Compute & Silicon
Why On-Chip Memory Is Getting Harder to Shrink
Logic transistors keep getting denser, but the fast SRAM memory built into chips has not kept pace. Here is why, and how designers are responding.
Every processor contains fast memory built right into the silicon. It is called SRAM, and AI accelerators rely on it heavily to keep their math units fed. For decades, SRAM shrank along with everything else on a chip. In recent process generations, that has become much harder.
What on-chip SRAM does
SRAM forms the caches and buffers that hold data a processor is about to use. It is far faster than external memory, and reading from it uses much less energy. The more data a chip can keep in SRAM, the less often it has to make costly trips to external memory.
In AI accelerators, SRAM buffers hold pieces of matrices, intermediate results and frequently reused weights. Some AI chip designs devote a large share of their area to SRAM precisely because it reduces data movement.
How an SRAM cell is built
A standard SRAM cell stores one bit using a small group of transistors, commonly six, arranged in a loop that holds its value steady while power is on. The cell must be stable enough to keep its value reliably, yet easy enough to read and write quickly.
Those transistors must be carefully balanced against each other. As transistors shrink, small variations from manufacturing have a bigger effect, which makes balancing them harder.
Why SRAM scaling has slowed
New process generations continue to deliver meaningful gains in logic density, the space taken by circuits that compute. SRAM cells, however, have shrunk much less in recent generations. Several factors contribute:
- SRAM cells already use very tightly packed layouts, leaving little room to squeeze further.
- Variation between transistors makes small cells less stable and less reliable.
- The wiring needed to connect the cells has its own limits.
- Lower operating voltages, useful for saving power, make stable SRAM cells harder to design.
The result is that SRAM takes up a growing share of the cost of an advanced chip, since it benefits less from each expensive new process.
How designers are responding
Chip architects have several options:
- Place SRAM on a separate die made with an older, cheaper process, then stack or connect it closely to the compute die using advanced packaging.
- Stack SRAM directly on top of a processor die, an approach already used in some commercial processors to add large caches.
- Use new transistor structures, such as gate-all-around designs, which may help improve SRAM density.
- Rethink software and chip architecture to make better use of limited on-chip memory.
Researchers are also exploring alternative memory technologies that might someday offer more density or lower power for some uses, though SRAM remains the standard for the fastest on-chip storage.
Why it matters for AI
AI workloads reward large amounts of fast memory close to computing units. If SRAM becomes relatively more expensive, designers must make careful choices about how much to include and where to put it. That is one reason memory placement, stacking and packaging have become central themes in AI chip design.
The slowing of SRAM scaling is a good example of how chip progress has become uneven. Some parts of a chip continue to improve rapidly, while others hit physical limits, pushing engineers to find creative solutions elsewhere in the system.
A Worked Example: Why Uneven Scaling Changes the Budget
Simplified numbers help show why SRAM’s slow shrinking matters. Imagine a chip design where, on an older process, half the area is logic and half is SRAM. The figures here are illustrative only.
- Move to a new process. Suppose logic shrinks to roughly half its old area, while SRAM shrinks only slightly, say to nine-tenths of its old area.
- Add it up. The logic half becomes a quarter of the original chip’s area. The SRAM half becomes about forty-five percent. The new chip is about seventy percent of the original size, not half.
- Look at the mix. SRAM now takes up nearly two-thirds of the new die, even though the design did not add any memory.
- Consider cost. Because the newest processes cost more per wafer, a design where most of the area barely shrank may not save much, or may even cost more, than staying on the older process for that part.
That arithmetic is why chip architects increasingly ask which parts of a design truly need the newest process and which could be made more economically elsewhere and joined through packaging.
Common Misunderstandings
“SRAM has stopped improving”
It still improves, just more slowly than logic in recent generations. Designers continue to find gains through new transistor structures, better circuit techniques and clever layout, but the pace is not what it once was.
“More cache always means faster AI”
Extra on-chip memory helps when a workload can reuse the data stored there. If data is used only once, a larger cache may add cost without much benefit. The value depends on how the workload accesses memory.
“Moving SRAM to another die is free”
Separating memory from logic adds packaging steps and some distance between them. Stacking keeps that distance very short, but it introduces its own manufacturing complexity and heat management questions.
What This Means for Readers
The slowing of SRAM scaling shows up in several trends worth recognizing:
- Processors marketed with stacked or extra-large caches are often responding to this very challenge.
- Chiplet designs that put memory and logic on different processes reflect the same economics.
- Announcements about new transistor structures may mention SRAM density specifically, because it has become a closely watched measure of a process’s value.
