Amazon delivered a stronger-than-expected performance from its cloud computing business in the second quarter, with Amazon Web Services (AWS) recording its fastest revenue growth since 2021 as surging demand for artificial intelligence infrastructure continues to reshape the global cloud market.

The results reinforce the view that hyperscale cloud providers remain among the biggest beneficiaries of the AI boom, as enterprises increase spending on AI training, inference and cloud-based applications.

AWS generated $42.23 billion in revenue during the quarter ended June, comfortably ahead of analysts’ expectations of $40.54 billion, according to StreetAccount.

Revenue rose nearly 37% year over year, a sharp acceleration from the 28% growth reported in the first quarter and the strongest expansion for AWS in almost five years.

AI Becomes A Major Growth Engine

Amazon said both AWS’ artificial intelligence business and its custom AI chips have each surpassed $25 billion in annualized revenue, more than doubling from a year earlier. The milestone illustrates how AI has evolved from an emerging opportunity into one of AWS’ largest growth drivers, as businesses increasingly deploy generative AI models and migrate AI workloads to the cloud.

The company continues to invest aggressively in proprietary silicon, including its Graviton processors and AI accelerators, to reduce reliance on third-party chips while improving performance and lowering operating costs for customers. Those investments are helping Amazon compete more effectively against rivals offering AI infrastructure built around chips from Nvidia and other suppliers.

The latest results come amid an increasingly competitive race among the world’s three largest cloud providers. AWS remains the industry’s largest cloud platform by revenue, with an annualized revenue run rate of approximately $148.4 billion.

That keeps it comfortably ahead of Microsoft, which reported on Wednesday that Azure and its broader cloud business generated more than $100 billion in revenue over the past year after cloud growth accelerated to 43%.

Meanwhile, Alphabet said last week that Google Cloud revenue surged 82% year over year to nearly $25 billion, lifting its annualized revenue to about $78 billion.

Although AWS remains the market leader, the faster growth rates reported by Microsoft and Google suggest competitors continue to gain traction as enterprises diversify cloud providers and expand AI deployments.

Following Alphabet’s earnings, Evercore analysts noted that Google’s rapid sequential revenue growth could eventually raise investor questions about cloud market share, although they maintained a positive rating on Amazon’s stock.

AWS Drives Amazon’s Profitability

Beyond its scale, AWS continues to be Amazon’s primary profit engine. The cloud division generated $16.62 billion in operating income during the quarter, significantly above analysts’ expectations of $13.62 billion.

AWS posted an operating margin of 36.8%, slightly ahead of Google Cloud’s 35.6% margin. The business accounted for nearly 61% of Amazon’s total operating profit, highlighting the extent to which the company’s earnings depend on cloud computing rather than its retail operations.

That profitability gives Amazon greater financial flexibility to fund massive investments in AI infrastructure while continuing to expand its e-commerce and logistics businesses.

Amazon significantly increased spending on infrastructure during the quarter as it races to meet growing demand for AI computing capacity. Capital expenditures rose 68% year over year to $54.21 billion, exceeding analysts’ expectations of $49.35 billion.

The spending reflects Amazon’s ongoing construction of data centers equipped with advanced AI chips, networking equipment, and power infrastructure needed to support complex AI models.

The elevated investment mirrors similar spending trends across the cloud industry. Microsoft recently reported quarterly capital expenditures of $41 billion, while Alphabet has also sharply increased investment in AI infrastructure, underscoring how hyperscalers are engaged in an unprecedented buildout of computing capacity.

Although the surge in capital spending has weighed on free cash flow across the sector, technology companies argue the investments are necessary to secure long-term leadership in artificial intelligence.

Amazon continued expanding AWS’ AI ecosystem during the quarter through new partnerships with leading AI developers. The company announced that AWS will begin hosting models from OpenAI, broadening the range of foundation models available through its cloud platform. Amazon also disclosed that Meta Platforms will deploy hundreds of thousands of AWS Graviton chips under a three-year agreement, providing another validation of Amazon’s custom silicon strategy.

These partnerships demonstrate that cloud providers are now competing not only on infrastructure capacity but also on the breadth of AI models, proprietary chips and software ecosystems they can offer enterprise customers.

AI Spending Shows No Signs of Slowing

Amazon’s results lend credence to the broader narrative emerging from recent earnings across the technology sector: enterprise demand for AI infrastructure remains exceptionally strong.

Despite investor concerns over the sustainability of AI-related capital spending, cloud providers continue to report accelerating revenue growth, expanding AI adoption and rising demand for computing capacity.

The quarter also highlights the increasingly central role of cloud computing in the AI economy. As companies build and deploy generative AI applications, demand for scalable computing power, specialized AI chips and cloud-hosted models continues to rise, benefiting hyperscalers with the financial resources to invest tens of billions of dollars in infrastructure.

OpenAI has significantly reduced the prices of two of its latest artificial intelligence models, escalating competition in the generative AI market as leading developers battle to deliver more computing power at lower cost and convince businesses that AI investments can generate measurable returns.

