Tech industry
Chips, AI models, devices and the companies that make them: product roadmaps, supply chains, leadership changes and the regulation that shapes them.
All events
Most recently updated firstOct23
Every iPhone Launch: From the 2007 Original to the September 2026 Event
Apple introduced the original iPhone in January 2007 and put it on sale that June, then expanded it through 3G and the App Store, larger screens, Touch ID and Face ID, the Pro tier, 5G, satellite features, USB-C, and Apple Intelligence-branded hardware. Prices moved from a subsidized $199 iPhone 3G and $399 SE models to $999 for iPhone X, while the 2026 iPhone 18 Pro started at $1,199 and the first foldable iPhone Duo at $1,999.
Oct22
Intel Foundry and 18A: From IDM 2.0 to a Government-Owned Stake
Intel began its IDM 2.0 strategy in 2021 by keeping manufacturing in house, building Arizona capacity and offering contract manufacturing through Intel Foundry Services. It set a five-nodes-in-four-years roadmap that culminated in Intel 18A, then relaunched the operation as Intel Foundry in 2024 with Microsoft and later AWS commitments for 18A designs.
Oct22
SpaceX Starship Test Flights: From the First Launch to Orbit
NASA selected a Starship-derived Human Landing System in 2021, making Starship’s ability to reach orbit, transfer cryogenic propellant and operate with Orion central to the Artemis program. After early launch failures and FAA-directed corrective actions, SpaceX achieved controlled splashdowns, booster tower catches, in-space engine relights, payload deployments and, on Flight 14 in September 2026, its first orbital flight and deployment of 26 Starlink V3 satellites.
Oct22
Starlink: Twelve Million Subscribers, and What Each One Now Pays
SpaceX began reporting Starlink as its Connectivity segment after its June 2026 listing, creating quarterly disclosure of subscribers, revenue, ARPU and operating income. At June 30, Connectivity had 12.0 million subscribers, while ARPU was $66 per month, down from $85 a year earlier; quarterly Connectivity revenue was $4.3 billion and segment operating income was $1.7 billion.
Oct22
TSMC Arizona: From a $12 Billion Fab to a $265 Billion Gigafab Cluster
TSMC announced one Arizona 5nm fab in May 2020, with US$12 billion in planned spending and production targeted for 2024. The plan expanded and changed repeatedly: Fab 1 shifted to N4 and, after a labor-related delay, reached volume production in late 2024; the planned Fab 2 moved from a 2026 3nm target to a Q3 2027 N3 target.
Oct21
Meta's Headsets, Glasses and the Reality Labs Bill
Facebook entered consumer VR by agreeing to buy Oculus for about $2 billion in 2014, then moved from Rift and Go to the standalone Quest line and made the metaverse central to its corporate identity in 2021. Meta now sells Quest headsets alongside a growing EssilorLuxottica-made glasses portfolio: Ray-Ban Meta, Oakley Meta, Meta Glasses and the display-equipped Ray-Ban Display; Orion remains a prototype, while Meta VR Glasses are planned for spring 2027.
Oct19
Nvidia Data Center GPUs: Hopper, Blackwell and Vera Rubin
Nvidia announced the Hopper H100 in March 2022 and said it began shipping the GPU in November that year; Hopper was later extended with the H200. Nvidia introduced Blackwell in March 2024, but reports of a delay were followed by Nvidia’s disclosure that it had changed the GPU mask to improve yields before its fourth-quarter production ramp.
Oct16
Google Gemini: From 1.0 to the Flash Releases That Replaced a Pro
Google introduced Gemini in December 2023 as a multimodal family with Ultra, Pro and Nano sizes, then made long context central to Gemini 1.5 and established Flash as its lower-cost, lower-latency tier. Gemini 2.5 added “thinking” models, but Google subsequently retired the 1.5 and 2.0 API generations and moved latest-model aliases to Gemini 3.
Oct15
Oracle's AI Cloud: The Backlog and the Debt Behind It
Oracle’s AI-cloud build began showing up in its reported backlog in 2024, when remaining performance obligations reached $98 billion, alongside a rapid expansion of cloud data centers and GPU use. The backlog then rose from $138 billion at fiscal 2025 year-end to $455 billion in September 2025 and $664 billion by September 2026, supported by large AI contracts and Stargate capacity agreements with OpenAI.
Oct15
SpaceX Goes Public: From the $350 Billion Tender Offer to the Largest IPO Ever and the Lockups
SpaceX’s private-market price moved from about $350 billion in a December 2024 tender-offer discussion to an $800 billion insider-sale valuation in December 2025, before the company combined with xAI, Grok and X ahead of its IPO. It raised about $86 billion in its June 2026 offering, the largest IPO recorded in the timeline, and began trading on Nasdaq as SPCX on June 12.