- Software that reuses data well inside on-chip memory can be as important as adding more memory.
Questions to ask when you read about chip memory
- Is the SRAM on the same die as the compute, stacked above it or on a separate chiplet?
- Which process is each piece made on, and why?
- Is the claimed gain about capacity, speed or energy per access?
Questions Readers Ask
Why not replace SRAM with DRAM on the chip?
DRAM is denser but slower and needs refreshing, and it is normally made on specialized processes that differ from logic processes. For the fastest storage right next to the math units, SRAM remains the practical choice.
Do the newer memory technologies have a timeline?
Research directions vary widely in maturity. Some have found use in specific products, while others remain in development. It is reasonable to treat broad replacement claims cautiously until products ship at scale.
This explainer is general education and not investment advice. For financial decisions involving chip companies, consult a qualified financial professional.
Latest stories
View allCompute & Silicon
Why a Cold Cache Changes the Meaning of a Fast Response Test
A service that has recently handled a request may respond differently from one meeting it for the first time. Data, compiled work or model state can already be available, so a warm test and a cold test answer different questions.
Photo: Kevin Ache / Unsplash
A service that has recently handled a request may respond differently from one meeting it for the first time. Data, compiled work or model state can already be available, so a warm test and a cold test answer different questions.
The operating distinction
Caches keep selected information close to where it will be used. Reusing that information can reduce retrieval or preparation work. A cold start may need to rebuild it, load a model or fetch data from a slower location. Neither condition is inherently the only valid test; the expected workload determines the mix.
Preparation can also occur outside the part of the system being timed. A demonstration may load information before starting its stopwatch, while a user experiences that loading as part of the wait. That does not invalidate the demonstration, but it changes the boundary of the claim.
A practical example
Imagine comparing two demonstrations of the same service. One repeats a familiar query while the other begins after a restart. Their response times may differ without any change in the underlying hardware. A fair comparison describes what was already loaded and what preparation was included.
Questions worth asking
A comparison should include preparation that matters to the intended user experience. Keeping warm and cold observations separate makes both useful without presenting the faster result as representative of every possible request.
- Was the test performed after restart or after repeated requests?
- Which preparation steps were included in the measured response?
- How often does the real workload encounter the cold condition?
What the evidence supports
A single fast example cannot establish the experience across restarts, uncommon requests and traffic changes. Conversely, an unusually slow first request should not automatically represent steady operation. Readers need the operating state, the sampling method and the service objective before drawing a conclusion.
A useful comparison can present both first-use and repeated-use observations under clearly described conditions. It should identify whether the test reuses the same request or exercises a broader set of inputs. The reader can then decide which observation resembles the intended application. The aim is an honest description of service behavior, not a contest that rewards whichever test omits the most preparation work.
Treat cache state as part of the test description, alongside the machine and the application.
Quick check
How much did you catch? Tap a question to see the answer.
1. Why do warm and cold tests give different results?
Answer: Data, compiled work or model state may already be loaded in a warm test.
2. Is a cold test the only valid one?
Answer: No. The expected workload decides the right mix.
3. What should a fair comparison describe?
Answer: What was already loaded and what preparation was included in the timing.
Where Warm State Lives in a Modern Service
The word cache can sound like a single component, but a typical service holds several layers of prepared state at once. Each layer can be warm or cold independently, which is one reason response tests can be difficult to compare.
- Model weights in memory. A model already loaded onto an accelerator is ready to run. A model that must be read from storage first adds loading time.
- Compiled or optimized code. Some systems prepare optimized execution paths the first time a particular shape of work appears and reuse them afterward.
- Request-level reuse. Inference systems may reuse computation for repeated prompt prefixes or recently seen inputs.
- Application data caches. Frequently retrieved documents, records or results may be held in faster memory near the service.
- Connection and session state. Established network connections, authentication tokens and open sessions avoid setup work for subsequent requests.
- Storage and operating system caches. Files read recently may already sit in memory, making later reads faster.
Designing a Fairer Response Test
Anyone evaluating a claim about response speed, or designing a test of their own, can work through a short sequence.
- Describe the starting state. State whether the test began after a restart, after a period of inactivity or after a run of similar requests.
- Define the timing boundary. Record exactly when the clock starts and stops, and which preparation steps fall inside or outside that window.
- Vary the inputs. Use a set of requests that resembles the real workload rather than repeating one query.
- Report a distribution. Present typical and slower responses, not only the fastest or the average.
- Estimate the real cold rate. Consider how often actual users will meet a cold condition, based on traffic patterns, scaling behavior and restarts.
- Keep the results labeled. Publish warm and cold observations side by side with their conditions.
