agreement's years-long stalemate. Trump's concession has crossed a long-standing red line of non-proliferation. (Source: Fortune / Reuters)
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