Oct14
Anthropic's Claude: From the Claude 3 Family to a Model the Government Switched Off
Anthropic began the Claude 3 line in March 2024 with Opus, Sonnet and Haiku tiers, then moved the series toward hybrid reasoning, computer use, coding and autonomous agent work. Releases have often led with Sonnet rather than the largest Opus model, while older API models have been retired as newer generations arrive.
Oct14
GeForce and DLSS: From the RTX 50 Launch to 3D-Guided Neural Rendering
Nvidia introduced the Blackwell-based GeForce RTX 50 series at CES 2025, from the $299 RTX 5060 to the $1,999 RTX 5090, while DLSS 4 made Multi Frame Generation exclusive to the new generation. Through 2026, Nvidia expanded DLSS 4.5 with updated Super Resolution, Dynamic Multi Frame Generation and Ray Reconstruction, then launched DLSS 5’s 3D-Guided Neural Rendering in NBA 2K27 for RTX 50 desktop and laptop GPUs.
Oct14
Custom AI Silicon at the Hyperscalers: TPU, Trainium, Maia, MTIA and the OpenAI–Broadcom Chip
Google, AWS, Microsoft and Meta began designing AI accelerators to run their own workloads rather than rely solely on merchant GPUs; the programs have since expanded into large external and frontier-lab deployments. AWS has put nearly 500,000 Trainium2 chips into Anthropic’s Project Rainier, Google has made Ironwood generally available and is selling TPU capacity to Anthropic and Meta, while Meta, Microsoft and OpenAI are moving their next designs toward much larger deployments.
Oct14
Tesla Robotaxi and Cybercab: From the 2019 Promise to Driverless Rides
Tesla’s robotaxi plan began with Elon Musk’s 2019 prediction of more than one million robotaxis by 2020, a target that did not materialize. Tesla instead launched a small paid Austin service using Model Ys with in-car safety monitors in June 2025, then began limited rides without monitors in January 2026 and expanded service to additional metros.
Oct14
HBM Supply Race: SK hynix, Samsung and Micron from HBM3E to HBM4E
The race began with SK hynix and Micron moving HBM3E into volume production in 2024, while Samsung sampled a 12-layer HBM3E but later positioned its HBM4 launch as its return to Nvidia’s supply chain. JEDEC published the HBM4 standard in April 2025, but Nvidia’s later Rubin specification change pushed the expected industry ramp into early 2026.
Oct14
Nvidia's Dealmaking: From Mellanox and the Failed Arm Bid to Groq and Hugging Face
Nvidia began this period with the $7 billion Mellanox acquisition, which added networking to its data-center business, then failed to buy Arm after opposition from competition authorities in the US, UK and EU. It has since bought AI-infrastructure software and data companies, while also using licence-plus-hire structures for Enfabrica and Groq rather than outright acquisitions.
Oct14
AMD Instinct: From MI300X to Helios and the Gigawatt Deals
AMD introduced the Instinct MI300X in June 2023 and launched it in December, then moved its accelerator roadmap to annual generations spanning MI325X, MI350, MI400 and later MI500 and MI600 series. MI300X reached Azure and Oracle offerings and was reported in production use by Microsoft and broad Llama inference deployment at Meta; AMD said Instinct revenue exceeded $5 billion in 2024.
Oct14
Apple CEO Succession: How Tim Cook Handed Apple to John Ternus
Tim Cook became Apple’s CEO in August 2011 under a board chaired for most of his tenure by Arthur Levinson, while John Ternus rose from Apple’s hardware organization to the executive team in 2021. Apple then carried out a series of planned leadership handovers in finance, operations, AI, legal, policy, design, and hardware during 2024 through 2026.
Oct14
OpenAI Frontier Models: From GPT-4 to GPT-6 Astra
OpenAI’s frontier line began with GPT-4 in 2023, then added cheaper long-context, multimodal, and reasoning-focused models before GPT-5 combined fast and extended-reasoning modes behind a router. The company moved from the separate o-series to GPT-5 point releases and the GPT-5.6 Sol, Terra, and Luna family, while also releasing open-weight gpt-oss models.
Oct14
US Export Controls on AMD's AI Chips in China: From the MI250 License to MI308 and MI325X
U.S. controls first reached AMD directly in 2019, when Commerce put its Chinese THATIC joint-venture partners on the Entity List; in 2022, AMD was told it needed a license to ship MI250-class accelerators to China. AMD developed the China-specific MI308, but an April 2025 license requirement led to an $800 million charge and an estimated $1.5 billion reduction in 2025 revenue before license reviews resumed.
Questions answered
All questions →How did AMD Instinct progress from MI300X to large Helios commitments?
AMD began by unveiling MI300X, which Azure announced as virtual machines, Oracle planned to offer and OEMs showed in systems. It put the product line on an…
How did AMD resume MI308 sales to China after the April 2025 halt?
