Alphabet’s Google Cloud on April 22 unveiled its eighth-generation tensor processing units (TPUs), splitting the line into two purpose-built chips: the TPU 8t for large‑scale model training and the TPU 8i for low‑latency inference and AI agents. Google said 8t delivers roughly 2.8x the training throughput of the prior generation and can be clustered into 9,600‑chip pods with networking to scale to logical clusters of more than one million chips. TPU 8i prioritises on‑chip memory and latency (including up to 384 MB SRAM and expanded HBM) and offers an estimated ~80% improvement in performance‑per‑dollar for inference. Google also flagged infrastructure updates — new network topologies and storage paths — to support the chips, confirmed continued availability of Nvidia GPUs in its cloud, and highlighted partner design and supply relationships with Broadcom, MediaTek and potential talks with Marvell. The systems and chips are due to be made generally available later this year.

Elon Musk said on April 22-23, 2026, that Tesla plans to use Intel’s next‑generation 14A semiconductor process to make chips for its Terafab project, making Tesla the first major external customer for the technology. Intel has joined Musk’s Terafab initiative alongside SpaceX to produce processors for vehicles, humanoid robots and space data centres. Musk also announced a roughly $3 billion research chip facility at Tesla’s Giga Texas campus to run a pilot line that would process a few thousand wafers a month; SpaceX would lead high‑volume manufacturing. Intel’s shares rose in after‑hours trading on the news. The move is framed as validation of Intel’s foundry efforts after years of yield and supply challenges; company leaders have said securing outside customers is crucial to sustaining their contract manufacturing push. Analysts and investors will watch production timing, yield performance for Intel’s 18A/14A nodes, who will fund and operate large‑scale fabs, and whether the Terafab plan involves licensing Intel technology or using Intel facilities.

Microsoft said on April 23, 2026 it will invest A$25 billion (about US$17.9 billion) in Australia by the end of 2029 to expand Azure AI supercomputing and cloud infrastructure, strengthen cybersecurity and boost AI skills training. The package — the company’s largest-ever commitment in Australia and building on a prior A$5 billion pledge in 2023 — includes plans to increase local Azure AI capacity by more than 140%, extend the Microsoft‑ASD Cyber‑Shield to additional government agencies, and train three million Australians in workforce-ready AI skills by 2028. The announcement, made in Sydney by CEO Satya Nadella alongside Prime Minister Anthony Albanese, is underpinned by a memorandum of understanding with the government. Analysts note the move is part of a broader global hyperscaler build‑out as rivals step up AI capex. Critics and local officials have flagged missing details on data‑centre locations, power sourcing, how much spending will stay in the domestic supply chain, and the likely number of permanent local jobs. NSW has launched scrutiny of the sector, with a parliamentary inquiry due to report in September 2026.

On Tesla’s Q1 2026 earnings call (April 22–23), CEO Elon Musk acknowledged that vehicles equipped with the company’s Hardware 3 (HW3) — roughly four million cars produced through 2023 — cannot run an unsupervised version of Full Self‑Driving (FSD). Musk said HW3 has only one‑eighth the memory bandwidth of Tesla’s Hardware 4 (AI4) and lacks the processing and camera capability required for unsupervised autonomy. Tesla will offer discounted trade‑ins and retrofit options — including replacing the onboard computer and cameras — and plans to build “microfactories” in major metropolitan areas to perform mass upgrades more efficiently than service centers. The company said it will continue to deliver incremental FSD feature updates to HW3 cars but that anything approaching unsupervised FSD will require hardware replacement. The admission comes amid customer frustration and pending legal actions in some markets from owners who bought FSD expecting future software‑only enablement.

Paris-based satellite startup Univity said on April 23, 2026 it closed a €27 million Series A round, bringing total secured funding to about €68 million when combined with a prior €31 million contract from the French space agency CNES. Investors in the round include Bpifrance’s Deeptech 2030 fund, venture firm Blast, Expansion and two family offices. Founded in 2022, Univity is developing a very low Earth orbit (VLEO) wholesale network intended to sell space-based internet and mobile services to telecom operators rather than direct to consumers. The company plans a fleet of up to 3,400 satellites in VLEO (roughly 375 km altitude), has signed commercial agreements with 16 operators across four continents, and intends to manufacture satellites near Toulouse to control costs. Current funds will support building and launching its first two demonstration satellites in a 2027–28 timeframe and the uniShape VLEO 5G demonstrator, ahead of a planned scale-up from 2028 that would use infrastructure financing from large investors and operators. Univity positions itself as a European alternative to consumer-facing systems like Starlink and Amazon Kuiper.