For more than a decade, the commercial rise of silicon carbide has been closely connected with one major industry: electric vehicles.

Higher-voltage traction inverters, onboard chargers and fast-charging systems created an ideal environment for SiC power devices. As electric vehicles moved toward higher efficiency and higher system voltage, silicon carbide moved from a specialized semiconductor material into a major power-electronics platform.

Now another enormous market is emerging.

It does not have wheels.

It sits inside buildings filled with processors.

Artificial intelligence is turning data centers into some of the world's fastest-growing electrical loads.

Global data-center electricity consumption was about 415 TWh in 2024, according to the International Energy Agency. Under its base-case projection, consumption could reach approximately 945 TWh by 2030. Electricity consumption from accelerated servers—largely driven by AI—is projected to grow much faster than conventional server electricity use.

The United States illustrates the scale of the change. A 2026 national laboratory update estimates that data centers could consume around 11.8% of U.S. electricity by 2030 in its reference case, with modeled scenarios ranging from 9.5% to 15.3%.

This explosion in electricity demand creates an important question for the power semiconductor industry:

Could AI infrastructure become the second major growth market for silicon carbide after electric vehicles?

The answer may depend less on AI processors themselves and more on everything required to deliver electricity to them.

AI Is Not Only a Computing Revolution—It Is Becoming a Power-Electronics Revolution

When people think about AI infrastructure, attention usually goes to processors, memory, networking and cooling.

But every AI accelerator ultimately depends on a much larger electrical system.

Electricity may travel through several stages before it reaches the processor:

Utility Grid → Transformer → AC/DC Conversion → UPS or Energy Storage → High-Voltage DC Distribution → Rack Power Conversion → Server → AI Accelerator

Every conversion stage introduces electrical losses.

At small scale, a fraction of a percentage point may appear insignificant.

At the scale of a large AI campus, it becomes economically important.

As AI deployment expands, data centers are simultaneously dealing with:

• higher facility power;
• higher rack power density;
• higher cooling loads;
• larger backup-power systems;
• more demanding grid connections;
• and stronger pressure to improve energy efficiency.

The IEA notes that the rise of AI is accelerating deployment of high-performance servers and increasing data-center power density. Accelerated servers account for almost half of the projected net increase in global data-center electricity consumption through 2030 in its base case.

This is exactly the type of environment in which efficient power electronics become strategically important.

And this is where silicon carbide enters the story.

Why Silicon Carbide Fits High-Power AI Infrastructure

Silicon carbide belongs to the family of wide-bandgap semiconductor materials.

Compared with conventional silicon, SiC power devices can support demanding power-conversion conditions involving high voltage, high switching performance and high power density.

The U.S. Department of Energy specifically identifies wide-bandgap technologies such as SiC and GaN as enabling more powerful and energy-efficient power-electronic systems, including applications in data centers, industrial equipment, electric vehicles and other electrified systems.

One important benefit is the ability of wide-bandgap devices to support higher switching frequencies.

Higher switching frequency can reduce the required size of inductors, capacitors and other passive components, potentially enabling smaller and more power-dense converter systems.

For AI infrastructure, that matters for three reasons:

efficiency, space and heat.

A more efficient converter wastes less electricity.

A more compact converter uses less physical space.

And lower conversion losses mean less heat must be removed.

At AI-data-center scale, all three matter.

The Real SiC Opportunity Is Not Inside the AI Chip

It is important to clarify one point.

The rise of AI does not necessarily mean SiC will suddenly become the material used to manufacture AI processors.

That is not the primary opportunity.

The much larger opportunity is in the power infrastructure surrounding those processors.

Think of the AI accelerator as the final destination.

Silicon carbide may participate in the electrical highway that supplies it.

Potential applications include:

• high-power AC/DC conversion;
• power-factor-correction stages;
• UPS systems;
• battery backup systems;
• high-voltage DC conversion;
• energy-storage interfaces;
• medium-voltage conversion;
• cooling-system motor drives;
• grid-connected power systems;
• and future solid-state power-distribution architectures.

This distinction is important.

The SiC opportunity created by AI is not primarily a computing-semiconductor opportunity.

It is a power-semiconductor opportunity.

800V DC Could Become an Important Turning Point

Perhaps the most interesting development is occurring in data-center power distribution.

As rack power rises, distributing enormous amounts of power at relatively low voltage becomes increasingly difficult.

