Alibaba’s Qwen challenge, Amazon’s $3T AI surge, and Apple’s subscription pivot

The AI race is no longer just about who has the biggest model or fastest chip. It’s now a full-stack story: cloud platforms, energy infrastructure, and even how we buy our phones are being reshaped around AI demand. In this breakdown, we’ll look at Alibaba’s new Qwen model, Amazon’s $3 trillion AI moment, Apple’s shift to hardware subscriptions, and the startups racing to power the AI boom.

Alibaba’s Qwen 3.8 Max takes aim at Anthropic

Alibaba has released Qwen 3.8 Max, its latest large language model, and is openly positioning it as a rival to leading U.S. “frontier” models like Anthropic’s Claude.

The company says Qwen 3.8 Max:

• Has around 2.4 trillion parameters, similar in scale to other recent frontier models like Kimi K3 from China.
• Performs strongly on benchmarks related to research reproduction and “agentic” computer use (letting AI act more like an assistant that can operate software and tools).
• Is priced aggressively at about $2 per 1 million input tokens, which is significantly cheaper than many U.S. frontier models.

Investors liked the news: Alibaba’s U.S.-listed shares jumped about 5%, and the stock is up roughly 35%–40% from its June lows, driven by renewed optimism around its AI and cloud businesses.

Why cheaper AI models matter

Alibaba is pitching Qwen 3.8 Max as “frontier performance at a fraction of the cost.” That has two big implications:

• Faster adoption: Lower prices can make it easier for startups and enterprises to experiment with advanced models, especially in cost-sensitive markets.
• Pricing pressure: If Qwen’s performance holds up in real-world use, it could force U.S. model providers to rethink their pricing, particularly for high-volume workloads.

Still, U.S. models retain advantages in areas like trust, reliability, ecosystem integrations, and turnkey features. Many enterprises will pay a premium for stability, compliance, and the convenience of not having to heavily customize or maintain models themselves.

If you want a deeper dive into how Apple, Google, OpenAI, and others are positioning their models, check out our broader roundup of the current landscape in the latest wave of new AI models.

Amazon passes $3 trillion on the back of AI and AWS

Amazon has joined the ultra-elite $3 trillion market cap club, becoming the fifth company ever to hit that milestone. The catalyst: a blowout earnings report driven by Amazon Web Services (AWS) and its AI business.

AWS CEO Matt Garman highlighted several key points:

• $25B AI revenue run rate: Amazon’s AI-related business on AWS has reached a $25 billion annual run rate, spanning both training and inference workloads.
• Broad-based demand: Growth isn’t just from a few frontier labs like OpenAI or Anthropic. Enterprises across finance, healthcare, retail, and media are ramping up AI usage.
• Inference is taking over: While training remains important, a growing share of spend is shifting to inference—actually running models in production to power apps, agents, and automation.

AWS chips and the economics of AI compute

Amazon is also leaning hard into custom silicon to control costs and performance:

• Trainium and Graviton: AWS’s in-house chips now represent a $25 billion run-rate business as well. Customers don’t buy the chips directly; they rent cloud capacity powered by them.
• Cost savings: For certain workloads, AWS says customers can cut inference costs by 20%–30% by using Trainium-based infrastructure instead of standard GPU-only setups.
• NVIDIA still central: AWS remains one of NVIDIA’s largest customers, offering Blackwell and other GPUs alongside its own chips, giving customers a mix-and-match option.

This hybrid approach—NVIDIA GPUs plus AWS custom chips—is central to Amazon’s pitch: choice, lower total cost, and a tightly tuned stack from data center to model runtime.

Amazon’s massive AI capex bet

To keep up with demand, Amazon has raised its 2024 capital expenditure plans to around $220 billion, most of it tied to AI and cloud infrastructure. According to Garman:

• Much of AWS’s capacity is already spoken for through 2027 and into 2028 via multi-year customer commitments.
• Demand for AI compute still significantly exceeds supply, pushing Amazon to build faster.
• Capex is likely to remain elevated as long as AI workloads continue to grow at this pace.

Amazon also recently signed an open-weights policy letter, signaling support for a balanced regulatory approach that doesn’t over-restrict open or semi-open models. The company wants to keep offering both closed frontier models and open-weight options via its Bedrock platform, so customers can choose what fits their needs and budgets.

Apple quietly turns hardware into a subscription

While Amazon and Alibaba fight over cloud and models, Apple is reshaping how people pay for devices. Instead of dropping a large lump sum for an iPhone or Mac, more customers are being nudged into monthly payments and upgrade programs.

Here’s what’s changing:

• From product to subscription: Apple is increasingly positioning iPhones, Macs, and Apple Watches as part of a recurring payment ecosystem, similar to how carriers have long sold phones on monthly plans.
• Regular upgrade cycles: With predictable annual or biannual hardware refreshes (iPhone and Apple Watch in September, Macs in fall or spring), subscriptions make it easier to keep users on the latest devices.
• Second-hand and parts value: Apple can refurbish or resell returned devices, or reuse parts, improving margins and supporting a growing secondary market.

