Robots, Chips & WattsRobots, Chips & Watts
RCW-0050 · The Daily · Field record 02 NOV 2026 · 08:00 ET

When the Network and the Grid Set the Terms

A $115 billion optical-connectivity deal, new AI infrastructure financing and grid-access rules reveal where physical AI's next constraints sit.

Record typeDaily brief
Pre-market
Systemsoptical interconnect and advanced semiconductors
AI infrastructure finance
grid interconnection and power flexibility
inference accelerators and HBM
industrial and energy AI deployment
autonomous systems and manufacturing integration
critical minerals and supply-chain security
RegionsUnited States · South Korea · United Arab Emirates · Japan · Malaysia · India · Australia · China · Taiwan · Europe · Southeast Asia
Source count18 attached
Mixed, primary-led: company and grid-operator releases, joint government statement, Reuters original reporting and survey-based secondary reporting; unconfirmed claims labeled.

00Opening read

The next AI infrastructure contest may be decided by everything surrounding the processor.

Marvell's plan to buy Lumentum puts an extraordinary price on the optical equipment that carries data between chips and across data centers. At the same time, Oracle and Alphabet are seeking more financing, while California is rewriting how large electricity users might reach the grid. The common issue is delivery. A chip needs bandwidth, electricity and an operator who can afford to run it.

There is another story beyond the data center. South Korea wants to spend more on chips and physical AI. A Korean inference-chip company says it is scaling production. Toyota is preparing to integrate autonomous-driving hardware directly into vehicles at its Japanese factory. The work is moving toward systems that operate reliably.

01What moved

Optical connectivity just became a much bigger corporate bet. Marvell and Lumentum announced a proposed $115 billion combination. Marvell brings custom chips and networking; Lumentum brings lasers and optical components. They want to improve the links that keep AI systems moving data. The deal needs approval and is not expected to close until 2028. Manufacturing and delivery will be the test.

The AI factory is returning to the credit market. Oracle is seeking fresh funding for its infrastructure expansion, and Alphabet launched a euro bond sale. Both matter because spending precedes much of the promised revenue. Investors want to know whether announced projects will produce cash. A contract is not a profitable, energized building.

California is testing a different answer to the grid queue. Its grid operator has approved a proposal that could let some large electricity customers connect sooner if they agree to operate flexibly while network upgrades are completed. That could accelerate construction, but operators must manage possible power restrictions. Federal review remains.

Korea is backing the whole industrial chain. The government is seeking approval for a larger AI and semiconductor budget, including physical AI and data centers. Separately, FuriosaAI announced a substantial funding round to expand production of its inference chips. The test is installed systems with memory, working software and paying customers.

AI agents are entering industrial planning. ADNOC, OpenAI and G42 announced an agreement to connect operational and commercial decisions across the energy company. Oilfield planning must respect equipment, safety and engineering rules. Measured results will determine its significance.

Autonomy is moving into the assembly process. Toyota and Waymo plan to test a Japanese-built autonomous vehicle fleet in 2027. Rather than add the driving hardware after the car leaves the factory, Toyota intends to integrate it during production. Repeatable factory integration could matter more than another demonstration drive.

02The throughline

Across these developments, the bottleneck keeps moving. First it was the accelerator. Then memory, cooling and electricity. Now the connections between machines, the financing behind facilities and the rules governing access to the grid are taking a larger share of attention.

That does not mean demand for AI hardware has disappeared. It means the buildout is entering a harder stage. Each layer must work with the others, at viable cost. A well-funded chipmaker still needs fabrication slots and customers. A new data center still needs reliable power. An autonomous system still needs repeatable manufacturing and safe operation.

The useful measure is becoming less about what has been announced and more about what can be delivered, qualified and kept running.

03What to watch

  • California ISO's planned November 16 federal filing, especially the operating conditions for faster grid access.
  • The final financing terms for Oracle and Alphabet, and whether future infrastructure spending produces stronger cash flow.
  • Production shipments and customer deployments from FuriosaAI, alongside Korea's budget approvals.
  • Toyota and Waymo's 2027 testing milestones, including factory integration and safety validation.

04Source trail

05Signal references

This record is cited as evidence by the following active signals.

SIG-R118OBSERVE
SIG-M142STRENGTHENS
SIG-W087STRENGTHENS
SIG-C204OBSERVE
SIG-A052STRENGTHENS
SIG-F031STRENGTHENS