Micron reported record fiscal Q4 revenue of $54 billion, up from $11 billion a year earlier, and told investors that memory supply will remain tight through 2028, with more than 75% of its 2027 output already committed to customers.

The company also signaled that 2027 HBM pricing will come in significantly higher than 2026 levels — the clearest confirmation yet that the AI-driven memory cycle still has years to run.

Micron's numbers anchor a week in which the entire manufacturing ecosystem posted record results. Revenue across foundries, IDMs, packaging and test hit $96.6 billion in Q2 2026, up 25% year over year, according to Counterpoint.

South Korea added its own datapoint: semiconductor exports surged more than 262% year over year to a record $60 billion in September, driven by stronger memory demand, higher shipment volumes and rising prices, according to the Ministry of Trade.

How tight does memory get?

HBM will consume nearly 30% of DRAM manufacturers' wafer capacity in 2027, up from about 20% today, a Samsung executive told Reuters. That shift crowds conventional DRAM supply just as demand accelerates.

TrendForce expects conventional DRAM contract prices to rise 10% to 15% in Q4 2026, with NAND flash up 15% to 20%. Enterprise SSD bit demand is projected to grow more than 80% in 2026.

On the test side, Teradyne introduced its Magnum E2 high-speed memory test system, supporting LPDDR6, DDR6, GDDR7 and NAND for AI and HPC workloads.

SK hynix's U.S. subsidiary Solidigm is weighing an IPO that could value the unit at up to $150 billion, Reuters reported. SK hynix said nothing has been decided.

Where is capacity being built?

TSMC is evaluating a multibillion-dollar manufacturing investment in Texas that would expand its U.S. footprint beyond the $265 billion Arizona buildout, Bloomberg reported.

Advanced packaging and substrates remain the hottest capex targets:

• Samsung Electro-Mechanics will invest about $4.9 billion to expand FCBGA package-substrate capacity in South Korea and Vietnam for AI servers.
• TOPPAN opened its first overseas FC-BGA substrate plant in Singapore, targeting large, high-layer-count substrates for AI processors and networking gear.
• Infineon opened the first phase of a backend hub in Bangkok, adding assembly, test and wafer-test capacity.
• VSMC, the Vanguard-NXP joint venture, opened its first 300mm fab in Singapore and is reportedly considering a second, per Bloomberg.

The power side of the AI buildout is scaling too. South Korea and the U.S. are advancing a $22 billion, 6.47-GW gas-fired power project in Texas — part of a broader investment package of up to $350 billion — to feed semiconductor plants and AI data centers. California, meanwhile, exempted photoresist solvent mixtures containing 70% or less gamma-butyrolactone from state permitting, easing a regulatory burden on fabs.

What is changing in design technology?

Lam Research proposed a self-aligned CFET architecture that could extend CMOS scaling below 10 angstroms. Its digital-twin modeling produced a 0.0094 μm² 6T SRAM bit cell — 52.7% smaller than N3 — while holding at least a 10 nm overlay window across critical contact layers.

In EDA, Synopsys rolled out a portfolio of domain-specific long-horizon agents that reason, plan and execute multi-step workflows across verification, implementation, analog, manufacturing and simulation. It also signed a multi-year agreement with OpenAI to develop GPT-Synopsys, a specialized model for semiconductor design workflows using its EDA tools.

Amazon broadened its Synopsys relationship to include application-optimized silicon IP for custom chips and AWS infrastructure.

What are the big deals?

AMD plans to acquire World Labs, which builds spatial-intelligence AI models simulating interactive 3D environments for robotics, in an $8.2 billion all-stock deal expected to close by the end of 2026.

Anthropic's IPO filing revealed it is the customer behind Broadcom's previously disclosed AI compute-financing arrangement worth up to $42 billion, under which Broadcom lends money to lease its chips, Reuters reported.

Samsung will invest $1 billion in Helix Digital Infrastructure, spanning hyperscale data centers, power generation and fiber networks. Applied Materials added Kioxia and Besi as partners at its $5 billion EPIC Center for next-gen memory and advanced packaging work.

Funding flowed toward physical AI and photonics: SiMa.ai raised $150 million (reaching $500 million total and a $1.45 billion valuation) for silicon exceeding 1,000 TOPS; optical interconnect startups CScale ($145 million) and Volantis ($88 million) both exited stealth or raised rounds aimed at AI scale-up and memory pooling.

Why does photonics suddenly have a standard?

JEDEC published the industry's first silicon photonics reliability standard, establishing common qualification requirements as optical interconnects scale for AI systems. Counterpoint says network bottlenecks now bind accelerator performance harder — GPUs spend roughly 12.5% of their time waiting on the network in some workloads — and expects near-packaged optics to sample in late 2026 or early 2027, followed by co-packaged optics.

Bain argues the financial picture has flipped: semiconductor and hardware stocks compounded at 24% annually from 2020 to 2026 versus 6% for software, reversing software's long-standing advantage.

What about security?

VUSec and SSSA researchers uncovered Branch Target Reuse, a new Spectre-v2 attack against just-in-time compilers. Their end-to-end Linux exploit leaked arbitrary memory — including a root password hash — with existing Spectre defenses enabled. Hardware vendors said current mitigation mechanisms can address BTR via software.

Nvidia launched its Open Agent Safety Platform, whose Sentry reference design runs on BlueField-4 DPUs to monitor and quarantine AI agents in silicon; Cadence is integrating it with its AI Super Agents for autonomous chip-design workflows. Fraunhofer, meanwhile, made its OpenTitan-based RISC-V Secure Element available as licensable IP after silicon validation.

The NSA set new post-quantum cryptography requirements for National Security Systems: quantum-resistant support in new commercial systems from 2027, with incompatible legacy systems phased out by 2030.

Who builds all of this?

SEMI says U.S. semiconductor firms could need 88,000 new engineers by 2029, with shortages now concentrated in advanced packaging and experienced manufacturing leadership. Nearly 30% of Europe's semiconductor workforce is expected to retire by 2030.

With more than three-quarters of Micron's 2027 output already committed and HBM eating a growing share of DRAM wafers, pricing power looks set to stay with memory makers — and with equipment and substrate suppliers — well into 2028.