Google has introduced two new variations of its in-house Tensor Processing Units (TPUs), expanding its artificial intelligence chip portfolio as competition with Nvidia intensifies. The new processors, called TPU 8tx and TPU 8i Max, are designed for specialised AI workloads, including confidential computing, advanced coding assistants, mathematical reasoning and drug discovery. The announcement reflects Google’s strategy of developing specialised hardware rather than relying on a single chip configuration for every AI application.
The two variants build on Google’s eighth-generation TPU family, which already includes TPU 8t for large-scale AI model training and TPU 8i for inference workloads. The TPU 8tx focuses on protecting sensitive information during AI processing, while the TPU 8i Max combines two processors into a single package to provide additional computing and memory capacity. The move also strengthens Google’s ambition to sell more of its custom AI chips to external customers through Google Cloud.
Key Takeaways
- Google has introduced two new TPU variants: TPU 8tx and TPU 8i Max.
- TPU 8tx targets confidential AI computing for government agencies and regulated industries.
- TPU 8tx combines Google’s TPU technology with AMD processors and hardware-level encryption.
- TPU 8i Max combines two TPU 8i chips in one package.
- TPU 8i Max is designed for long-context AI workloads, including coding, mathematics and drug discovery.
- The two variants expand Google’s eighth-generation TPU portfolio to four configurations.
- The new systems are redesigned configurations of existing processors, not entirely new chip generations.
- Google is expanding its TPU business beyond internal applications through Google Cloud.
- Bloomberg Intelligence estimates TPU-related sales could exceed $135 billion by 2028, including associated software and hardware.
Google Introduces TPU 8tx and TPU 8i Max
Google’s latest announcement expands its custom AI semiconductor portfolio with two specialised configurations.
The company has been developing its own Tensor Processing Units for more than a decade. Initially designed to support Google’s internal computing requirements, TPUs have become an important part of its AI infrastructure.
The latest additions reflect changing requirements in the artificial intelligence industry.
Businesses increasingly need processors that can handle different types of workloads, from training large AI models to running autonomous agents and processing sensitive information.
Instead of developing one processor for every application, Google is introducing specialised configurations that address specific computing challenges.
Google’s chip chief Amin Vahdat told Bloomberg that the strategy reflects the company’s belief that specialised processors can deliver greater value than a single chip designed to serve the entire market.
Google’s Eighth-Generation TPU Portfolio
| Chip | Primary purpose | Main application |
|---|---|---|
| TPU 8t | AI model training | Training large and complex models |
| TPU 8i | AI inference | Running AI models and reasoning workloads |
| TPU 8tx | Confidential computing | Secure AI processing for sensitive data |
| TPU 8i Max | High-memory computing | Long-context AI agents and complex workloads |
The four configurations are designed to address different requirements across Google’s AI infrastructure ecosystem.
TPU 8tx Focuses on Secure AI Computing
One of the biggest challenges facing businesses adopting artificial intelligence is protecting sensitive information.
Government agencies, financial institutions, healthcare organisations and other regulated businesses often work with confidential information that cannot be exposed during AI processing.
Traditional encryption protects information while it is stored or transferred across networks. However, data generally needs to be decrypted while a processor performs calculations.
This creates a potential security vulnerability during computation.
Google’s TPU 8tx is designed to address this challenge through confidential computing.
How TPU 8tx Works
The TPU 8tx combines Google’s TPU technology with processors from Advanced Micro Devices (AMD).
It introduces hardware-level security features intended to protect sensitive data and AI model information while computations are taking place.
The system is designed for organisations that want to run AI workloads without exposing confidential information during processing.
Potential applications include:
- Government AI systems handling sensitive information.
- Financial institutions processing confidential customer data.
- Healthcare organisations working with protected medical information.
- Businesses running proprietary AI models.
- Companies training or fine-tuning models using private datasets.
The chip is also designed to support model fine-tuning, model distillation and reinforcement learning.
Google says the system can deliver up to twice the speed of its predecessor, with virtually no performance trade-off from the security features.
The combination of AMD processors and Google’s custom chip technology also demonstrates how semiconductor companies are increasingly collaborating to build specialised AI computing systems.
TPU 8i Max Targets Long-Context AI Applications
The second new variant, TPU 8i Max, focuses on applications that require AI models to process large amounts of information simultaneously.
