The class of 2026 is entering a job market that has changed shape around them while they were studying. Unemployment among recent graduates aged 22 to 27 climbed to 5.7%, well above the 4.2% rate for all workers. Nearly 43% of new graduates are now underemployed, the highest rate since the pandemic. Entry-level job postings in the US are down 35% since early 2023. This is not a recession story. Companies are not struggling. They are finding that one person with AI tools can do what previously required three. The writing tasks, the research summaries, the first-draft spreadsheet models, the initial client decks were the apprenticeship. They are now automated. In the UK, tech graduate roles fell 46% in 2024, with a further 53% projected drop by 2026. In the US, junior software development postings are down 67% from their peak. The fields still hiring share one characteristic: they require judgment, physical presence, or human accountability AI cannot replicate. Among employers, 35% of entry-level postings now require demonstrated AI skills, nearly double the share from a year ago. The "Founding Engineer" job title is up 390% as a role for new graduates, a signal that some portion of the entry-level cohort is creating roles rather than waiting for them.

For two years, the AI investment story has rested on a quiet assumption: that interest rates would fall, making it cheap to borrow the capital needed to build data centres, buy chips, and fund the infrastructure AI requires. On Friday that assumption took a serious hit. The US government reported employers added 172,000 jobs in May, more than double what economists had expected. What followed was the worst day for tech stocks since April 2025. The Nasdaq fell 4.2%. The Philadelphia Semiconductor Index plunged over 10%, its worst session since March 2020. Nvidia fell 6%. Micron fell 13%. Marvell collapsed 17%. AMD fell 11%. Well over $1 trillion was erased from equity markets in a single session. Alphabet, Microsoft, Meta and Amazon are collectively spending over $750 billion on AI infrastructure this year. Most of that is financed through borrowing. Every quarter-point rise in interest rates adds billions to the cost of that financing. The 10-year Treasury yield jumped above 4.5%. The 30-year crossed 5%. Markets are now pricing in a meaningful probability of a rate hike before end of 2026. Investors sold AI stocks and rotated into Coca-Cola, Colgate and Johnson and Johnson.

Broadcom reported Q2 2026 revenue of $22.2 billion, up 48% year over year. AI chip revenue grew 143% in a single year. Free cash flow hit a record $10.26 billion. Earnings per share beat estimates. By any conventional measure, an outstanding quarter. The stock fell 12% anyway. The stock was not priced for a great quarter. It was priced for a perfect one. Wall Street had modelled AI chip guidance of $17.2 billion for the next quarter. Broadcom guided $16 billion. It left its full-year AI revenue target unchanged at $56 billion rather than raising it. CEO Hock Tan confirmed that Google plans to diversify its chip suppliers rather than relying exclusively on Broadcom. And he acknowledged that AI chip revenue carries lower profit margins than Broadcom's software business, meaning more AI revenue does not automatically mean more profit. The stock's reaction triggered a broader sell-off across chip stocks.

Apple's Worldwide Developers Conference opens June 8 and it may be the most consequential keynote the company has held in a decade. Two years ago Apple promised a smarter, more capable Siri. Those features never arrived. Apple agreed to a proposed $250 million settlement with iPhone buyers who accused the company of false advertising after the AI Siri features promoted during the iPhone 16 launch remained unavailable for nearly two years. Apple finalised a deal in January 2026 to pay Google approximately $1 billion per year for a custom Gemini model that will power a completely redesigned Siri. The new Siri will live in the Dynamic Island with a standalone chatbot app supporting back-and-forth conversation, file access, on-screen awareness, and actions across all apps. For the first time, Apple will allow users to set third-party AI services as the default for Apple Intelligence features. Morgan Stanley maintains a bull case of $440 on Apple stock predicated on an AI re-rating. Wedbush has a Street-high $400 target.

For four decades, every Windows laptop and desktop has run on an x86 processor made by either Intel or AMD. At Computex 2026 in Taipei, Nvidia CEO Jensen Huang unveiled a new PC processor made alongside Microsoft, entering an arena that has long been ruled by Intel, AMD, Qualcomm and Apple. The chip, called RTX Spark, will power a new line of Windows laptops from Dell, HP, Lenovo, ASUS and MSI arriving in fall 2026. It is not built on x86. Arm-based processors like Nvidia's are gaining ground over traditional x86 chips, while the overall CPU market is growing into what Jensen Huang says will be a $200 billion industry. The RTX Spark combines CPU cores with Nvidia's Blackwell AI processor and up to 128 gigabytes of unified memory, designed to run AI agents locally on device without sending data to the cloud. Huang described this as the PC industry's first major transformation in four decades. Intel is responding with its Wildcat Lake chips. AMD has not yet announced a direct counter.