Common mistakes
A frequent mistake is running a test many times and reporting the best result, which almost always reflects fully warm conditions. Another is forgetting that automatic scaling can introduce cold starts whenever new capacity is added during a traffic increase. A third is comparing two systems where one test included model loading and the other did not. A fourth is assuming that a cold start happens only after a deliberate restart, when cache eviction under memory pressure can produce similar effects.
What This Means for Buyers and Readers
For a business choosing an inference service or reading a vendor’s performance claim, the cache question is practical. A service that is consistently fast for repeated requests may still feel slow to users who arrive with varied or uncommon requests, or who arrive just after a scaling event. Asking how the provider handles cold conditions, and how often they occur in practice, gives a more realistic sense of what users will experience.
Fast is a description of a particular state. The useful question is how often the real workload is in that state.
Questions Readers Ask
Is a warm test misleading?
Not if it is labeled. Warm results describe steady operation for repeated work, which is useful. They become misleading only when presented as the whole experience.
Can cold starts be eliminated?
They can be reduced by keeping capacity loaded and preparing state in advance, but this usually carries a cost in reserved resources.
Why does the first request after a restart take longer?
The service may need to load models, prepare execution paths, open connections and fill caches before it can respond.
What should a buyer ask a provider?
Ask what conditions their published response figures assume and how the service behaves after restarts, scaling events and long idle periods.
AI infrastructure explainers and news on iPhone, iPad and Apple Watch.
Compute & Silicon
An Idle Accelerator Can Still Be Waiting on Useful Work
A low utilization reading does not identify a single problem. An accelerator can be waiting for data, another machine, a scheduling decision or a request from a user. Understanding the waiting state matters more than assigning blame to the chip.
Photo: Anne Nygård / Unsplash
A low utilization reading does not identify a single problem. An accelerator can be waiting for data, another machine, a scheduling decision or a request from a user. Understanding the waiting state matters more than assigning blame to the chip.
The operating distinction
Useful work passes through several stages before an accelerator starts arithmetic. Data must be available, memory must have room and dependent tasks must finish. A scheduler may also hold capacity for a workload with a deadline. That reserved headroom can look similar to inefficiency on a broad average.
Different kinds of waiting call for different responses. Adding processors will not necessarily repair a slow input pipeline, and changing scheduling can affect one customer's response while improving another's throughput. The resource graph and the service objective need to be read together.
A practical example
Imagine an inference service that keeps spare resources available for sudden requests. Filling every available slot with background work could increase the average utilization while making urgent requests slower. A training job with predictable input and long execution periods would present a different operating problem.
Questions worth asking
Reserved headroom, missing input and inefficient execution are different explanations for unused capacity. The next operational question should follow the observed cause rather than assume that every idle interval represents the same problem.
- What resource or dependency is the accelerator waiting for?
- Does the measurement include reserved capacity and maintenance periods?
- Did higher utilization improve completed work or merely lengthen queues?
What the evidence supports
Utilization needs a definition and a time window. A device-level percentage alone cannot show whether a service met its response target or whether a job completed on schedule. Compare the waiting causes and user outcome before comparing the headline averages.
A useful account identifies the interval when the observation was made and describes the mix of jobs present. Readers can then distinguish an idle system with no demand from a busy service deliberately retaining headroom. Both can show unused device capacity, but they tell different stories about the amount of work waiting and whether customers are receiving the experience the service was designed to provide.
The useful target is productive work within the service requirement, not a percentage detached from its purpose.
Quick check
How much did you catch? Tap a question to see the answer.
1. Does low utilization point to a single problem?
Answer: No. An accelerator may be waiting on data, another machine, scheduling or users.
2. Why can reserved headroom look inefficient?
Answer: It can resemble idle capacity on a broad average.
3. Will adding processors fix a slow input pipeline?
Answer: Not necessarily.
Sorting Idle Time Into Categories
A single utilization line becomes far more useful once the unused intervals are split into a few named categories. The categories do not need to be elaborate. They only need to separate explanations that call for different responses, so that a review meeting can discuss causes instead of a number.
- No demand. Nothing was waiting. The service was available and nobody needed it during that interval.
- Deliberate headroom. Capacity was held back on purpose so urgent or latency-sensitive requests could start quickly.
- Input starvation. Work existed, but data, model weights or request batches were not arriving fast enough to keep the device busy.
- Coordination waits. The device was ready but waiting on another machine, a synchronization step or a dependent task.
- Scheduling gaps. Work was queued, yet placement, priority or fragmentation kept it from landing on available hardware.
- Planned unavailability. Maintenance, testing, updates or recovery activity removed the device from service.