The latest pricing overhaul is part of a growing industry shift that has seen the competitive edge being determined not only by model intelligence but also by the economics of deploying AI at scale. As enterprises expand AI adoption, they are placing greater emphasis on reducing inference costs, improving efficiency and lowering the total cost of ownership.

Chief Executive Officer Sam Altman announced the reductions on Thursday, saying OpenAI wants to offer “the best price/intelligence tradeoff at every level.”

The company slashed the cost of GPT-5.6 Luna by 80%, reducing pricing to $0.20 per million input tokens and $1.20 per million output tokens. GPT-5.6 Terra received a 20% price cut, bringing costs down to $2 per million input tokens and $12 per million output tokens.

OpenAI also introduced a new “Fast” mode for GPT-5.6 Sol, its flagship frontier model. The option delivers up to 2.5 times faster performance through the API for twice the price while maintaining the same level of intelligence, giving developers the flexibility to prioritize speed for latency-sensitive applications.

The lower pricing extends beyond application programming interface (API) customers. OpenAI said Luna and Terra’s reduced costs will also be reflected in how usage is calculated for paid subscribers using Codex and ChatGPT Work, making advanced AI capabilities more affordable for software engineering, automation and enterprise workflows.

The company attributed the lower prices to improvements across its AI stack rather than reductions in model capability.

“Our efficiency edge comes from improving the models, the inference systems that run them, and the agentic harness that connects them to tools and context,” OpenAI said in a statement.

The company said advances in model architecture, optimized inference systems, improved routing, and better context management have enabled it to generate responses using fewer computing resources while maintaining performance. More efficient routing ensures hardware utilization remains high, while smarter context management allows AI agents to avoid repeating completed tasks, reducing unnecessary token generation and computing costs.

The announcement comes roughly three weeks after OpenAI launched the GPT-5.6 family following a temporary pause in the broader rollout at the request of the U.S. government. The models represent the company’s latest generation of AI systems designed for coding, reasoning and enterprise applications.

AI Pricing Emerges As The Industry’s Next Battleground

Over the past two years, model providers competed primarily by releasing increasingly capable AI systems. That contest is now evolving into a battle over who can deliver the highest intelligence at the lowest cost, a transition driven by rising enterprise AI adoption and surging demand for computing infrastructure.

Every AI query requires expensive graphics processing units (GPUs), electricity and networking resources. Lowering inference costs enables providers to improve margins while making their services more attractive to businesses that process billions of tokens each month.

Jacob Bourne, senior analyst at EMARKETER, said the latest move reflects growing pressure from enterprise customers seeking stronger returns on AI spending.

“The era of tokenmaxxing is over,” Bourne said.

“Enterprises have figured out how easy it is to burn tokens without getting value back, and they’re pushing back on those increasing AI bills.”

Organizations are now evaluating AI projects based on productivity gains and financial returns rather than raw model capability, forcing AI providers to optimize both performance and operating costs.

OpenAI’s pricing move is believed to have also been spurred by growing competition in the AI industry.

Chinese startup Moonshot AI recently released its open-weight Kimi K3 model, intensifying pressure on proprietary AI developers such as OpenAI by offering developers greater flexibility and lower deployment costs. Open-source and open-weight models continue to gain traction among enterprises seeking greater control over infrastructure and long-term operating expenses.

Anthropic, another leading AI developer, has been balancing subscription and usage-based pricing while managing finite computing capacity, illustrating the industry’s challenge of expanding access without overwhelming available infrastructure.

Google has likewise emphasized affordability as a competitive advantage, repeatedly highlighting the cost efficiency of its Gemini models as customers increasingly scrutinize AI spending.

Microsoft has adopted a similar strategy. During the company’s quarterly earnings call on Wednesday, Chief Executive Officer Satya Nadella repeatedly emphasized “cost efficiency” as a defining characteristic of Microsoft’s MAI-Thinking-1 model.

“We are building a new model system where the harness, context, memory, and action space are separate from any one model family, thereby moving the frontier on the cost to outcome curve,” Nadella said.

These point to a broader architectural shift across the AI industry. Rather than relying solely on larger models, developers are separating reasoning engines from memory, context management and external tools, allowing AI systems to complete more sophisticated tasks while using computing resources more efficiently.

Overall, OpenAI’s latest pricing cuts signal that the economics of artificial intelligence are becoming as important as the capabilities of the models themselves.

As AI applications move deeper into software development, customer service, research, finance and enterprise automation, businesses are demanding predictable operating costs alongside high performance. Lower token prices reduce barriers to adoption, encourage larger AI workloads and strengthen customer retention, particularly as organizations deploy AI agents that continuously process large volumes of data.

A U.S. federal judge has cast significant doubt on the Trump administration’s justification for banning the federal government from using Anthropic’s artificial intelligence technology, marking a pivotal moment in a closely watched legal battle that could reshape how Washington regulates and procures advanced AI systems.