For Apple, this means payments from customers never really stop. Over time, that can translate into higher revenue and more predictable cash flow, which Wall Street loves.

Apple is also pushing hard into on-device AI and a revamped Siri. The new Siri is finally becoming competent enough to feel like a real assistant, even if it’s still not as flexible as leading chatbots. For more on that strategy, see our breakdown of why Apple’s next trillion-dollar bet is on-device AI.

Palantir’s AI momentum and valuation questions

Palantir continues to be a lightning rod in the AI software space. The company has built its reputation on government and defense contracts, but its fastest-growing segment is now U.S. commercial customers.

Key dynamics to watch:

• Commercial vs. government: Analysts expect Palantir’s commercial growth to surpass government growth for the first time, a big psychological and strategic shift.
• What the software does: Palantir’s platforms help organizations pull data from many sources, clean and integrate it, and then apply AI models (including third-party LLMs) to make decisions, run simulations, or automate workflows.
• Valuation reset: The stock once traded at ~240x earnings; it’s now closer to 67x—still expensive, but far less frothy. Investors are watching whether the company can keep justifying its premium with sustained AI-driven growth.

Nuclear and batteries: powering the AI era

AI doesn’t just need GPUs and models—it needs enormous amounts of electricity. Two startups featured in the discussion highlight how energy infrastructure is being rethought for the AI age.

Valar Atomics: nuclear for data centers

Valar Atomics is building what it calls America’s first “gigesite” network: vertically integrated industrial nuclear power sites designed to feed AI data centers.

Recent milestones:

• A $1 billion Series B round led by Sequoia, plus a $200 million credit facility from JPMorgan.
• A live demo where an advanced nuclear reactor directly powered an NVIDIA Blackwell chip—symbolic, but important for investor confidence.
• An ambition to move from building one reactor a year to one every six months, then monthly, aiming to become “the biggest energy company on Earth.”

Valar’s long-term plan is to finance reactors largely through debt and customer contracts, rather than relying solely on equity. The pitch: AI leaders need massive, reliable power near their data centers, and nuclear is uniquely suited to that.

Base Power: home batteries as grid infrastructure

Base Power (often stylized as BASE) is tackling the grid from the opposite direction: the home. The company just raised a $1 billion Series D at a $13 billion valuation to scale its residential battery systems.

What makes Base different from a typical home battery:

• Grid-first design: The battery is larger than typical home units and is treated as grid infrastructure, not just a backup box in your garage.
• Owned and operated by Base: Instead of you buying the hardware outright, Base owns and operates a fleet of batteries across homes, using them as a virtual power plant to support the grid during peak demand.
• Not tied to solar: While it can work alongside solar, the core idea is about distribution and time-shifting energy—batteries “move energy through time” much like wires move it through space.

Base manufactures its batteries in Texas and keeps engineering and manufacturing tightly co-located to iterate quickly. It works closely with utilities (municipal, co-ops, and investor-owned) as a provider of choice, helping them handle surging demand from AI, electrification, and EVs.

Security, chips, and the AI talent gap

Several other developments round out the picture of an AI-driven economy:

• Crypto hardware hack: CoinKite, maker of the Coldcard hardware wallet, disclosed a software flaw that exposed some Bitcoin recovery keys, leading to an estimated $110 million theft from around 5,000 wallets. It’s a reminder that even “offline” hardware security can be undermined by software bugs.
• AI chip startups surge: A South Korean AI chip company reportedly quadrupled its valuation to $2.2 billion after landing more than 30 contracts across eight countries, underscoring intense global competition in AI accelerators.
• AI skills vs. MBAs: In a survey of 1,000 finance executives, 86% said AI training is now more valuable than an MBA for new hires, and over 90% are willing to pay more for employees with strong AI literacy.

Apple, Amazon, Alibaba: three different AI plays

Put together, these stories show how differently the tech giants are approaching the same AI wave:

• Alibaba: Competing on model performance and price, especially in China and emerging markets, while trying to reignite growth in its cloud business.
• Amazon: Building the infrastructure layer—cloud, chips, and platforms—for everyone else’s AI, and spending heavily to stay ahead.
• Apple: Turning hardware into a subscription-like service and betting on tightly integrated, on-device AI to quietly win the consumer experience.

Behind them, energy startups like Valar Atomics and Base Power are racing to make sure the grid doesn’t become the bottleneck for AI progress. And across industries, companies are scrambling to hire people who understand AI tools as well as, or better than, traditional business credentials.

The next phase of the AI boom won’t just be about smarter models—it will be about who can build the most resilient, affordable, and scalable stack from silicon to software to electricity.