Hardware

Modular data center startup Armada closed a $230 million oversubscribed Series B at a $2 billion pre-money valuation, with Johnson Controls signing on to manufacture units at a new 400,000-square-foot Arizona factory. The company's containerized AI compute units serve the U.S. Navy, offshore oil rigs, and remote industrial sites — markets where hyperscale cloud infrastructure cannot reach.

Samsung Foundry and Cadence Design Systems unveiled a joint Physical AI chiplet platform at CadenceLive 2026, built on Samsung's 5nm SF5A process. The pre-verified modular design targets real-time camera processing, AI radar, and autonomous systems, with tape-out planned for early 2027 and volume production in the second half of that year.

Chinese authorities have certified nine domestically designed AI processors — including Huawei's Ascend 910 and Alibaba's T-Head Zhenwu M890 — under the country's Anke security framework, creating the first formal 'AI training and inference chips' procurement category. The move accelerates China's systematic effort to replace foreign silicon in government and state-enterprise AI infrastructure.

NVIDIA confirmed during its record-breaking Q1 FY2027 earnings call that the Vera Rubin platform will ship in Q3 2026 with a full volume ramp by Q4. The seven-chip architecture promises 50 petaFLOPs of NVFP4 inference per GPU and 3.6 exaFLOPs of rack-scale throughput — potentially the most consequential AI hardware generation since the Transformer took hold.

A March 2026 paper from Oratomic, Caltech, and UC Berkeley shook the cryptography world by showing that Shor's algorithm could crack RSA-2048 and elliptic-curve encryption with as few as 10,000 neutral-atom qubits — a tenfold improvement over previous estimates. Nature called it 'a real shock.' With Q-Day now a plausible 2030 milestone, organizations face a narrowing window to migrate to post-quantum cryptography before years of intercepted data become readable.

SpaceX launched the first Starship V3 on May 22 from its new Pad 2 at Starbase, reaching space and completing a controlled splashdown in the Indian Ocean despite losing one engine and its Super Heavy booster. The mission deployed 22 mock Starlink satellites — including two 'Dodger Dog' camera spacecraft designed to automate heat shield inspection — marking a pivotal step toward rapid reusability at commercial scale.

Weeks after recalling its entire 3,791-vehicle fleet over a flooded-road software flaw, Waymo's OTA patch failed a second time in Atlanta on May 21. The company suspended robotaxi service across six cities and halted all freeway rides nationwide — with no permanent fix yet in sight.

Amazon's Q1 2026 earnings revealed its Trainium custom silicon business has surpassed a $20 billion annual revenue run rate, growing at triple-digit percentages year-over-year, with $225 billion in committed backlog from customers including OpenAI and Anthropic. CEO Andy Jassy called it one of the top three data center chip businesses in the world — and said the real value is twice that if sold externally.

Anthropic is in early-stage negotiations to rent Azure servers powered by Microsoft's Maia 200, a second-generation custom AI accelerator that could cut per-token inference costs by 30–50% compared to NVIDIA H200 instances. The deal would make Anthropic the first major external customer for Microsoft's custom silicon program, deepening the two companies' existing $30 billion cloud relationship.

AMD has begun production ramping its 6th Gen EPYC processor, codenamed Venice, on TSMC's N2 process technology — making it the first high-performance computing chip in the industry to reach production at the 2nm node. The 256-core chip claims a 70% performance leap over its predecessor and marks a pivotal transition from FinFET to gate-all-around nanosheet transistors, the most significant architectural change in chip fabrication in over a decade.

Nvidia's Q1 fiscal 2027 results delivered $81.6 billion in revenue and $75.2 billion from data centers alone — yet the stock slipped as CEO Jensen Huang publicly acknowledged the company has 'largely conceded' China's AI chip market to Huawei following years of tightening US export controls. The dual narrative defines where the world's most powerful chip company stands heading into the second half of 2026.

The Commerce Department is distributing $2 billion in CHIPS Act funding to nine quantum computing companies, with IBM receiving the largest share at $1 billion. In an unprecedented move, the government will take minority equity stakes in every recipient—a model borrowed from the semiconductor grants playbook and now applied to the emerging quantum sector for the first time.