Modern AI systems increasingly work with lengthy documents, large software codebases, extensive datasets and complex conversations.
This creates a challenge known as the memory bottleneck.
An AI processor may have substantial computing power, but its performance can be limited if it cannot access the information required for calculations quickly enough.
Google designed TPU 8i Max to address this limitation.
Two TPU Chips Combined Into One Package
Unlike a completely new processor architecture, TPU 8i Max combines two existing TPU 8i chips into a single package.
This configuration provides additional computing and memory resources.
Google says the new system doubles computing power, static random-access memory (SRAM) and high-bandwidth memory bandwidth compared with the existing TPU 8i.
The increased capacity is intended to support AI applications that need to work with much larger context windows.
A context window determines how much information an AI model can consider while responding to a request.
For example, a coding assistant may need to examine an entire software project rather than just a few lines of code.
Similarly, a drug-discovery system may need to process large scientific datasets and relationships between different biological structures.
Applications of TPU 8i Max
| Application | How the chip can help |
|---|---|
| Coding assistants | Process larger software projects and codebases |
| Mathematical reasoning | Handle complex calculations and extended reasoning tasks |
| Drug discovery | Analyse large scientific datasets |
| AI agents | Maintain and process more information during multi-step tasks |
| Long-context models | Work with larger amounts of information in a single request |
The design also increases computing density by placing more silicon within a comparable rack footprint.
This is important for data-centre operators because space, electricity and cooling capacity are major constraints when deploying large AI computing systems.
Why Google Is Developing Specialised AI Chips
The introduction of TPU 8tx and TPU 8i Max reflects a broader change in the AI semiconductor industry.
AI workloads are becoming increasingly diverse.
The computing requirements of training a large language model are different from those involved in operating an AI assistant or processing sensitive enterprise information.
A single general-purpose processor may not provide the best balance of performance, memory capacity, security and energy efficiency for every task.
Google believes specialised processors can improve efficiency by matching hardware configurations to specific workloads.
The company’s strategy also gives it greater control over the infrastructure used to operate its AI products.
Google already uses TPUs for services such as Search, YouTube and Gemini.
By developing its own processors, the company can optimise the relationship between hardware, software, networking and data-centre infrastructure.
This approach could reduce its dependence on external chip suppliers while creating another source of revenue through Google Cloud.
Google Is Challenging Nvidia in the AI Chip Market
The announcement comes as Nvidia continues to dominate the market for advanced AI accelerators.
Nvidia’s graphics processing units (GPUs) are widely used by technology companies, cloud providers and AI laboratories to train and operate large models.
However, Google is increasingly positioning its TPUs as an alternative for customers seeking different price-performance and infrastructure options.
The company’s advantage comes from its ability to design processors alongside its own software and cloud infrastructure.
Google has also started making TPUs available to external customers instead of reserving them primarily for internal use.
Google TPU Versus Nvidia GPU
| Feature | Google TPUs | Nvidia GPUs |
|---|---|---|
| Main developer | Nvidia | |
| Primary market | Google Cloud and selected external customers | Global AI and data-centre market |
| Hardware strategy | Specialised AI processor configurations | Broad GPU and AI computing platforms |
| Software ecosystem | JAX, PyTorch and Google Cloud tools | CUDA and a broad AI software ecosystem |
| Main opportunity | Integrated cloud and AI infrastructure | Widely adopted AI computing hardware |
The competition is not limited to chip performance.
Developers and enterprises also consider software compatibility, availability, energy efficiency, infrastructure costs and the difficulty of moving existing AI workloads between platforms.
Nvidia benefits from its established CUDA ecosystem and extensive customer relationships.
Google’s advantage is its ability to combine custom silicon with its cloud services and AI models.
The new TPU configurations could help Google attract customers whose workloads require specialised security or large memory capacity.
Google Is Turning TPUs Into a Bigger Cloud Business
Google’s custom chip strategy is increasingly becoming a commercial opportunity.
For years, the company’s TPUs primarily supported its internal services.
Now, Google is making more of its computing infrastructure available to external customers through Google Cloud.
This creates a new revenue opportunity beyond advertising and traditional cloud services.