Even a rough split changes the conversation. If most idle time belongs to deliberate headroom, the question becomes whether the reserve is sized correctly for the response target. If most belongs to input starvation, the attention moves to storage, networking or preprocessing. Neither answer requires buying or removing accelerators.
A Step-by-Step Review of Unused Capacity
An operator or analyst trying to understand a low reading can work through a short sequence before drawing conclusions.
- Fix the time window. State whether the figure covers an hour, a day, a month or a reporting quarter. Overnight lulls and weekend demand can dominate a long average.
- State what counts as busy. Some measurements record whether any work is scheduled on the device, while others try to estimate how much of its arithmetic capability is in use. The two can tell very different stories.
- Separate the workload mix. Training, batch processing and interactive inference have different shapes. Averaging them together hides the behavior of each.
- Place the waits on a timeline. Line up idle intervals against data transfers, checkpoint writes, job starts and request bursts. Patterns often appear where they would never show in a single percentage.
- Check the outcome. Ask whether jobs finished on schedule and whether response targets were met during the same window.
- Choose the response that matches the cause. Only then decide whether the answer is more capacity, better scheduling, a faster input path or no change at all.
Common mistakes
The most frequent error is treating utilization as a goal in itself. A team rewarded for a higher figure can fill reserved headroom with low-priority work, which raises the average while making urgent requests wait. Another mistake is comparing utilization across facilities or providers without confirming that the definitions match. A third is ignoring the cost of reaching a higher number, such as longer queues or more frequent preemption of jobs that then have to restart.
What This Means for Readers Following the Industry
Utilization figures appear in earnings discussions, vendor presentations and infrastructure announcements. For a reader trying to understand the business, the figure is a starting point for questions rather than a verdict. A high number may reflect strong demand, or it may reflect a fleet with too little headroom for spikes. A low number may signal weak demand, or a service that keeps capacity in reserve because its customers expect fast responses.
The more useful question is usually what the figure was designed to show and what it leaves out.
This is general education about how infrastructure metrics are read, not a view on any company or security. Anyone weighing a financial decision should consult a qualified financial professional about their own situation.
Questions Readers Ask
Is a fully utilized accelerator always the efficient outcome?
Not necessarily. A device kept constantly busy may leave nothing available for sudden requests. Whether that is efficient depends on the service the device is meant to support.
Can idle time be measured precisely?
It can be measured, but the result depends on the definition, the sampling interval and the tools used. Short waits can disappear inside coarse samples.
Why would a provider intentionally leave capacity unused?
To meet response commitments, absorb failures, allow maintenance or accept new work without delay. Each of those purposes has value that an average cannot display.
What is the single best follow-up question?
Ask what the accelerator was waiting for during the idle intervals. The answer points toward the next useful decision.
AI infrastructure explainers and news on iPhone, iPad and Apple Watch.
Compute & Silicon
Custom AI ASICs vs. GPUs
Custom AI ASICs can deliver strong efficiency on a narrow workload, while GPUs remain valuable because their software ecosystem and flexibility absorb uncertainty.
The custom-ASIC-versus-GPU debate is often framed as a contest between a cheaper chip and a more expensive one. That misses the real decision. A GPU is a general-purpose accelerator with a large software ecosystem. An application-specific integrated circuit is a narrower machine designed around a workload, a power target, or a system architecture.
The winner depends on whether the workload is stable enough to justify specialization. Training, inference, recommendation, search, and video pipelines do not stress hardware in exactly the same way. A model that is efficient today can change after a new architecture, quantization method, context length, or customer demand appears.
What the GPU buys
A GPU’s first advantage is optionality. It can run different model families, numerical formats, kernels, and frameworks. It can be reassigned when demand shifts. That matters for a cloud provider serving many customers and for a company whose AI roadmap is still moving.
The second advantage is software. Mature compiler tools, libraries, kernels, orchestration systems, and developer familiarity reduce the risk between purchasing hardware and putting it into production. A GPU’s list price is not the whole cost; neither is an ASIC’s lower unit cost if engineers spend months rewriting software.
The third advantage is scale. GPUs can be linked into standard or proprietary fabrics, with established server, cooling, and support ecosystems. A customer can often rent them before it owns them, which converts a large capital decision into a more flexible operating expense.
The GPU’s weaknesses are also clear. The chip may carry circuitry for workloads the customer does not use. Its memory and network design may be optimized for a broad market rather than one application. High demand can create shortages, and power density can push a facility’s cooling and electrical systems beyond their original design.
