MediaTek Targets $2.5 Billion in Quarterly AI Chip Sales, With $16 Billion Goal for 2027

September shipments open a bigger chapter in cloud computing

MediaTek expects its first custom artificial intelligence accelerator to generate US$2.5 billion in revenue in the fourth quarter of 2026, following initial shipments in September. Chief Executive Officer Rick Tsai set out the quarterly target during the company's October 30 earnings call, alongside a much larger ambition: approximately US$16 billion from the same program in 2027.

The figures in MediaTek's earnings call remarks put a specific timetable on its expansion into chips for data centers. The company is moving from early deliveries to a business large enough to reshape its sales and profitability, while its mobile phone division faces weaker demand.

An ASIC, short for application specific integrated circuit, is a chip designed for a particular job. MediaTek's AI accelerators are custom chips built for customers' computing infrastructure, where they help perform the enormous numbers of calculations needed by AI systems.

Tsai, MediaTek's chief executive, described the initial shipment schedule and quarterly sales expectation in his prepared remarks:

We have already commenced initial shipments in September and are confident of delivering 2.5 billion US dollars revenue in the fourth quarter.

MediaTek has also secured a second cloud ASIC customer outside China, with joint design work underway and revenue expected in 2028, Chief Financial Officer David Ku confirmed. Together, the production ramp, another customer and a schedule extending to 2029 give the company several routes to growth, each dependent on completing complex designs and securing enough manufacturing capacity.

How large could the accelerator business become?

MediaTek expects its first accelerator program to reach the upper end of its revenue target range in 2027, approximately US$16 billion. Tsai said that would represent around 20% of the AI accelerator ASIC market.

Those two figures imply a market of roughly US$80 billion under the company's assumptions. The comparison gives readers a sense of the scale MediaTek is pursuing: a substantial share of the market for custom AI accelerators, a specific segment within the wider AI chip industry.

The US$16 billion annual target is also 60% above the US$10 billion that would result from maintaining the expected fourth quarter sales pace of US$2.5 billion for four quarters. MediaTek therefore anticipates further expansion after the first large quarter of deliveries.

Tsai linked that confidence to the supply outlook, saying the company was working with manufacturing partners to obtain additional capacity. Strong orders must be matched by production slots, packaging capacity and components before they become shipped products and recorded revenue.

Fourth quarter growth follows stronger earnings

MediaTek forecast total fourth quarter revenue of NT$210.1 billion to NT$221.8 billion, using an exchange rate of NT$31.7 to US$1. That would mean growth of 25% to 32% from the third quarter and 40% to 48% from a year earlier.

The company entered that period with third quarter revenue of NT$168.1 billion, up 10.4% from the previous quarter and 18.3% from the same period of 2025. Revenue exceeded the earlier guidance ceiling of NT$159.8 billion by about 5.2%, helped by stronger customer demand and improving supply.

Ku reported third quarter operating income of NT$27.1 billion and net income of NT$28.2 billion. Earnings per share reached NT$17.50, compared with NT$15.28 in the previous quarter and NT$15.84 a year earlier.

At the exchange rate used for fourth quarter guidance, US$2.5 billion equals NT$79.25 billion, or approximately 36% to 38% of projected total quarterly revenue. Management also described cloud ASICs as accounting for about 30% of fourth quarter sales; the basis for that lower share was not explained.

MediaTek now expects its full year revenue growth in US dollars to exceed its previous target in the high single digits. Seasonal weakness in consumer electronics is expected to offset some of the accelerator business's growth.

Why lower gross margins can still mean better profits

MediaTek's data center business carries a lower gross margin than the company average, yet management expects cloud ASICs to improve the company's operating margin beginning in the fourth quarter. Tsai set a target for the full year operating margin to exceed 20% in 2027.

Gross margin measures the share of revenue left after the direct cost of making and supplying products. Operating margin goes a step further, subtracting expenses such as research, development and administration. A business can have a lower gross margin while still contributing a higher operating margin, depending on the operating expenses needed to support its sales.

MediaTek's third quarter gross margin was 46.3%, while its operating margin was 16.1%. A full year operating margin above 20% would place the 2027 target more than 3.9 percentage points above that quarterly benchmark.

The company's published fourth quarter gross margin guidance is 45%, plus or minus 1.5 percentage points, meaning 43.5% to 46.5%. Management also described an expected range of 44% to 46%. Its published operating expense ratio forecast is 26%, plus or minus two percentage points.

Ku said cloud ASIC operating margins would contribute positively to the company average from the fourth quarter. The business is therefore expected to change both the size of MediaTek's revenue and how much operating profit the company keeps from each dollar of sales.

A development schedule stretching into 2029

MediaTek's second accelerator project is scheduled for volume production in the first quarter of 2028. Several dies, the individual pieces of silicon that can make up a complex chip, had already completed tape out, with a more complicated die scheduled to follow a few weeks after the October call.

Tape out is the stage when a chip design is sent for manufacturing. Engineers still need to check that the manufactured silicon works and that production can deliver sufficient numbers of usable chips.

Management's schedule brings together several milestones:

• September 2026: Initial shipments of the first AI accelerator began.
• Fourth quarter 2026: Revenue from that program is expected to reach US$2.5 billion.
• 2027: The first program targets approximately US$16 billion in annual revenue.
• First quarter 2028: The second accelerator project is scheduled to enter volume production.
• 2028: Revenue is expected to begin from a second cloud ASIC customer outside China.
• Late 2029: A third generation accelerator being planned with key customers is expected to enter mass production.

The second accelerator generation and the second customer are separate developments within the company's expansion plans. MediaTek has not publicly named the new customer. Tsai expects another year of strong growth in 2028, while some supply arrangements are still being finalized.

