Amazon considers selling its in-house AI chips, eyeing $50 billion annual revenue

Andy Jassy, Amazon's CEO, revealed in his annual shareholder letter on Thursday that the company is considering a shift from its previous model of exclusively using its own chips to selling its self-developed chips directly to third-party customers. The chip division's annualized revenue has already surpassed $20 billion, and if the business opens to external sales, it could surge to $50 billion .
From Renting to Selling: A Shift in Business Model
For years, Amazon's chip strategy has diverged sharply from those of Nvidia or Intel. Instead of selling physical hardware, Amazon previously 'rented' the computing power of chips like Trainium and Inferentia through its AWS cloud service. In his letter, Jassy said that due to extremely strong market demand, Amazon may soon begin selling entire rack-mounted chips directly to third parties.
Surge in Production Driven by Computing Power Demand
Amazon's self-developed chips are currently in short supply:
Trainium 2: The second-generation AI training chip, with a cost-performance ratio about 30% better than similar GPUs, is already sold out.
Trainium 3: Scheduled to ship in early 2026, it offers 30% to 40% better performance, and its current subscription rate is nearly full.
Trainium 4: Although still about 18 months from official launch, a significant portion of its production capacity is already reserved.
Alternatives to Counter Nvidia's Dominance
As global demand for computing power in AI model training explodes, companies are urgently seeking alternatives to Nvidia. Jassy noted that AWS's current production capacity bottleneck is preventing some demand from being met. To address this, the company plans to double its total power capacity by the end of 2027. He emphasized that AI represents a once-in-a-lifetime opportunity, and Amazon will not be conservative in its approach to this technological wave.
This strategic shift means Amazon is transitioning from a single cloud service provider to a more aggressive hardware supplier, directly entering the core AI infrastructure market dominated by Nvidia.
Related article
Lingguang App Refreshes Circle Community With Hot List, Follow Features and PC Support
On July 16, AIBase discovered that Ant Group’s full-modal AI assistant, the "Lingguang App," has updated its instant application community, "Lingguang Circle." This latest version introduces features like a "Hot List," "Editor's Picks," and the abili
Horizon Unveils White-Box Licensing for Core i7, Accelerating Mass Production of L3 and L4 AI Models
Horizon Robotics is evolving from a Chinese autonomous driving innovator into the intelligent driving brain for a global automotive leader in China. As reported by Gelonghui, Horizon and Volkswagen Group officially announced on July 22 that their joi
xAI Co-Founder Steps Down as Only Five Original Team Members Remain
xAI, Elon Musk’s artificial intelligence venture, has undergone another round of significant staff turnover. Co-founder Toby Pohlen has officially announced his exit. Sharing his thoughts on X, Pohlen reflected on his three-year tenure, noting he han
Related Special Topic Recommendations
Comments (0)
0/500

Andy Jassy, Amazon's CEO, revealed in his annual shareholder letter on Thursday that the company is considering a shift from its previous model of exclusively using its own chips to selling its self-developed chips directly to third-party customers. The chip division's annualized revenue has already surpassed $20 billion, and if the business opens to external sales, it could surge to
From Renting to Selling: A Shift in Business Model
For years, Amazon's chip strategy has diverged sharply from those of Nvidia or Intel. Instead of selling physical hardware, Amazon previously 'rented' the computing power of chips like Trainium and Inferentia through its AWS cloud service. In his letter, Jassy said that due to extremely strong market demand, Amazon may soon begin selling entire rack-mounted chips directly to third parties.
Surge in Production Driven by Computing Power Demand
Amazon's self-developed chips are currently in short supply:
Trainium 2: The second-generation AI training chip, with a cost-performance ratio about 30% better than similar GPUs, is already sold out.
Trainium 3: Scheduled to ship in early 2026, it offers 30% to 40% better performance, and its current subscription rate is nearly full.
Trainium 4: Although still about 18 months from official launch, a significant portion of its production capacity is already reserved.
Alternatives to Counter Nvidia's Dominance
As global demand for computing power in AI model training explodes, companies are urgently seeking alternatives to Nvidia. Jassy noted that AWS's current production capacity bottleneck is preventing some demand from being met. To address this, the company plans to double its total power capacity by the end of 2027. He emphasized that AI represents a once-in-a-lifetime opportunity, and Amazon will not be conservative in its approach to this technological wave.
This strategic shift means
Lingguang App Refreshes Circle Community With Hot List, Follow Features and PC Support
On July 16, AIBase discovered that Ant Group’s full-modal AI assistant, the "Lingguang App," has updated its instant application community, "Lingguang Circle." This latest version introduces features like a "Hot List," "Editor's Picks," and the abili
Horizon Unveils White-Box Licensing for Core i7, Accelerating Mass Production of L3 and L4 AI Models
Horizon Robotics is evolving from a Chinese autonomous driving innovator into the intelligent driving brain for a global automotive leader in China. As reported by Gelonghui, Horizon and Volkswagen Group officially announced on July 22 that their joi
xAI Co-Founder Steps Down as Only Five Original Team Members Remain
xAI, Elon Musk’s artificial intelligence venture, has undergone another round of significant staff turnover. Co-founder Toby Pohlen has officially announced his exit. Sharing his thoughts on X, Pohlen reflected on his three-year tenure, noting he han





Home