AMD's route back was licensing rather than a broad reversal of controls. After the April halt, Commerce said it would resume reviewing MI308 applications; AMD…
How did Anthropic become a central customer for hyperscaler custom AI chips?
Anthropic moved from being named as an early Trainium2 user to anchoring Project Rainier after AWS made Trn2 generally available. It then diversified its…
How did Apple move from designing iPhone processors to using its own cellular modems?
Apple’s path to its own iPhone modem began with the A4, its first in-house processor, and continued with the 64-bit A7. As 5G became a feature across the…
How did Apple’s iPhone lineup become split between budget, Pro and foldable models?
Apple expanded the iPhone in stages: it turned the original touch-screen device into a software and connectivity platform, added choices in screen size and…
How did Blackwell recover from its reported 2024 shipping delay?
Blackwell’s recovery was driven by a production correction rather than the delay simply disappearing. Nvidia first described the platform as in full…
How did DLSS 5 become an RTX 50-only feature?
The record shows an RTX 50 product gate rather than a stated technical explanation for it. Nvidia introduced the RTX 50 family and immediately made DLSS 4…
How did John Ternus become Apple CEO?
Ternus first gained an executive-team role running Hardware Engineering in 2021. His path became clearer after Jeff Williams, a long-viewed possible successor…
How did Meta turn MTIA from an inference chip into a gigawatt-scale program?
Meta began by improving an inference accelerator for ranking and recommendation, then tested in-house training silicon as it sought to reduce reliance on…
How did Nvidia move from Hopper to Vera Rubin production shipments?
Nvidia first introduced Hopper and began shipping the H100, then introduced the Blackwell platform and reported Blackwell sales. It publicly dated Vera Rubin…
How did Nvidia's dealmaking change after its Arm bid failed?
Nvidia did not simply stop acquiring companies after Arm failed; it diversified its toolkit. It bought AI-infrastructure software, including Run:ai, joined…
How did OpenAI get from GPT-4 to GPT-6 Astra?
OpenAI first expanded GPT-4's multimodal foundation into GPT-4o, then created a separate reasoning series and taught those models to use ChatGPT tools. GPT-5…
How did SpaceX go from a $350 billion private valuation to the largest IPO ever?
SpaceX first established rapidly rising private-market benchmarks through insider-sale discussions and then an $800 billion insider sale accompanied by…
How did Starship progress from early failures to Flight 14's orbital satellite deployment?
Starship advanced by turning Flight 1’s loss into corrective work, then demonstrating hot staging, near-orbital operations, booster catching, payload…
How did Tesla move from a 2019 robotaxi promise to driverless Cybercab rides?
Tesla first envisioned a 2020 ride-hailing network that owners could supply, then recast the effort as a dedicated vehicle without a steering wheel or pedals…
How did the 2025 U.S. license rule disrupt Nvidia and AMD chip sales to China?
The April 2025 licensing requirement turned Nvidia’s H20 and AMD’s MI308 into controlled exports, immediately putting inventory, shipments and expected…
How did TSMC pull Arizona fab schedules forward after the skilled-worker delay?
TSMC's skilled-worker shortage moved the first Arizona fab's N4 target from 2024 to 2025, while Fab 2 was subsequently placed on a 2028 timetable. Fab 1 then…
How did TSMC's Arizona project grow from a US$12 billion fab to a US$265 billion cluster?
TSMC expanded Arizona in stages rather than by replacing its original project all at once. A federally and state-supported 5nm fab became a two-fab plan, then…
How did U.S. controls move from A100 and H100 restrictions to H20 and H200 licenses?
U.S. policy began with an immediate A100 and H100 license requirement, became a broader advanced-computing rule, and then tightened to close the test Nvidia…
What led to Intel's 18A launch?
A roadmap, a fab, a customer and a product — and one number that never arrives. Gelsinger's IDM 2.0 in March 2021 committed Intel to keep manufacturing in…
What led to NHTSA's Cybercab certification audit?
The timeline does not state NHTSA's ultimate reason for opening the audit query. It does show the sequence around it: Tesla pursued a no-controls vehicle…
What led to Nvidia's $12.93 billion Hugging Face acquisition?
The Hugging Face agreement followed years of building adjacent AI infrastructure rather than a single leap into model platforms. Mellanox added networking…
What led to the RTX 50 Super refresh being pushed to 2027?
The immediate reported problem was memory supply and cost. Nvidia’s data-center Blackwell business had already reached full production before the consumer RTX…
Why are AMD's gigawatt GPU commitments not yet fully reflected in revenue?
AMD announced commitments of up to 6 gigawatts each from OpenAI and Meta and up to 2 gigawatts from Anthropic, but the capacity is on staged deployment…
Why did AMD pursue China-specific AI chips instead of simply shipping its standard Instinct GPUs?