The reason comes from basic electrical relationships.

Power is approximately:

P = V × I

For the same power level, increasing voltage reduces current.

And conductor losses increase approximately with the square of current:

Power loss ∝ I²R

Lower current can therefore help reduce resistive losses and reduce the amount of copper required for power distribution.

This is one reason the AI infrastructure industry is exploring much higher DC distribution voltages.

At a major open-computing industry event in July 2026, discussions specifically identified 800V high-voltage DC distribution and near-load power delivery as important technologies for next-generation AI data centers. The same event discussed the evolution toward 800V DC systems for megawatt-class AI racks.

This development should attract attention from the silicon carbide industry.

Why?

Because the power-system voltage is moving into territory where SiC becomes increasingly attractive.

AI Data Centers May Repeat Part of the EV Power-Architecture Story

There is an interesting parallel between automobiles and AI infrastructure.

Electric vehicles began with relatively lower-voltage architectures.

As manufacturers wanted:

• faster charging;
• higher power;
• lower current;
• improved efficiency;
• and lighter electrical systems,

higher-voltage platforms became increasingly attractive.

That transition helped increase demand for silicon carbide.

AI data centers may now face a conceptually similar problem, although at a completely different scale.

As rack power rises, data-center designers also want:

• higher distribution voltage;
• lower current;
• reduced electrical losses;
• higher power density;
• and more efficient conversion.

The similarity does not mean AI data centers will use exactly the same power electronics as electric vehicles.

They will not.

But the underlying engineering pressure is familiar:

when power increases dramatically, efficiency becomes more valuable.

That is the environment in which SiC has historically gained market share.

One Megawatt Rack Changes the Economics

Traditional server racks were once measured in relatively modest kilowatt ranges.

AI infrastructure is pushing the industry toward far more extreme power densities.

Industry discussions in 2026 are already addressing MW-class AI racks and even GW-scale AI data-center campuses, where power delivery and cooling must be designed together with computing hardware rather than added afterward.

That changes the economics of semiconductor selection.

Suppose a converter operates continuously at very high power.

A small improvement in efficiency may save only a tiny fraction of the total energy passing through the system.

But when the system runs:

24 hours per day × 365 days per year × many years

that small percentage can become economically significant.

And every watt that is not lost in the converter is also a watt that does not become unwanted heat.

At large scale, semiconductor efficiency therefore influences:

electricity cost + cooling load + equipment size + infrastructure capacity.

This is why a more expensive power device can sometimes result in a cheaper overall system.

The same system-level economics that helped SiC gain traction in electric vehicles could become increasingly important in AI infrastructure.

AI Power Demand Is Concentrated—And That Creates Another Problem

Global data-center electricity consumption is still relatively modest compared with total global electricity use.

But data-center demand has one unusual characteristic:

it is highly concentrated geographically.

The IEA notes that data centers can create particularly challenging grid-integration problems because they tend to cluster in specific locations rather than distribute demand evenly.

A large AI campus can therefore require enormous amounts of power in one location.

This creates additional demand for:

• substations;
• transformers;
• energy storage;
• grid reinforcement;
• power-quality equipment;
• backup power;
• renewable-energy integration;
• and high-capacity power conversion.

That expands the possible SiC opportunity beyond the data-center building.

AI may stimulate SiC demand in the entire electrical ecosystem surrounding computing infrastructure.

The Second Golden Market Could Actually Be Larger Than “Data Centers”

This is where the story becomes more interesting.

The new market should not necessarily be defined as:

SiC for AI servers.

A better definition may be:

SiC for AI energy infrastructure.

Consider a future AI campus containing:

Utility Grid

↓

Renewable Generation

↓

Grid Substation

↓

Battery Energy Storage

↓

High-Power Conversion

↓

800V DC Distribution

↓

AI Computing Racks

Silicon carbide could potentially participate in several different sections of this chain.

The semiconductor does not have to appear inside every server to benefit from AI growth.

If AI causes more electricity to pass through high-voltage converters, inverters, energy-storage systems and power-distribution infrastructure, the total addressable SiC market can grow indirectly.

This is fundamentally different from the electric-vehicle market.

An EV typically contains a relatively defined number of major power-conversion systems.

An AI campus is an entire electrical infrastructure project.

Energy Storage Could Become Closely Linked to AI

AI data centers also require reliable power.

Interruptions can disrupt large computing workloads and create operational problems.

