Artificial intelligence (AI) is a technology that mimics human thought and information processing, which is beneficial for businesses in a variety of cutting-edge industries. Large volumes of data are processed by AI chips in developing AI technology. Companies that design
The AI industry garnered a lot of attention from the media and investors in 2023. We continue to discover new applications for AI technology, thus the market’s potential size is yet unclear.
Here are some of the latest developments in AI chips:
- MediaTek expands into AI data-center chips: MediaTek announced a $5 billion financing plan to accelerate development of custom AI chips (ASICs) for cloud providers. The company expects more than $2 billion in AI chip revenue during 2026 and aims to capture 15–20% of the AI ASIC market by 2027.
- Samsung says AI chip demand remains strong: Samsung reported a sharp increase in semiconductor profits, driven by AI-related demand. The company also warned that supply constraints for advanced chips could continue through 2028, highlighting ongoing pressure across the AI hardware supply chain.
- Advantest raises its outlook: Japanese chip-testing equipment maker Advantest increased its fiscal forecast because demand for testing hardware used in AI accelerators and data-center chips continues to grow.
- AI infrastructure investment continues: The AI boom is driving investment not only in chip designers but also in packaging, testing, and server manufacturing. Companies across the semiconductor ecosystem are expanding capacity to meet demand from AI data centers
Why Amazon builds its own chips
In the past, the market for AI chips hasn’t been eager to reduce prices. Rarely do market leaders cause their own disruptions. Amazon made a different decision.
Amazon creates chips from the ground up for particular applications instead of depending only on off-the-shelf CPUs.
Three chip families with different but complementary functions are the end result:
- The websites, apps, databases, and increasingly the agentic AI workloads that drive contemporary software are all managed by Graviton. 98% of our top 1,000 EC2 clients utilize Graviton, and revenue commitments have nearly tripled from quarter to quarter. The most recent version, Graviton5, is growing almost twice as quickly as Graviton4 and offers up to 25% higher performance than its predecessor.
AI chips built to meet customer demands at unprecedented scale
The majority of the inference in the company’s managed AI service, Amazon Bedrock, which is utilized by hundreds of thousands of users, is done on Trainium dollar compared to alternatives, are among the growing number of AI start-ups utilizing Trainium.
Individual chips, however, don’t tell the whole story. Amazon links them to progressively more potent systems:
- Compared to Trainium2 UltraServers, Trn3 UltraServers offer up to 4.4 times more computational power by combining up to 144 Trainium3 chips into a single integrated system. Customers may now train models in weeks rather than months thanks to this.
- The largest AI computing cluster in the world, Project Rainier was
- Energy efficiency: Trainium3 produces more than five times as many output tokens per megawatt as Trainium2; at scale, this efficiency
AI, silicon, and quantum: Amazon’s semiconductor business’s future
The roadmap for chips from Amazon keeps getting faster. Trainium4 is currently under development. Graviton is still developing for the age of agentic AI. Additionally, the corporation is merging these technologies so they complement one another under Peter
an advantage in how we build these models,” stated Peter DeSantis, SVP, Amazon AI Models, Chips, & Quantum Computing.
Why the Chip Selloff May Be Overdone
The larger chip selloff seems to be more motivated by profit-taking and valuation issues than by declining company fundamentals.Chip stocks were under pressure due to a number of factors, including unsatisfactory investor responses to SK Hynix’s earnings report despite record profitability, doubts about the rate of AI monetization, worries about hyperscaler capital Investors now primarily focus
That distinction is important.
The AI revolution hasn’t stopped since then. Instead, after semiconductor stocks saw enormous gains over the past year, expectations have now become extremely high. After such strong rallies, corrections of this size are not unusual, particularly in one of the market’s fastest-growing sectors.
The Oversold AI Leader: NVDA
Nvidia is still the clear leader in AI accelerators, despite the fact that its shares have been affected by the wider semiconductor slump.
