Alibaba built 10,000 AI chips — Samsung forecasts 8x profit

Broadcom's stock jumped 6% in a single session. Samsung just forecast an 8-fold increase in quarterly profits. And Alibaba quietly opened a new data center stocked with 10,000 chips it built entirely on its own. In the week of April 7–8, 2026, the global race for AI infrastructure didn't just accelerate — it rewrote who gets to compete at all.

For most people, AI feels like software: open an app, type a prompt, get an answer. But underneath every response from Claude, ChatGPT, or Gemini is a supply chain of rare materials, advanced manufacturing, and politically charged trade policy. This week made that supply chain impossible to ignore.

Broadcom's 6% day — and what Google and Anthropic just committed to

Broadcom — a semiconductor company that designs custom AI chips called ASICs (Application-Specific Integrated Circuits: chips engineered for one company's exact AI workload, rather than general-purpose computing) — saw its stock surge 6% after reports confirmed expanded supply agreements with both Google and Anthropic on April 7, 2026.

These aren't small purchase orders. When a hyperscaler (a term for companies operating at Google- or Amazon-scale infrastructure) commits to a Broadcom chip deal, they're locking in billions of dollars of manufacturing capacity years in advance. It's the AI equivalent of reserving every seat on every flight out of the airport before a holiday rush — you pay a premium to guarantee access when it matters most.

Anthropic's growing chip footprint carries a second layer of meaning. The company simultaneously signaled confidence in the broader AI enterprise ecosystem, with market analysts describing its commentary as an "AI stamp of approval" for cybersecurity stocks and established tech names. The implication: AI adoption is now mature enough to lift an entire category of companies, not just the AI labs at the center.

• Broadcom stock: +6% in a single session on April 7, 2026
• New deals confirmed: Expanded chip supply agreements with both Google and Anthropic
• Market read: Custom chip commitments signal that AI capex (capital expenditure on physical infrastructure) is structural — not a speculative bubble
• Downstream effect: Cybersecurity and enterprise tech stocks gained on Anthropic's market endorsement

Samsung's 8x profit forecast — the view from inside the AI chip factory

Samsung's quarterly earnings forecast landed with a number few analysts expected: an 8-fold jump in profits, driven almost entirely by AI chip demand. Samsung manufactures HBM (High Bandwidth Memory — a type of chip that sits directly next to AI processors and feeds them data at extremely high speeds, acting like a very fast short-term memory for the AI). Demand for HBM has outpaced global supply for over a year.

Samsung shares rose 5% on the forecast alone. That's the market pricing in a world where AI model training and inference (inference means running the model to generate answers for users, as opposed to training it on datasets) are consuming memory chips at a rate no analyst projected two years ago.

The scale of this shift is hard to overstate:

• A single large AI training run can consume more memory bandwidth than entire generations of research supercomputers from a decade ago
• Samsung's HBM is embedded inside Nvidia's most advanced AI GPUs (graphics processing units — originally designed for video games, now repurposed as the central engine of AI computation)
• An 8x profit swing is not incremental improvement — it reflects a complete restructuring of who the chip industry's most important customer is
• Samsung shares gained 5% on the announcement, with the company forecasting record earnings across the coming quarters

The bottom line: whoever controls AI memory chip supply controls the pace at which AI models can be trained and deployed globally. Samsung and SK Hynix currently hold that position for most of the Western AI ecosystem — and this week's numbers show how valuable that position has become.

Alibaba's counter-move — 10,000 chips the U.S. can't reach

While Broadcom and Samsung dominated Western financial media, a quieter but more strategically significant event unfolded in China. Alibaba launched a new data center equipped with 10,000 of its own proprietary AI chips — designed and built without U.S. components subject to export controls, and therefore entirely outside Washington's reach.

U.S. export controls restrict which advanced chips American companies can sell to Chinese buyers. Nvidia's H100 and H800 GPUs, for instance, are banned from export to China. The underlying assumption: if China can't purchase cutting-edge chips, it can't build cutting-edge AI systems.