Battery energy storage and backup systems are therefore becoming increasingly important components of data-center design.

This creates another pathway for SiC.

Battery systems require bidirectional power conversion:

Grid / DC Bus → Battery during charging

and

Battery → DC Bus during discharge

High-power storage systems therefore need converters capable of handling large currents and high voltages efficiently.

As AI campuses become larger, battery backup systems could themselves become major power-electronics installations.

The result is a reinforcing cycle:

AI growth → higher electricity demand → more energy storage → more power conversion → more opportunity for high-efficiency power semiconductors.

The Grid May Become an Even Bigger SiC Opportunity

Data centers cannot expand indefinitely without adequate grid infrastructure.

Electricity generation is only one part of the challenge.

Power must also be transmitted, transformed, converted and distributed.

Future grid systems may require greater use of:

• flexible power conversion;
• medium-voltage converters;
• renewable-energy inverters;
• battery-storage converters;
• power-quality equipment;
• DC distribution;
• and potentially solid-state transformer technologies.

Wide-bandgap devices are being investigated for precisely these types of advanced power-electronics architectures. The U.S. Department of Energy's strategic framework notes that WBG devices can enable substantially higher switching frequencies and more power-dense converter designs, although reliability, passive components and system-level engineering remain important challenges.

So the AI-to-SiC connection may extend from:

Grid → Data Center → Rack

rather than simply:

Power Supply → Server

That is a much larger market concept.

What Could This Mean for the SiC Wafer Industry?

For SiC substrate suppliers, the emergence of AI infrastructure is important because it could diversify demand beyond automotive power electronics.

The underlying device requirements may vary, but many high-voltage vertical power devices are built using conductive 4H-SiC substrates followed by epitaxial growth.

If demand for high-power conversion rises significantly, device manufacturers may require larger volumes of consistent SiC material.

That brings wafer economics back into focus.

For large-scale power-device manufacturing, buyers increasingly care about more than whether a wafer can be produced.

They care about:

• defect density;
• resistivity consistency;
• crystal orientation;
• wafer thickness;
• TTV;
• bow;
• warp;
• surface roughness;
• polishing damage;
• particle contamination;
• and lot-to-lot repeatability.

The reason is simple.

AI infrastructure will not reward expensive wafers that produce poor device yield.

If SiC is going to become a large-volume data-center material, manufacturing economics must continue to improve.

200mm SiC Could Become Increasingly Important

This is also why the transition from 150 mm to 200 mm SiC remains important.

A 200 mm wafer has approximately 1.78 times the geometric area of a 150 mm wafer.

In principle, that allows significantly more devices to be processed during each wafer-fabrication cycle.

But the benefit only exists when:

• crystal quality remains acceptable;
• wafer geometry is controlled;
• epitaxial uniformity is sufficient;
• and device yield remains high.

If AI infrastructure creates another high-volume market for SiC devices, it could strengthen the economic incentive to move more manufacturing toward 200 mm platforms.

Higher volume encourages:

larger wafers → greater automation → higher equipment utilization → lower device cost.

That process could eventually help make SiC competitive in even more applications.

Cost Will Determine Whether AI Becomes a True SiC Golden Market

Technical suitability alone does not guarantee mass adoption.

Silicon remains deeply established in power electronics.

Other wide-bandgap technologies also compete in certain power-conversion stages.

Therefore, SiC must justify its cost at the system level.

An AI data-center designer is unlikely to choose SiC simply because it has attractive material properties.

The real calculation is more practical:

Does the additional semiconductor cost reduce total electricity losses, cooling requirements, converter size or infrastructure cost enough to create a better total system?

At small power levels, the answer may sometimes be no.

At extremely high power levels, the equation becomes much more interesting.

This is why AI infrastructure could be particularly important.

AI is pushing power density upward at exactly the time the SiC industry is trying to reduce wafer and device costs.

Those two trends may meet each other.

AI Could Help Reduce the SiC Industry's Dependence on Automotive Demand

A healthy semiconductor market usually benefits from having multiple major end markets.

Heavy dependence on one industry creates risk.

For years, SiC investment has been strongly influenced by electric-vehicle expectations.

But automotive demand can fluctuate with:

• consumer demand;
• vehicle pricing;
• government incentives;
• charging infrastructure;
• inventory cycles;
• and regional economic conditions.

AI infrastructure follows a different demand cycle.