Alibaba's T-Head semiconductor unit launched the Zhenwu M890 AI accelerator, claiming triple the performance of its predecessor and targeting Nvidia's H20 in China's restricted chip market. Paired with the Qwen 3.7-Max model that ran uninterrupted for 35 hours calling over 1,000 tools, the announcements represent the most comprehensive vertical AI stack from any Chinese tech giant to date.

Samsung Electronics reached a last-minute tentative deal with its 48,000-member union on May 21, averting what would have been the largest work stoppage in semiconductor history. A 6.2% pay raise and a new profit-sharing bonus averaging $338,000 per chip division worker pushed Samsung shares up 8%, but the agreement still needs a union ratification vote before it becomes final.

London-based Fractile has raised $220 million in a Series B round led by Accel, Factorial Funds, and Founders Fund, hitting a $1 billion post-money valuation. The company is building inference chips that co-locate memory and compute on a single die — eliminating the expensive DRAM bottleneck that slows and inflates the cost of running frontier AI models. Anthropic is reportedly in early talks to become a customer.

Mind Robotics, a Rivian spinout founded by RJ Scaringe less than six months ago, has raised a $400 million Series B led by Kleiner Perkins, pushing total funding past $1 billion and valuation to $3.4 billion. The company is deploying foundation-model-powered robots in manufacturing — using Rivian's own Illinois plant as its live training ground.

TSMC's N2 process node, which uses Gate-All-Around transistors and carries a $30,000-per-wafer price tag, is ramping toward 140,000 wafers per month by year-end 2026—yet orders are already booked through 2028. Apple has locked up over half the initial allocation for its A20 and M6 chips, while Nvidia's Rubin AI architecture depends on N2 for the next generation of data center GPUs.

AI chipmaker Cerebras Systems priced its IPO at $185 per share on May 13, raised $5.55 billion, and soared 68% on its first trading day before giving back roughly 10% — a wild debut that underscores both voracious investor appetite for AI hardware and the unanswered questions about a company almost entirely dependent on OpenAI.

South Korea's National Samsung Electronics Union plans an 18-day walkout starting May 21 at Samsung's Pyeongtaek campus, a dispute rooted in how the AI chip boom's profits are being distributed. Analysts warn the action could remove 3–4% of global DRAM supply and push memory prices sharply higher, rippling through PC and smartphone markets worldwide.

Washington approved H200 chip exports to ten major Chinese firms including Alibaba, Tencent, and ByteDance — yet zero deliveries have occurred. Beijing is quietly steering its tech giants away from the deal, and Nvidia's CEO joined Trump's Beijing summit hoping to break the stalemate. Nvidia's China AI market share has effectively collapsed to zero.

The Wall Street Journal reports that Google and SpaceX are in advanced talks to launch orbital AI data centers into space, pairing Google's Project Suncatcher satellite initiative with SpaceX's unprecedented launch capacity. The potential deal — disclosed days before SpaceX's $1.75 trillion IPO — would create the first commercial space-based AI compute platform and force the entire cloud industry to reckon with a new dimension of infrastructure competition.

Apple and Intel have struck a preliminary agreement for Intel to manufacture Apple's base M-class processors at its Arizona fabs starting in 2027, using Intel's cutting-edge 18A-P process. The deal — facilitated by direct White House involvement — represents a landmark supply chain diversification for Apple and a validation of Intel's foundry revival under CEO Lip-Bu Tan.

Google unveiled the Googlebook at its Android Show I/O Edition on May 12, 2026 — a new laptop category powered by Gemini Intelligence that succeeds the 15-year-old Chromebook. With hardware partners including Dell, HP, Lenovo, Asus, and Acer, and a fall 2026 launch window, Googlebook takes direct aim at Apple's MacBook franchise with features like AI-powered Magic Pointer and seamless Android device integration.

Nvidia and Corning announced a multiyear partnership on May 7 that gives Nvidia rights to invest up to $3.2 billion in the glass and fiber company. Corning will 10x its US optical connectivity capacity, expand fiber production by more than 50%, and open three new facilities in North Carolina and Texas—creating over 3,000 jobs—as AI data centers demand an entirely new generation of optical interconnects.