During a hearing on Thursday, U.S. District Judge Rita Lin said the administration had failed to present sufficient evidence to support its decision to classify Anthropic as a supply chain risk and prohibit federal agencies from using the company’s AI models.

The judge’s remarks suggest the government faces an uphill battle in defending one of the most consequential procurement actions taken against a leading U.S. AI developer, particularly as artificial intelligence becomes increasingly integrated into national security, intelligence and defense operations.

At issue is whether the federal government can effectively blacklist a domestic AI company based on concerns over future operational risks and disagreements over how its technology should be deployed, rather than on evidence of actual security vulnerabilities or misconduct.

Judge Lin repeatedly questioned the government’s rationale, signaling skepticism over several of the Pentagon’s core arguments.

One of the administration’s central claims is that Anthropic could pose a national security risk because it might interfere with AI models used in military operations. Government lawyers argued that the company could potentially disable, modify, or otherwise influence its systems during wartime.

Judge Lin said she had seen no evidence supporting that assertion.

She noted there was no proof that Anthropic could alter a model after it had been delivered to the government or activate what she described as a “kill switch” that would disrupt military operations.

The government’s broader argument that Anthropic’s public criticism of the Department of Defense justified its exclusion from federal contracts also drew scrutiny.

Government attorneys contended that the company’s public opposition to certain military uses of artificial intelligence contributed to the decision to ban its technology.

Judge Lin described that reasoning as “really troubling,” warning it could create a precedent in which federal contractors face retaliation simply because they disagree with government policy.

This touches on broader constitutional concerns surrounding free speech and whether companies can continue to criticize government policy without jeopardizing their eligibility for federal contracts.

Contract Dispute Evolved Into Legal Battle

The dispute originated during negotiations between Anthropic and the Department of Defense over potential military contracts.

Anthropic told the Pentagon it did not want its frontier AI models used for mass surveillance of Americans or to support systems responsible for targeting decisions or the use of lethal force, arguing current AI technology has not matured sufficiently for such high-consequence applications.

The Defense Department rejected that position, maintaining that private technology companies should not dictate how the U.S. military employs legally acquired technology. Pentagon officials also argued that any deployment of Anthropic’s AI would comply with U.S. law, military rules of engagement and established oversight procedures.

Negotiations eventually collapsed, after which the administration designated Anthropic a supply chain risk and barred federal agencies from using its technology.

Anthropic responded by filing two lawsuits in March challenging both the procurement ban and the government’s risk designation.

Judge Lin previously issued a preliminary injunction blocking enforcement of the ban while litigation proceeds. Thursday’s hearing focused on whether that temporary protection should become permanent.

A separate lawsuit involving related issues continues in federal court in Washington, D.C.

However, the legal battle underscores a rapidly emerging conflict between frontier AI developers and governments seeking broader access to increasingly powerful artificial intelligence systems.

Unlike conventional defense contractors, companies developing foundation models often maintain detailed acceptable-use policies that restrict applications they consider unsafe or ethically problematic. Many leading AI firms, including Anthropic, OpenAI and Google, have published policies limiting the use of their models for certain military, surveillance or weapons-related purposes, although those policies have evolved as governments have become increasingly important customers.

The Pentagon, meanwhile, has accelerated efforts to integrate generative AI into military planning, intelligence analysis, logistics, cybersecurity and decision-support systems as part of a broader modernization strategy aimed at maintaining the United States’ technological advantage.

That has created friction between commercial AI developers seeking to preserve ethical safeguards and defense agencies that argue operational decisions must ultimately remain under government control.

National Security Argument Faces Judicial Scrutiny

A central issue in the case is whether concerns raised by the Defense Department amount to legitimate national security risks or remain largely speculative. Experts have noted that many government AI deployments rely on locally hosted or isolated versions of models that cannot simply be altered remotely by their developers.

Judge Lin’s questioning reflected that distinction, suggesting the government had not demonstrated that Anthropic retained the technical ability to manipulate AI systems after delivery.

The case may therefore establish an important legal threshold for future government actions involving AI procurement, requiring agencies to support national security restrictions with concrete technical evidence rather than hypothetical scenarios. It could also influence how federal agencies define “supply chain risk” for software and artificial intelligence providers.

Historically, such designations have been reserved for companies accused of espionage, foreign government influence, or demonstrated cybersecurity vulnerabilities. Applying the label to a U.S.-based AI developer over disagreements concerning acceptable use represents a significant expansion of that authority.

The dispute comes at a time when competition among leading AI developers has become increasingly intertwined with national security policy.

The Trump administration, meanwhile, has made accelerating AI adoption across the federal government a strategic priority while simultaneously tightening scrutiny of technology suppliers involved in sensitive government work.

Against that backdrop, the Anthropic case has become one of the first major legal tests of how far the federal government can go in restricting access to advanced AI technologies based on perceived national security concerns.