Alibaba's 10,000-chip facility challenges that assumption head-on. The chips, reportedly called Zhenwu (named after a Chinese deity associated with power and protection), likely trail Nvidia's latest hardware in raw per-chip performance — but deploying 10,000 units in a single facility changes the competitive math entirely:

• Scale offsets performance gaps: More chips running in parallel can match fewer faster chips across many AI training and serving workloads
• Domestic supply eliminates political risk: No U.S. trade restriction can affect chips China designs and manufactures entirely within its own borders
• Software optimization is a force multiplier: Chinese AI labs have accumulated years of experience extracting maximum performance from hardware constrained by export controls
• Strategic precedent if it works: A successful 10,000-chip domestic AI facility validates China's path to full AI independence from Western supply chains

The bottleneck nobody talks about: Taiwan's packaging plants

There's a structural detail that rarely makes headlines but shapes everything downstream: even chips manufactured in U.S. factories often travel to Taiwan for advanced packaging (the process of physically connecting multiple chips together into a single functional unit — think of it as final assembly that turns raw components into the finished product). Taiwan's TSMC and its supplier ecosystem currently dominate this step globally.

This creates a paradox. The U.S. is committing hundreds of billions of dollars to build domestic chip fabs (fabrication plants — the actual factories where chips are physically manufactured). Yet many of those finished chips still require Taiwan's packaging infrastructure before they're ready to use. It's a bottleneck that neither capital nor policy has fully resolved.

For the AI infrastructure race, this dependency matters in three concrete ways:

• Supply chain disruptions in Taiwan directly throttle AI chip availability for every major U.S. AI lab — regardless of where the chips were originally made
• Geopolitical tensions around Taiwan remain the single largest wildcard in long-range AI infrastructure planning for companies globally
• This week's U.S.-Iran ceasefire announcement reduced some of the geopolitical risk premium — Alphabet, Meta, Amazon, and Nvidia all gained as markets digested the reduced Middle East supply chain threat

The infrastructure scorecard: who's ahead, who's at risk

This week's data drew unusually clear lines between AI infrastructure leaders and laggards. Here's how the field looks right now:

• Broadcom ✅ — Clear winner. Locked deals with two major AI labs in one week; stock +6%.
• Samsung ✅ — Clear winner. 8x profit forecast, +5% stock gain, irreplaceable HBM chip position in the Western AI supply chain.
• Google ✅ — Investing aggressively. New Broadcom chip commitment; CEO Sundar Pichai signaling AI startup investment activity.
• Meta ⬆ — Catching up fast. Launched its first major new AI model since the $14 billion acquisition of AI researcher Alexandr Wang.
• Anthropic ⬆ — Expanding quietly but consistently. New Broadcom deal secured; growing enterprise credibility driving broader market confidence.
• Apple ⚠ — Losing ground. Stock fell after foldable iPhone delays; analysts say the company has effectively "blown a five-year AI hardware lead."
• OpenAI ⚠ — Managing legal and security distractions simultaneously. Active Elon Musk lawsuit; limiting Mythos AI model rollout over cyberattack vulnerabilities.
• Alibaba 🌏 — Strategic wildcard. 10,000 proprietary chips, entirely domestic supply chain, entirely outside Western control. If this scales, it changes the global AI race permanently.

The pattern is consistent across every data point from this week: companies controlling physical AI infrastructure — chips, power, data center capacity — are pulling ahead. Companies whose edge is only in software are being outpaced by those who control the hardware underneath. If you're building with AI tools today, the capabilities you'll have access to in 12 months are being decided right now, in chip supply agreements and data center groundbreakings half a world away. Watch the infrastructure companies — they move before the model releases do. Track this race at AI for Automation News, or start with our beginner's guide to AI infrastructure to understand how hardware shapes the tools you use every day.

Related Content — Get Started | Guides | More News