Cristiano Amon disclosed that Qualcomm is developing undisclosed AI form factors with 'pretty much all' major AI companies, explicitly naming OpenAI and Meta. The new Snapdragon Wear Elite chip—supporting 2B-parameter models on-device—is the silicon foundation for what Amon calls the 'ecosystem of you': glasses, earbuds, and autonomous agents replacing the smartphone as the center of digital life.

Nintendo has announced a global price increase for the Switch 2, raising the US price from $449.99 to $499.99 effective September 1, 2026. The company cited 'changes in market conditions' — a phrase widely understood to encompass Trump-era tariffs, trade war pressures, and rising memory costs — in what marks the console's first price hike since its 2025 launch.

NVIDIA and IREN Limited announced a landmark strategic partnership on May 7 to deploy up to 5 gigawatts of next-generation AI factory infrastructure across IREN's global data center pipeline. The deal includes NVIDIA's option to invest up to $2.1 billion in IREN equity plus a separate $3.4 billion five-year Blackwell GPU cloud contract — sending IREN's stock surging nearly 20% and marking NVIDIA's deepest yet push to become the central organizer of the AI factory ecosystem.

Dutch quantum hardware startup QuantWare closed a $178 million Series B — the largest private round ever for a dedicated quantum processor company — to build KiloFab, a purpose-built factory targeting 10,000-qubit processors and a 20x jump in production capacity. Intel Capital joined as a lead new investor.

Advanced Micro Devices reported blowout first-quarter 2026 results on May 5, posting $10.3 billion in revenue—a 38% year-over-year gain—while confirming a landmark $60 billion, multi-year agreement to supply Meta Platforms with custom MI450 GPUs and Helios rack-scale infrastructure. The results signal a structural shift in AI chip competition, as AMD mounts a credible challenge to Nvidia's long-standing dominance in hyperscaler data centers.

Arm Holdings released its fourth-quarter fiscal 2026 results on Wednesday after market close, with analysts expecting $1.47 billion in revenue — a 19% year-over-year increase — and adjusted EPS of $0.58. The report comes as ARM stock has surged 84% year-to-date on the strength of triple-digit data center royalty growth, rising adoption of ARM-based custom silicon at hyperscalers, and the company's first-ever move to build its own CPU, announced in March.

Oregon startup Panthalassa closed a $140 million Series B led by Peter Thiel to build floating, autonomous ocean platforms that generate electricity from wave energy and use seawater cooling to run AI inference workloads. The company plans to deploy its Ocean-3 pilot nodes in the northern Pacific by end of 2026, with commercial operations in 2027, aiming to deliver compute power at as little as $0.02 per kWh.

AI chipmaker Cerebras Systems has refiled its IPO prospectus targeting a $4 billion raise at a $40 billion valuation, underpinned by a landmark inference compute agreement with OpenAI worth over $10 billion through 2028. With $510 million in 2025 revenue and a $24.6 billion backlog, the Nasdaq listing could become the largest AI chip IPO in U.S. history.

Tesla has begun volume production of the Cybercab at Giga Texas after engineering it to comply with all existing federal safety standards without requiring a regulatory waiver — a maneuver that removes the government-imposed 2,500-vehicle production ceiling that has constrained rivals. But the safety numbers from Tesla's supervised robotaxi fleet raise questions about whether the Cybercab's commercial ambitions are running ahead of its autonomous driving software.

Qualcomm's Q2 FY2026 earnings revealed an unexpected pivot: the company has secured a custom silicon deal with a major unnamed hyperscaler, with first shipments expected in December 2026. The disclosure, paired with record automotive revenue and a $20 billion buyback authorization, sent shares surging 15% despite overall revenue declining year-over-year — and positions Qualcomm as a new challenger in the custom chip market long dominated by Broadcom and Marvell.

Amazon's Q1 2026 results revealed two intertwined stories: AWS grew 28% to $37.6 billion — its fastest pace in nearly four years — while the company's custom chip business (Trainium, Graviton, Nitro) quietly crossed a $20 billion annual revenue run rate growing at triple-digit rates. CEO Andy Jassy hinted the business could eventually sell chips externally, at a potential $50B valuation.

