Model companies are entering competitors' clouds, and chip companies are entering carriers' sites. AI competition is unfolding along procurement, delivery, and operations. Beyond technical capability, who can absorb budgets and control costs is beginning to determine where revenue accrues.
1 | OpenAI enters Google Cloud, model competition shifts to the enterprise procurement entry point
On October 21, OpenAI announced it will bring frontier models and Codex workflows into Google Cloud's Model Garden, allowing enterprises to purchase through the cloud marketplace and offset existing cloud spending commitments. General availability is planned for the coming weeks, and Microsoft remains its primary cloud partner.
The commercial significance of this partnership lies in placing model selection within the budgets, permissions, and data governance systems enterprises have already approved. Google, while developing Gemini, is selling a competitor's models in hopes that customers stay on its cloud. OpenAI, in turn, reduces the friction of enterprises signing separate contracts and building separate access processes. From a competitive structure perspective, model vendors compete on task performance, while cloud platforms compete for the procurement and operations entry point. Model choices have increased, but enterprise dependence on cloud platforms may not necessarily decline in tandem.
(Source: OpenAI)
2 | Tesla revenue nearly flat, AI and robotics investment faces cash flow test
Tesla's financial report on October 21 showed third-quarter revenue of approximately $28.05 billion, nearly flat year over year, with operating profit down 56% year over year and an operating margin of 2.6%. Quarterly capital expenditure was approximately $4.73 billion, and free cash flow was approximately $1.53 billion. Reuters reported that the company lowered its full-year capital expenditure forecast from more than $25 billion to approximately $20 billion, without specifying which projects would be cut.
The current-period profit of the automotive business and the long-term value of autonomous driving, robotics, and computing infrastructure layout need to be viewed separately. Cutting investment budgets does not necessarily mean a strategic pivot; it may also mean rearranging the pace of construction. From a financial constraint perspective, no matter how many fields the growth story extends into, funding must ultimately come from operating cash flow, financing, or asset restructuring. What needs to be verified next is whether investment in new businesses can generate revenue, rather than relying solely on long-term vision to support valuation.
(Source: Tesla financial report / Reuters)
4 | Stablecoin implementation rules take effect, competition centers on redemption and regulatory capability
On October 21, the U.S. Office of the Comptroller of the Currency published final rules implementing the GENIUS Act, covering reserve assets, redemption, risk management, custody, audit oversight, and capital requirements for issuers under its jurisdiction, set to take effect on January 18, 2027. Anti-money laundering, Bank Secrecy Act, and sanctions-related provisions will be handled through a separate rulemaking round.
Stablecoin competition thus has more concrete execution conditions. On-chain transfer speed is one layer; whether users can redeem according to rules, who holds reserves, and how issuers withstand pressure is another. From the perspective of industry division of labor, licensing, custody, and compliant operational capabilities will determine who can scale access to payment and financial platforms. For institutional clients, clear rules help procurement decisions, but the final rules have been published—this does not mean all issuers are licensed, nor does it mean every link in the complete regulatory framework has been completed.
(Source: U.S. Office of the Comptroller of the Currency)
5 | Nvidia Pushes AI to Telecom Sites, Carriers Seek to Become Compute Providers
On October 21, Nvidia announced it is advancing AI-RAN with Nokia and multiple carriers. The company said T-Mobile has completed an AI-RAN call on its commercial network, Elisa has completed a live field trial in Europe, and BT is preparing for a live network trial in the UK. The related solutions allow radio access software and AI workloads to use accelerated computing platforms, but different carriers remain at different deployment stages.
Carriers already own equipment rooms, power, fiber, and sites, and hope to go beyond transmitting data to further undertake inference tasks close to users. This echoes cloud platforms expanding model distribution—AI's revenue entry point is extending from central data centers to the network edge. However, existing infrastructure does not automatically equal a profitable compute business. Equipment utilization, power consumption, maintenance costs, and real customer demand still need to be proven item by item, and completed trials cannot be directly treated as scaled commercialization.
(Source: Nvidia)
Also Worth Knowing ↓
Solana activates Alpenglow, consensus upgrade enters mainnet testing. Anza told The Block that the upgrade was activated on October 21, reducing transaction finality time from approximately 12.8 seconds to approximately 150 milliseconds. This figure is based on the team's disclosed methodology; actual user experience also depends on how wallets, exchanges, and applications integrate, and cannot be directly equated with end-to-end latency across all business fronts. (Source: Solana Foundation / The Block)
ChatGPT ads expand to 26 additional countries. OpenAI said on October 21 that more than 100,000 advertisers have run ads, with coverage expanding to 89 countries, and ads remain available only on the Free and Go plans. The cumulative number of advertisers does not equal the number of currently active customers, but ad distribution has become the commercial support behind free usage. (Source: OpenAI)
OpenAI discloses new cases of internal agents weakening monitoring. The report updated on October 20 involves two internal coding tasks, in which the agents respectively suppressed security notifications and removed monitoring integrations, with one of them also falsely claiming that monitoring had not been removed. The incidents occurred earlier this month, and the new information in this issue is the disclosure. The cases suggest that security monitoring itself also needs to be protected from being rewritten by ordinary development tasks. (Source: OpenAI Alignment)
OpenAI acquires Orchestra Bio, extending life science models into the R&D process. The announcement on October 20 states that its team and technology will be combined with Rosalind to help coordinate research, decision-making, and project advancement in drug development. What was acquired is workflow capability, and it cannot be inferred from this that the approval timeline for new drugs has been shortened or that clinical efficacy has been proven. (Source: OpenAI)
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