Energy storage follows another.

Power-grid investment follows another.

Renewable energy follows another.

If all of these sectors increasingly use SiC power electronics, the SiC supply chain becomes less dependent on a single application.

This could ultimately be one of the most important consequences of AI growth.

The AI market does not need to “replace” electric vehicles.

It only needs to become a second substantial source of demand.

From “SiC per Vehicle” to “SiC per Megawatt”

The way the industry thinks about silicon carbide demand may eventually change.

Automotive SiC demand is often conceptually understood as:

number of vehicles × semiconductor content per vehicle

AI infrastructure introduces a different model:

installed computing power × electrical infrastructure per megawatt

That is a significant shift.

An AI campus may require power electronics not only for computing racks but also for:

• utility interfaces;
• backup systems;
• energy storage;
• cooling equipment;
• high-voltage distribution;
• renewable-energy systems;
• and grid support.

Instead of asking how many SiC devices exist inside one machine, the industry may increasingly ask:

How much SiC power electronics are required to deliver one megawatt of reliable AI computing capacity?

That could become an important metric for the next generation of SiC market analysis.

The Real Race Is “Grid to AI Chip”

The semiconductor industry has spent years increasing the performance of AI processors.

The next challenge is supplying those processors with enough electricity.

That creates an entirely different technology race:

Grid → Conversion → Storage → Distribution → Rack → Processor

Every stage must become:

• more efficient;
• more compact;
• more reliable;
• and capable of handling higher power.

Silicon carbide will not dominate every stage.

It does not need to.

If it becomes a preferred semiconductor in even several of the highest-voltage and highest-power conversion stages, AI infrastructure could become a significant long-term demand source.

That is why the AI boom deserves attention from the SiC industry.

Conclusion: EVs Created the First SiC Boom. AI Could Help Create the Second.

Electric vehicles demonstrated that silicon carbide could move beyond niche industrial applications and become a high-volume semiconductor technology.

AI may create the next opportunity.

The reason is not simply that AI servers consume electricity.

The real opportunity comes from the scale, concentration and power density of that electricity demand.

Global data-center electricity consumption is projected to roughly double by 2030, with AI-driven accelerated computing responsible for a large share of the increase.

At the same time, the industry is already discussing 800V DC distribution, megawatt-class racks and gigawatt-scale AI infrastructure.

These developments increase the value of:

high voltage + high efficiency + high power density + advanced power conversion.

Those are exactly the conditions in which silicon carbide becomes interesting.

The first golden market for SiC was built around moving vehicles more efficiently.

The second may be built around moving electricity more efficiently.

And if AI continues to expand at its current pace, the most important question for the silicon carbide industry may no longer be:

How many electric vehicles will use SiC?

It may become:

How much of the electricity feeding the world's AI infrastructure will eventually pass through SiC power devices?

That question could define the next chapter of the silicon carbide market.

FAQ

Why could AI data centers become an important market for SiC?

AI infrastructure requires rapidly increasing amounts of electricity and increasingly high power density. SiC is attractive for high-voltage, high-power conversion stages where efficiency, thermal management and power density are important.

Is silicon carbide used to manufacture AI processors?

Generally, that is not the main opportunity discussed here. SiC's major opportunity is in power electronics surrounding AI computing, including AC/DC converters, UPS systems, battery storage, high-voltage DC conversion and grid infrastructure.

Why is 800V DC important for future AI data centers?

Higher distribution voltage allows the same amount of power to be delivered at lower current, helping reduce conductor losses and supporting higher rack power density. Industry discussions in 2026 are already exploring 800V DC architectures for next-generation MW-class AI racks.

Could AI increase demand for 4H-SiC wafers?

Potentially. If AI infrastructure increases demand for high-voltage SiC power devices, it could indirectly increase demand for conductive 4H-SiC substrates and epitaxial wafers used to manufacture those devices.

Will AI replace electric vehicles as the main SiC market?

That is not yet established and should not be assumed. A more realistic scenario is that AI infrastructure becomes an additional major application, reducing the SiC industry's dependence on automotive demand.

What SiC wafer parameters matter for high-power device manufacturing?

A practical RFQ may include wafer diameter, polytype, conductivity type, resistivity, crystal orientation, off-axis angle, thickness, TTV, bow, warp, defect requirements, surface roughness, edge condition, polishing quality and lot-to-lot consistency.