Meta has secured an agreement with Overview Energy for up to 1 gigawatt of space-based solar power, using a constellation of satellites to beam near-infrared light down to terrestrial solar farms — including at night. The orbital demonstration is planned for 2028, with commercial grid delivery targeting 2030, as Meta races to solve the energy math behind its $115–135 billion AI capex plan.

Google, Amazon, Microsoft, and Meta are deploying custom AI accelerators that cut compute costs by 30–65% versus Nvidia GPUs. As AI shifts from training to inference at scale, analysts project Nvidia's inference market share could collapse from over 90% to just 20–30% by 2028—the most consequential hardware power shift since the GPU era began.

Japan Airlines Ground Service and GMO AI & Robotics Trading are launching Japan's first humanoid robot trial at Haneda Airport in May 2026. Two robots — the Unitree G1 and UBTECH Walker E — will handle cargo transport and baggage loading as the aviation industry grapples with a chronic labor shortage that human hiring alone cannot solve.

BMW Group has launched its first European humanoid robot pilot at the Leipzig plant, deploying Hexagon Robotics' AEON units in high-voltage battery assembly and exterior parts manufacturing. The April 2026 test phase—ahead of a full summer pilot—marks a milestone for physical AI entering the continent's largest industrial sector, with implications for automation, labor policy, and the global race to deploy AI in the physical world.

Tesla's first-quarter 2026 earnings beat analyst expectations, but the real news was Elon Musk's $25 billion capital expenditure commitment — nearly three times 2025 spending — along with confirmation that Optimus humanoid robot production lines will begin running in late July. The announcement marks Tesla's most decisive pivot yet from electric vehicle maker to an integrated AI, robotics, and autonomous transport platform.

Researchers at the University of Cambridge have engineered a hafnium oxide-based memristor that mimics how neurons simultaneously store and process information, potentially slashing AI hardware energy consumption by up to 70%. The device operates at switching currents one million times lower than conventional memristors and was published in Science Advances on April 22, 2026.

Intel reported Q1 2026 revenue of $13.58 billion, crushing analyst estimates of $12.42 billion, as data center revenue soared 22% to $5.1 billion. CEO Lip-Bu Tan declared the CPU 'indispensable' to the AI era, marking Intel's sixth consecutive quarter beating its own forecasts and sending shares up 20% after hours.

SK Hynix reported first-quarter 2026 net profit of 40.35 trillion won ($27.3B), up 398% year-over-year, on revenue that grew 198% to 52.58 trillion won. The South Korean memory giant commands a 57% share of the global HBM market and plans to ship HBM4E samples in H2 2026, with mass production targeted for 2027.

Google made its seventh-generation Ironwood TPU generally available at Cloud Next 2026, delivering 4.6 petaFLOPS per chip and 42.5 exaFLOPS at pod scale — a 10× leap over TPU v5p. The company simultaneously previewed its eighth-generation split architecture: a Broadcom-designed training chip and a MediaTek-designed inference chip, both on TSMC's 2nm process, targeting 2027. A $750 million partner fund and expanded A2A agent protocol round out Google's most comprehensive challenge yet to Nvidia's AI infrastructure dominance.

Photonic quantum computing startup Xanadu Quantum Technologies saw its stock surge nearly 250% after NVIDIA released Ising — a family of open-source AI models for quantum error correction and calibration. The rally minted CEO Christian Weedbrook a billionaire just weeks after Xanadu's Nasdaq debut via a $302M SPAC transaction, and pushed the company's market cap toward $16 billion.

Google is negotiating a deal with Marvell Technology to jointly develop two new AI chips: a memory processing unit designed to eliminate inference bottlenecks alongside its existing TPUs, and a cheaper, more efficient inference-optimized TPU. The talks add Marvell as a third custom chip design partner alongside Broadcom and MediaTek, and sent Marvell shares surging while Broadcom stock fell sharply.

TrendForce projects China's humanoid robot production will grow 94% this year, with Unitree and AgiBot together capturing nearly 80% of global shipments. Unitree's $610 million Shanghai IPO filing — potentially the first for a pure-play humanoid robotics company — caps a year of explosive growth: 335% revenue surge, 674% profit increase, and 5,500+ units shipped in 2025.

