Theme

Silicon fractures into vertical empires

Explore this theme’s evidence over time →

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to December 12, 2025 (theinnermostloop.substack.com)

Buyers stop being buyers. Large players bypass the GPU market with custom accelerators or move design in-house entirely, turning one supplier’s monopoly into several vertical stacks.

Development timeline

Newest reporting dates first. Reporting dates may differ from event dates.

2026-10-13: Microsoft plans multi-gigawatt Maia inference deployments

2026-10-05: OpenAI models reportedly reduce Jalapeño die area

2026-09-17: Apple may revive Xserve with NVLink

2026-09-17: SK Hynix considers Intel’s Ohio site

2026-08-27: A lab's own inference chip beats every incumbent part tested

2026-08-21: A state approves up to 7,000 robotaxis for one city

2026-08-21: A chipmaker seeks up to $100 billion for custom AI silicon

2026-08-17: A fab is framed as a bet that pays off however a geopolitical question resolves

2026-08-13: A lab moves to buy a company to squeeze more inference from existing silicon

2026-08-08: A fab will host a particle accelerator as a shared lithography light source

2026-08-06: A fab breaks ground aiming at a terawatt a year of AI compute

2026-08-06: A chipmaker buys a startup that etches model weights into silicon

2026-08-05: A buyer commits to one chip vendor exclusively, inverting diversification

2026-07-30: A chipmaker's operating profit rises nineteen-fold on AI memory

2026-07-26: A lab asks a memory maker to supply its own chip effort

2026-07-22: High-NA EUV scanners enter high-volume manufacturing

2026-07-20: A lab freezes its model's blueprint directly into a chip

2026-07-14: A foundry's revenue jumps 68% on sold-out leading-edge capacity

2026-07-08: A chip division's year beats everything it earned across four decades

2026-07-07: Chinese firms plan to route 46% of accelerator budgets to domestic chips

2026-07-05: A startup reorganizes itself as a fab that builds fabs

2026-07-04: A challenger GPU serves an open model at half the cost as agents write kernels

2026-07-04: A lab explores custom silicon as buying compute becomes becoming compute

2026-06-28: A company pledges $9B over five years toward a 2029 quantum machine

2026-06-26: A sub-one-nanometer node crams 100 billion transistors onto a fingernail

2026-06-25: A lab tapes out its first inference chip in nine months

2026-06-22: A memory maker overtakes its parent rival for the first time since 2000

2026-06-17: A rocket company buys an IDE maker for $60B

2026-06-14: A datacenter is built from two thousand retired phones

2026-06-12: A lab signs a dozen letters of intent to lease its own datacenters

2026-06-12: A memory-chip maker overtakes an automaker as Japan's most valuable firm

2026-06-08: A hyperscaler orders three million TPUs from a rival foundry

2026-06-02: One architecture now spans rack, desk, laptop and body

2026-05-28: An ARM server chip outscores the best x86 parts

2026-05-26: A qubit is patterned with production EUV lithography

2026-05-25: A chipmaker announces a scaling law to replace Moore's Law

2026-05-20: A $25B joint venture targets half a gigawatt of custom silicon

2026-05-18: The President says the government should have asked for more of Intel

2026-05-15: The first mixture-of-experts model is trained end to end off Nvidia

2026-05-11: A rival software stack improves 75-fold in fourteen days

2026-05-09: A 245-terabyte drive ships as a quantum firm files to list

2026-05-08: Nvidia puts $3.2B into optical fiber as copper runs out of bandwidth

2026-05-06: Apple explores Intel and Samsung as US fabs beyond TSMC

2026-04-23: A memory maker's revenue nearly triples in a year

2026-04-20: Intel erases every dollar lost in the dot-com crash after 26 years

2026-04-17: A chipmaker files to list with warrants that scale with its customer's spend

2026-04-16: A memristor runs reliably hotter than the surface of Venus

2026-04-12: A small quantum computer shows broad advantage for machine learning

2026-04-09: The first model class trained at scale on Blackwell arrives mid-handoff

2026-04-08: A network fast-tracks its whole platform to post-quantum security

2026-04-05: Tesla's research fab puts logic, memory, packaging and masks in one building

2026-04-02: Chinese accelerators take 41% of their home market

2026-04-02: Shor's algorithm is shown viable at ten thousand qubits

2026-04-01: Breaking elliptic curve crypto may need 20x fewer qubits than thought

2026-04-01: An autonomous warship startup raises $1.75B

2026-03-29: A quantum processor matches neutron scattering on fifty qubits

2026-03-28: A quantum-resistant deadline moves forward six years

2026-03-25: Arm abandons neutrality to build an AGI CPU

2026-03-24: A memory maker orders $7.9B of lithography tools

2026-03-23: Leading-edge capacity is booked out through 2028

2026-03-22: A fab is announced for a terawatt a year, 80% of it for space

2026-03-17: A superconducting optoelectronic network claims a millionfold speedup