What the ASIC buys
A custom ASIC can remove general-purpose features and use the saved area or power for the specific operations that matter. It may deliver more performance per watt, better latency, or lower cost per inference when the workload is predictable.
An ASIC can also be part of a broader system strategy. Google and other hyperscalers have built custom accelerators because owning the silicon roadmap can reduce dependence on a merchant vendor and let the company tune hardware to its software stack. Intel described an expanded 2026 collaboration with Google around custom ASIC-based infrastructure processing units that offload networking, storage, and security functions from host CPUs. That is a reminder that “custom AI silicon” is not limited to a chip that performs matrix multiplication.
The costs appear before production
ASIC design requires architecture, verification, physical design, software, packaging, testing, and manufacturing commitments. A company must pay those costs before it knows how many units it will ship.
The minimum efficient volume can be high. If a custom chip saves $1,000 per accelerator but costs $300 million to develop and qualify, the business needs more than 300,000 units before ignoring financing, support, and inventory risk. If the workload changes before that volume is reached, the savings may disappear.
Advanced packaging and high-bandwidth memory can add another constraint. Micron reported in March 2025 that its HBM output for calendar 2025 was sold out and that it was already discussing 2026 demand with customers. A custom ASIC that cannot secure memory, substrates, and packaging is not a complete product.
A practical comparison must use system metrics
Compare the two options at the system level. Measure useful work per rack, tokens per second per watt, cost per million tokens, time to train, memory capacity, software migration effort, and failure-recovery behavior.
Do not compare a peak chip number from a data sheet with a measured production result from another platform. Ask whether the benchmark uses the same model, batch size, sequence length, precision, compiler version, and utilization target.
A GPU may win on time to market even if an ASIC wins on energy. The value of six months of earlier service can be material for a cloud provider. Conversely, an inference fleet with a stable model and high utilization may benefit from a specialized accelerator.
The market is moving toward heterogeneous systems
The likely outcome is not one winner. General-purpose GPUs will remain useful for training, new models, and workloads with uncertain demand. Custom ASICs will take more of the predictable inference, networking, storage, and recommendation work. CPUs, DPUs, and specialized accelerators will handle other parts of the system.
Intel’s Gaudi 3 product materials emphasize standard Ethernet, open software, and a migration path for customers moving from GPU-based systems. That positioning shows the market’s pressure point: customers want acceleration without locking every layer of the stack to one vendor.
The decision rule is simple. Use GPUs when flexibility, software, and deployment speed are worth paying for. Use an ASIC when the workload is stable, volume is large, memory and packaging are secured, and the organization can support a separate software and operations path.
The best analysis treats silicon as a fleet decision, not a chip-shopping exercise. The question is which architecture produces reliable, adaptable work at the lowest total cost over the period in which the workload is likely to remain stable.
There is also a strategic lock-in cost. A GPU fleet may be portable across cloud providers and model frameworks, while a custom accelerator can require a dedicated compiler, kernel library, scheduler, and support team. That lock-in may be acceptable when demand is large and stable. It is dangerous when the business case assumes a model architecture will remain unchanged for five years. The best ASIC programs preserve a fallback path: compatible APIs, exportable model formats, or enough GPU capacity to absorb new workloads while the specialized stack catches up.
Sources
- Micron, “Fiscal Q2 2025 Investor Materials,” March 20, 2025: investors.micron.com
- TSMC, “2025 Annual Report,” 2026: investor.tsmc.com
- Intel, “Intel Gaudi AI Accelerator Products,” accessed October 2026: intel.com
- Intel, “Intel and Google Deepen Collaboration to Advance AI Infrastructure,” April 9, 2026: newsroom.intel.com
- IEA, “Energy and AI,” April 10, 2025: iea.org
BOSSTOX Research articles are drafted with AI assistance and reviewed by our editors before publication. See our editorial standards.
-
Power & Grid1 month agoTransformer Lead Times Are Becoming a Data-Center Schedule Risk
-
Power & Grid1 month agoGas Turbines on Data-Center Campuses
-
Cooling2 months agoWater Use and Dry Coolers: The Tradeoff AI Sites Cannot Avoid
-
Power & Grid1 month agoBattery Storage at AI Sites
-
Compute & Silicon1 month agoGPU Supply Chains: How AI Hardware Gets Made, Moved, and Bottlenecked
-
Power & Grid4 weeks agoFERC’s Large-Load Orders: The November 16 Grid Deadline Behind the AI Build-Out
-
Capital & Backlog4 weeks agoAI Infrastructure’s Rough Week: Chips Fall 6.7% as Meta Lifts Capex Floor
-
Cooling4 weeks agoLiquid-Cooling Leak Detection and Reliability