A wave of European AI chip startups — led by Dutch company Euclyd, backed by ex-ASML CEO Peter Wennink — are raising $100 million or more each, targeting what they say is a 100x power efficiency gap in Nvidia's inference stack. As AI inference workloads scale globally, a new generation of purpose-built silicon is emerging to challenge GPU supremacy.

Huawei has begun mass producing the Ascend 950PR, an AI inference chip delivering 2.8 times the FP4 performance of Nvidia's H20, paired with the Atlas 350 accelerator card. ByteDance has committed $5.6 billion for 750,000 units in 2026, while Alibaba and Tencent follow — marking China's most credible challenge yet to Nvidia's dominance in its own backyard.

Elon Musk is intensifying supplier engagement for Terafab, the $20 billion vertically integrated semiconductor complex announced in March 2026 at Tesla's GigaTexas site. With Intel already signed as a manufacturing partner, the initiative targets one terawatt of compute output per year — roughly doubling U.S. capacity — in a direct challenge to the existing global chip supply chain.

Meta Platforms announced price increases of $50–$100 on its Quest 3 and Quest 3S VR headsets, effective April 19, blaming a global DRAM shortage driven by AI infrastructure demand. The price hike is partly self-inflicted: Meta's own $115–135 billion AI capital expenditure plan for 2026 is one of the largest single contributors to the memory market tightness now forcing its consumer hardware division to raise prices.

Science Corporation, founded by ex-Neuralink president Max Hodak, is preparing to place its first 520-electrode recording sensor in a human brain. Yale neurosurgery chair Dr. Murat Günel has been appointed medical director, with an opportunistic trial design that piggybacks on existing cranial surgeries to minimize additional risk.

On April 19, Beijing's E-Town district hosts the world's largest humanoid robot half-marathon, with over 100 teams — a nearly fivefold increase from last year — competing over a 21-kilometer urban course. The event, spanning autonomous and remote-controlled categories, is the most ambitious real-world stress test of China's booming humanoid robot industry.

NVIDIA has released Ising, a family of open-source AI models designed to accelerate the path to fault-tolerant quantum computers — delivering error correction that is 2.5x faster and 3x more accurate than leading open-source tools. The launch sent quantum computing stocks surging and marks a pivotal convergence between classical AI infrastructure and quantum hardware development.

Elon Musk announced on April 15 that Tesla's next-generation AI5 processor has completed tape-out, delivering up to 8× the compute, 9× the memory, and 5× the bandwidth of its predecessor. The chip — nearly two years behind the original schedule — will be fabricated at TSMC's Arizona plant and Samsung's Texas facility, with engineering samples expected by late 2026.

The world's only maker of extreme ultraviolet lithography machines posted €8.8 billion in Q1 net sales and lifted its full-year outlook to €36–40 billion, but shares fell as China's share of system revenue plummeted from 36% to 19% and US lawmakers push to extend export bans to legacy DUV equipment.

Meta Platforms announced a major expansion of its partnership with Broadcom on April 14, committing to deploy over one gigawatt of custom AI accelerators based on Broadcom's XPU platform through multiple MTIA chip generations. The multi-year deal extends through 2029 and signals Meta's serious intent to reduce its dependence on Nvidia as the AI infrastructure arms race intensifies.

IonQ announced on April 14 that it has achieved the first-ever photonic interconnect between two separate commercial trapped-ion quantum systems, demonstrating quantum entanglement between two independent computers at a distance. The same day, the company was selected for DARPA's HARQ program to build multi-qubit heterogeneous quantum networks — sending its stock surging nearly 20%.

Bank of America analyst Vivek Arya raised his 2026 global semiconductor revenue target to $1.3 trillion—a $300 billion upward revision made just four months after the prior estimate—citing AI-driven surges in compute and storage demand. The bank names Nvidia, Broadcom, Marvell, AMD, Lam Research, and KLA as the top beneficiaries, while warning that cloud capex must surpass $1 trillion by 2027 to sustain the current growth trajectory.

Bloomberg's Mark Gurman revealed Apple is testing at least four frame styles for AI-powered smart glasses set for a 2027 launch, powered by a custom N401 chip and dual cameras. The no-display device positions Apple squarely against Meta's wildly popular Ray-Ban smart glasses, while a simultaneous leadership shake-up sees AI chief John Giannandrea retire.