According to Bloomberg Intelligence estimates cited by Bloomberg, TPU-related sales could exceed $135 billion by 2028, including associated software and hardware, compared with an estimated $12 billion in 2026.
These are analyst projections rather than confirmed Google revenue targets.
The company has also attracted interest from major AI and technology businesses.
Customers and reported TPU users include Apple, Meta, Oracle, Anthropic and OpenAI, although their individual arrangements and workloads differ.
Oracle has announced plans to make Google TPUs available to its cloud customers, while Anthropic has disclosed a substantial future purchasing option involving Google’s processors.
The expansion of external customers is important because it could turn Google’s internal semiconductor capabilities into a larger commercial business.
Why AI Companies Want Alternatives to Nvidia
Demand for AI computing has increased significantly as companies develop larger models and deploy AI applications across business operations.
Nvidia remains a major supplier, but relying on a single hardware ecosystem can create challenges.
These include:
- Supply constraints: AI companies may struggle to secure enough processors when demand exceeds available manufacturing capacity.
- Infrastructure costs: Advanced AI accelerators require substantial investment.
- Vendor dependence: Businesses may want alternatives to reduce dependence on one supplier.
- Specialised workloads: Different AI applications can benefit from different processor designs.
- Energy efficiency: Data-centre operators need to reduce electricity consumption as computing demand increases.
Google’s TPU expansion could provide another option for enterprises and AI developers.
However, competing with Nvidia requires more than developing powerful processors. Google must also make its hardware accessible, reliable and compatible with the software developers already use.
The Bigger Picture
Google’s introduction of TPU 8tx and TPU 8i Max highlights the next phase of competition in the artificial intelligence semiconductor market. The industry is moving beyond the race to build faster general-purpose AI processors towards specialised computing systems designed for security, memory-intensive applications and autonomous AI agents.
For Google, the opportunity is particularly significant because it can combine custom chips, Gemini models and Google Cloud infrastructure into one integrated offering. If the company attracts more external customers, TPUs could become a much larger commercial business and provide enterprises with an alternative to Nvidia’s dominant GPU ecosystem. However, software adoption, chip availability and manufacturing capacity will determine how quickly Google can turn its technological advantage into market share.
Frequently Asked Questions
1. Which two new AI chips has Google introduced?
Google has introduced TPU 8tx and TPU 8i Max, two specialised variations of its eighth-generation Tensor Processing Units.
2. What is Google TPU 8tx used for?
TPU 8tx is designed for confidential AI computing. It targets government agencies and regulated industries that need to protect sensitive information while AI models process it.
3. What is Google TPU 8i Max?
TPU 8i Max combines two TPU 8i chips into a single package to provide additional computing power and memory capacity. It is designed for long-context AI applications.
4. How does TPU 8tx work with AMD?
TPU 8tx combines Google’s custom TPU technology with AMD processors and hardware-level security features designed to protect information during computation.
5. How is TPU 8i Max different from TPU 8i?
TPU 8i Max combines two TPU 8i chips in one package. Google says this doubles computing power, SRAM capacity and high-bandwidth memory bandwidth compared with the standard TPU 8i.
6. Is Google trying to compete with Nvidia?
Yes. Google is expanding its TPU business to attract customers that currently depend on Nvidia GPUs. Its specialised processors provide another option for AI training, inference and enterprise workloads.
7. Who uses Google’s TPU chips?
Google uses TPUs internally for products such as Gemini, Search and YouTube. External customers and reported users include companies such as Apple, Meta, Oracle and Anthropic.
8. When will TPU 8tx and TPU 8i Max become available?
Google has not announced complete public availability details, pricing or supported regions for the two new configurations. The reported rollout is expected to take place in 2027.
Looking Ahead
Google’s specialised TPU strategy could strengthen its position in the global AI infrastructure market. TPU 8tx addresses an important requirement for businesses handling sensitive information, while TPU 8i Max targets the growing demand for AI systems capable of processing large amounts of context. These capabilities could become increasingly valuable as companies deploy autonomous agents and more complex AI applications.
The next challenge will be turning these technical capabilities into widespread commercial adoption. Google must ensure that its chips remain competitive in performance, pricing, availability and software compatibility. If it succeeds, the company could become a more significant competitor to Nvidia while expanding Google Cloud’s role in the rapidly growing AI computing market.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