Efficiency becomes a selling point

MediaTek says its second accelerator project is designed to deliver several times more performance per watt and a much lower total cost of ownership. Performance per watt measures how much computing work a chip performs for the electricity it consumes.

Total cost of ownership, often shortened to TCO, includes the expense of buying and running a system over time. For an AI data center, that can include electricity, cooling and the equipment surrounding the processors. A more efficient chip can help a customer perform more calculations within a fixed power budget.

MediaTek has not given a numerical efficiency multiple or a detailed cost comparison for the new accelerator. The improvements remain design targets as the project moves toward production.

Citi maintained a forecast of US$40 billion in MediaTek AI ASIC revenue for 2028, reflecting expectations for rapid expansion. That is the bank's projection; management's stated US$16 billion target concerns the first accelerator program in 2027. The different years and business scope matter when comparing the figures.

Faster connections and more complex packaging

MediaTek's technical investment extends beyond the calculations performed inside a processor. The company is developing the connections that move information between chips, the packaging that brings components together and the systems that allow many processors to work as a group.

Its work includes 448G SerDes, 3.5D advanced packaging and co packaged optics. SerDes circuits convert data into fast streams for transmission and then reconstruct it at the receiving end. MediaTek says its 448Gbps SerDes intellectual property has completed tape out and verification, following production of an earlier 224G technology.

Advanced packaging places multiple pieces of silicon and memory close together so they can exchange data quickly. MediaTek also cited integration capabilities for HBM5, a generation of high bandwidth memory, while co packaged optics brings optical connections close to computing components.

Packaging remains a production concern. Citi assessed risks involving yields and manufacturing cycles for EMIB-T packaging as being within its expectations. Yield means the share of manufactured components that function properly. Lower yields or longer production cycles can limit the number of finished products available for delivery.

Possible use of CoWoS or CoPoS packaging for a later accelerator has also been discussed in market forecasts. MediaTek's broader commitment is to invest in advanced packaging, engineering talent and industry partnerships as its designs become more demanding.

NVIDIA and Alphabet help fund the expansion

MediaTek completed a US$3.9 billion overseas convertible bond offering during the third quarter, with NVIDIA and Alphabet participating. Convertible bonds are debt that can be exchanged for shares under agreed conditions.

Tsai said the funding strengthened relationships with major AI infrastructure partners and increased MediaTek's ability to secure capacity for critical components. Access to capital and access to production are closely connected when suppliers must reserve scarce capacity ahead of large product launches.

MediaTek is also expanding from individual chips toward complete systems and racks. Through NVIDIA's NVLink Fusion program, it aims to connect custom processors to the NVLink ecosystem. The company also says it supports UALink, giving customers another approach to connecting AI computing hardware.

Tsai said several prospective customers had responded positively to the NVLink Fusion business model. MediaTek expects these relationships to reduce the work customers need to do before their equipment can start operating and generating revenue.

TSMC remains MediaTek's most important supplier partner, according to Tsai, although he said the company does not depend on a single supplier. That approach accompanies its effort to coordinate manufacturing, packaging and other components across a wider supply network.

Cloud growth arrives as phone sales weaken

MediaTek's existing businesses show why the new accelerator revenue matters. Mobile phone chips accounted for 37% of third quarter revenue, with sales down 1% from the previous quarter and 17% from a year earlier.

Smart Edge Platforms accounted for 58% of revenue and grew 20% from the previous quarter and 63% from a year earlier. New projects entering production included a 5G modem for a new smartphone customer, AI tablets, AI PCs and automotive cockpit systems. Power management chips contributed the remaining 5%, with growth helped by gains in data center business.

MediaTek's cooperation with NVIDIA also reaches consumer computing. The company said RTX Spark AI PCs using the N1X chip, which combines NVIDIA graphics technology with MediaTek's system on a chip, became commercially available through major PC manufacturers in mid October. Development of a second generation chip is underway.

Samsung's Galaxy Tab S12 series uses MediaTek's Dimensity 9500, marking a third consecutive generation of Samsung AI tablets powered by MediaTek flagship chips. These device projects give the company growth opportunities alongside the much larger orders it is pursuing in cloud infrastructure.

Demand and financing carry their own risks

Tsai expects demand for AI computing to keep growing over the next one to three years, despite concerns that cloud providers could slow capital spending. He argued that high performance chips developed closely with major customers give MediaTek a stronger position during market volatility.

Customers still have to finance their equipment purchases and earn enough from AI services to support continued investment. MediaTek may consider financing support for suitable emerging developers of frontier AI models, the advanced models at the leading edge of capability, to help them reach stable cash flow.

Such support would add another financial relationship alongside chip design and supply. MediaTek has not announced recipients, amounts or terms. For now, its next measurable milestones are the fourth quarter shipment ramp, progress toward the 2027 revenue target and completion of designs scheduled for production in early 2028.

Key Points

• MediaTek expects US$2.5 billion in fourth quarter 2026 revenue from its first AI accelerator, following initial September shipments.
• The first accelerator program targets approximately US$16 billion in 2027, which the company estimates would equal around 20% of the AI accelerator ASIC market.
• Total fourth quarter revenue is forecast at NT$210.1 billion to NT$221.8 billion, up 25% to 32% from the third quarter.
• Cloud ASICs are expected to improve operating margins from the fourth quarter, with the company targeting a full year operating margin above 20% in 2027.
• A second accelerator project is scheduled for volume production in the first quarter of 2028. A second customer outside China is expected to begin contributing revenue in 2028.
• NVIDIA and Alphabet participated in MediaTek's US$3.9 billion convertible bond offering as the company works to secure production capacity.