Framework Computer has announced a live hardware launch event for April 21, hinting at major modular laptop and desktop upgrades alongside a significant Linux-focused announcement. CEO Nirav Patel has published a stark warning: the AI boom's winner-takes-all dynamic over chips and storage could drive consumers toward cloud-leased, locked-down devices — and Framework is positioning its Next Gen lineup as a defense of user ownership.

South Korean AI chip maker Rebellions has closed a $400 million pre-IPO round at a $2.34 billion valuation, backed by Samsung, SK Hynix, and Saudi Aramco. Alongside the funding, the company unveiled RebelRack and RebelPOD — rack-scale inference systems built around its Rebel100 NPU, which the company claims matches Nvidia H200 performance at 3.2× better power efficiency. A new partnership with SK Telecom and Arm targets sovereign AI and global telecom infrastructure.

Google and Intel announced an expanded multiyear partnership on April 9 that commits Google Cloud to multiple generations of Intel Xeon 6 processors for AI workloads while deepening joint development of custom infrastructure processing units. The deal gives Intel a meaningful foothold in a data-center market long dominated by Nvidia and signals that CPUs are returning to relevance in AI infrastructure.

Taiwan Semiconductor Manufacturing Company reported first-quarter 2026 revenue of $35.6 billion, a 35% year-on-year surge that beat analyst consensus, driven by insatiable demand for AI accelerators and the company's monopoly on advanced packaging technology. TSMC shares crossed 2,000 New Taiwan dollars for the first time, and the full earnings call is set for April 16.

Apple is advancing development of its first custom AI server chip, codenamed Baltra, while Samsung Electro-Mechanics has begun supplying experimental glass substrate samples for packaging. The collaboration marks Apple as the last major tech hyperscaler to build its own AI silicon, while the glass substrate technology represents a potential generational leap in chip packaging that could reshape AI hardware economics in the 2027–2028 timeframe.

Global semiconductor sales hit $88.8 billion in February 2026, up 61.8% year-over-year and 7.6% month-over-month, according to industry data. With AI accelerators on pace to account for roughly half of all semiconductor revenues this year, the industry is tracking toward its first $1 trillion annual revenue milestone—even as critical bottlenecks in high-bandwidth memory and advanced packaging constrain how fast AI infrastructure can actually scale.

NVIDIA's next-generation Vera Rubin NVL72 AI supercomputer has entered full production ahead of its originally guided H2 2026 schedule. The rack-scale system delivers 3.6 exaflops of inference compute and up to 5x greater inference performance and 10x lower cost per token compared to Blackwell, with AWS, Google Cloud, Microsoft Azure, and CoreWeave among the first deployment wave.

Nvidia took a $2B stake in Marvell and launched NVLink Fusion, letting third-party AI accelerators plug into Nvidia's interconnect fabric. The deal signals a pivot: long-term dominance may rest on owning the networking layer, not GPU monopoly.

Intel will repurchase Apollo Global Management's 49% stake in its Leixlip, Ireland fabrication plant for $14.2 billion — paying a $3 billion premium over what Apollo paid just 18 months ago. The deal hands Intel full ownership of its most advanced EUV fab as the company positions itself to compete for foundry customers against TSMC.

Intel is buying back Apollo Global Management's 49% stake in its Fab 34 facility in Leixlip, Ireland for $14.2 billion — a 27% premium over what Apollo paid in 2024. The deal signals renewed confidence in Intel's turnaround under CEO Lip-Bu Tan and strong demand for CPU and AI chip production.

After years of being NVIDIA's also-ran, AMD's AI GPU is landing real enterprise contracts. The CUDA moat is real but it's eroding, and the price difference is becoming impossible to ignore.

While everyone fights over NVIDIA and AMD GPUs, a handful of startups are building AI chips on the open RISC-V architecture. They're years from maturity — but the implications for AI sovereignty are enormous.

The GPU shortage isn't just an NVIDIA problem — it's reshaping the entire AI industry's power structure. Companies that secured allocation early are pulling away. Everyone else is scrambling.