2026-03-17: Nvidia projects a trillion dollars of AI processor sales through 2027

2026-03-17: Nvidia builds a module for orbital datacenters

2026-03-13: Data centers are built from lab-grown human neurons

2026-03-12: Intel accelerates fully homomorphic encryption five thousandfold

2026-03-10: A quarter of iPhones are now made in India

2026-03-08: Laser-heated storage ships at scale toward 100-TB drives

2026-03-04: Apple reorganizes its pro silicon around running models locally

2026-03-04: Peking University reaches a 1-nm ferroelectric transistor at 0.6 volts

2026-03-03: Nvidia commits $4B to optical interconnects

2026-03-02: In-vacuum cryoelectronics unlock trapped-ion scaling

2026-03-02: The first 2-nm phone chip arrives with 6G commitments

2026-02-28: Nvidia prepares an inference processor with Groq silicon inside

2026-02-27: Meta rents Google TPUs to diversify away from Nvidia

2026-02-24: ASML raises EUV source power to a thousand watts

2026-02-24: Meta contracts up to 6 GW of AMD GPUs

2026-02-23: A company bakes any model into custom silicon in two months

2026-02-18: New silicon cuts cost per token 35-fold against Hopper

2026-02-17: Heat is made to flow from cold to hot

2026-02-17: SpaceX and its own subsidiary compete for a drone swarm prize

2026-02-16: SpaceX hires crystal growers for orbital wafer fabs

2026-02-08: Carmack and Musk brainstorm replacing DRAM with fiber loops

2026-02-03: SpaceX acquires xAI to build a sentient sun

2026-01-31: Models distill to 4-bit with almost no loss

2026-01-30: Musk weighs merging SpaceX with xAI or Tesla for orbital datacenters

2026-01-29: Tesla names chips its bottleneck and plans a terafab

2026-01-27: Microsoft ships Maia 200 as Nvidia and CoreWeave add 5 GW

2026-01-27: An optical unit claims ten times Rubin's throughput

2026-01-26: Stretchable transistors are built from 2D flakes

2026-01-24: ASML passes a $500B valuation as Delft builds a 6-qubit silicon circuit

2026-01-15: OpenAI adds 750 MW of low-latency compute with Cerebras

2026-01-14: Specialty silicon raises against the GPU monopoly

2026-01-11: Nanoimprint lithography reaches 1.4-nm logic

2026-01-11: Sandia maps the motor cortex onto neuromorphic silicon

2026-01-10: D-Wave buys Quantum Circuits for $550M

2026-01-08: A robocar ships with 8,000 TOPS

2026-01-06: AMD previews a 2-nm MI500 and ships edge silicon

2026-01-06: Qualcomm puts Dragonwing silicon inside humanoids

2026-01-06: Vera Rubin enters production at 50 petaflops

2026-01-05: A Korean inference chip enters mass production

2026-01-03: Anthropic buys a million TPUs and skips the cloud

2026-01-01: A photonic reservoir device is ten times more efficient per operation

2026-01-01: TSMC pulls in its 1.4-nm fab on better yields

2025-12-31: China fuses flash storage directly to GPUs

2025-12-30: Nvidia buys a $5B stake in Intel

2025-12-30: TSMC starts volume production of gate-all-around N2

2025-12-29: Groq's LPUs are slated for Nvidia's 2028 GPUs

2025-12-29: SK Hynix weighs the first US 2.5-D packaging line

2025-12-28: GPU links hit the copper cliff, and terahertz radio answers

2025-12-27: Intel shows cellphone-sized multi-chiplet packages with HBM5

2025-12-27: NVIDIA and SK Hynix push GPUs to talk directly to storage

2025-12-26: NVIDIA is set to displace Apple as TSMC's largest customer

2025-12-25: NVIDIA buys Groq for $20B, its largest acquisition

2025-12-20: LightGen runs generative tasks entirely in light

2025-12-18: Reprogrammable 2D waveguides run neural inference in light

2025-12-17: Amazon discusses $10B into OpenAI tied to Trainium adoption

2025-12-15: SkyWater fabricates a 3D carbon nanotube chip domestically

2025-12-13: The US launches a Pax Silica Initiative

2025-12-12: Anthropic orders $10B of Google TPUs

2025-12-12: Rivian drops NVIDIA for its own autonomy chip

Knowledge graphopen in browser →

Referenced by

The substrate becomes a design target of the thing it runs

Cite or correct this record

Citation formatCitation text

Silicon fractures into vertical empires
Innermost Loop KB — secondary record
Reporting date: unknown
KB permalink: https://nicholsn.github.io/innermost-loop-kb/themes/vertical-silicon
Attributed sources:
- Welcome to December 12, 2025 — https://theinnermostloop.substack.com/p/welcome-to-december-12-2025
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.