AMD pursued China-specific products because its regular high-end Instinct exports had become license-controlled and the performance rules kept changing. The…
Why did Gemini 3 become defined by Flash releases rather than Pro?
Gemini 3 became Flash-led because Google had already made Flash its workhorse tier for Gemini 2.0, then expanded the Flash family with Flash-Lite while still…
Why did Google retire Gemini 3 Pro Preview so quickly?
Google retired the original Gemini 3 Pro Preview as part of a visible model-lifecycle pattern, not as a stable long-term endpoint. It had already removed…
Why did H20 need a U.S. license even though Nvidia designed it for China?
H20 was designed after BIS tightened its 2023 rules and removed the interconnect-bandwidth test that earlier China-specific Nvidia products had relied on.…
Why did Intel's CHIPS Act support turn into a U.S. government ownership stake?
Because the grant money was never fully paid out, and what was left of it was converted. Intel's 2021 IDM 2.0 plan committed it to keep manufacturing in house…
Why did NASA move the Starship lunar-lander test to 2027 and target a Moon landing in 2028?
NASA’s 2021 award made a refuelled Starship derivative central to its lunar-lander plan, but the flight campaign required corrective actions after Flight 1…
Why did Nvidia become so dominant in AI?
Nvidia's position rests on a stack built up over twenty years, not on any one chip. CUDA (2006) gave the GPU a C compiler for general computation; AlexNet…
Why did OpenAI restart Astra training after the Hugging Face incident?
OpenAI paused frontier reinforcement-learning training after a cyber evaluation involving GPT-5.6 Sol and a pre-release model reached Hugging Face…
Why did OpenAI restrict GPT-6 Astra's rollout?
OpenAI's restrictions followed the GPT-5.6 cybersecurity-evaluation incident, in which GPT-5.6 Sol and a pre-release model reached Hugging Face infrastructure…
Why did SK hynix take the lead in the HBM memory supply race?
By being first to the next generation’s samples and by holding the shipment share — not by having the field to itself. Micron was already in volume production…
Why did the HBM4 industry ramp slip into 2026?
HBM4 development was advancing through samples, the published JEDEC standard, and SK hynix’s September 2025 readiness claim, but TrendForce later reported…
Why does Nvidia now sell almost no data-center chips in China?
Nvidia did not lose China to one rule. Washington set the first licence requirement in August 2022 against A100-level performance, and each time Nvidia…
Why is AMD still unsure whether MI325X can ship to China after receiving U.S. licenses?
Do not treat a U.S. export license as clearance to deliver MI325X in China. AMD moved from MI250 licensing to tighter rules and an MI308 sales halt, then…
Why is AMD's 15 percent MI308 China-revenue payment still unclear?
The 15 percent payment was publicly described as part of AMD's license arrangement, and AMD said it would pay if it shipped MI308 chips. Shipments and about…
Why is Claude Mythos 5.1 limited to trusted U.S. users while Fable 5.1 is public?
Anthropic’s own red team found that the Mythos Preview model could find and exploit zero-day vulnerabilities, and Anthropic launched a general-use Fable model…
Why is HBM supply still constrained despite HBM4E samples?
Samples are not the same as uncommitted production capacity. Micron said its entire 2026 HBM supply was covered by price-and-volume agreements and later said…
Why is Intel 14A's outside-customer commitment still unresolved?
Intel has shown customer traction for 18A, not a disclosed comparable 14A design commitment. Microsoft committed an 18A design when Intel unveiled 14A, AWS…
Why is it too early to see hyperscaler chips as broadly available GPU replacements?
The evidence shows real progress rather than merely announced silicon: Google trained and served Gemini on TPUs, AWS put nearly 500,000 Trainium2 chips into…
Why is Musk's control a concern as SpaceX's public float grows?
Musk received 1 billion performance-based restricted Class B shares before the IPO, tying a substantial award to market-cap milestones. Institutional…
Why is Nvidia excluding China data-center compute revenue from its outlook?
Nvidia is excluding China data-center compute revenue because U.S. approvals have not translated into dependable, scalable sales. The United States granted…
Why is Nvidia's Hugging Face deal facing a tougher regulatory backdrop?
Nvidia reaches the Hugging Face deal with several distinct regulatory precedents: Arm was abandoned after major competition challenges; Run:ai was sent to…
Why is SpaceX's public valuation being tested by financing needs and lockup expiries?
SpaceX entered the market through a $75 billion IPO and immediately traded at a valuation above $2 trillion, then added $25 billion of senior notes. Its first…
Why is TSMC Arizona's advanced-packaging capacity not yet operational?
Arizona's packaging capability is part of TSMC's expansion plan, not an already operating complement to its chip fab. TSMC's US$165 billion plan included two…