For two years, AI agents have been one of the most talked-about concepts in technology and one of the least actually deployed. A chatbot answers your questions. An agent takes actions on your behalf without being asked at every step. It books the meeting, writes and tests the code, files the report, escalates the issue. The gap between that description and reality has been wide. That gap is closing. According to the 2026 Gartner Hype Cycle for Agentic AI, only 17% of organisations have deployed AI agents to date, yet more than 60% expect to do so within two years, the most aggressive adoption curve Gartner has recorded for any emerging technology. Gartner also forecasts that by end of 2026, 40% of enterprise applications will contain task-specific AI agents, up from less than 5% in 2025. This week the physical infrastructure for that shift arrived. Nvidia hand-delivered its first Vera CPUs to Anthropic, OpenAI, SpaceX and Oracle. Google's Gemini Spark runs across Gmail, Calendar and Drive without being invoked. Coding tools from Cursor, Claude Code and Codex are autonomously writing, testing and submitting code.

Two years ago, AI coding tools were clever autocomplete. Today they plan, write, test, debug, and submit code for review. They work through the night. They do not need onboarding. They do not ask for a salary review. Claude Code leads benchmark performance at 80.8% on SWE-bench Verified, the industry's standard software engineering test. Codex from OpenAI comes second at 71%. Cursor leads on speed and user experience with $2 billion in ARR and 67% of Fortune 500 companies deploying it. Windsurf was at the centre of one of the most dramatic acquisition battles of 2026. OpenAI attempted a $3 billion deal that collapsed when Microsoft demanded IP rights. Google then moved with a $2.4 billion deal to hire the founding team. Replit remains the dominant platform for education and beginners. The AI coding agent market doubled in size between mid-2025 and early 2026. Companies are no longer hiring cohorts of junior developers to do the work these tools now handle.

In almost every category this briefing covers, AI is reducing costs: fewer junior hires, more automated workflows, leaner teams. Cybersecurity is the exception. CrowdStrike reported fiscal Q1 results with annual recurring revenue up 24% to $5.5 billion and record net new ARR growth of $256 million, up 32% year over year. Palo Alto Networks reported a 31% revenue increase to $3 billion in the same period. Both beat estimates. CrowdStrike CEO George Kurtz explained it directly: AI has entered cybersecurity across two dimensions. You need cybersecurity to secure AI itself, because deploying AI across the enterprise without it is simply too risky. And accelerated AI use has created an explosion in new attack surfaces. Every AI agent deployed in an enterprise is a new entry point. Every automated workflow is a potential vulnerability. Wedbush Securities argues the proliferation of AI agents creates entirely new attack vectors that legacy security tools were never designed to address, accelerating consolidation toward platform players.

For thirty years, the central processing unit was the brain of every computer. Then AI arrived and the GPU became the engine of the new era, because GPUs are extraordinarily efficient at the parallel mathematical calculations AI models require. Now a third phase is beginning. Nvidia announced its Vera CPU, its first in-house processor designed specifically for AI agents, is now in full production. Agents plan sequences of actions, remember context across sessions, call external tools, manage orchestration across multiple systems, and operate continuously without human instruction. That pattern of work is better suited to CPUs than GPUs. Nvidia's VP of Hyperscale Ian Buck hand-delivered the first Vera systems personally to Anthropic in San Francisco, OpenAI in Mission Bay, SpaceXAI in Palo Alto, and Oracle in Santa Clara. The chip runs 1.8 times faster than Intel and AMD for AI agent tasks and is the first processor Nvidia has designed entirely from scratch. The total market for server CPUs could reach $211 billion by 2030, driven primarily by agentic AI demand.

This is the most consequential economic argument of 2026, and it does not have a settled answer. The first position is called the Jevons Paradox, named after a 19th-century economist who noticed something counterintuitive: when steam engines made coal more efficient to burn, Britain did not use less coal. It used dramatically more, because cheaper energy enabled entirely new industries. Applied to AI: cheaper legal services, financial analysis, and software development will not reduce demand for those services. It will expand the market because they become affordable to people who could never previously access them. Apollo Global Management's chief economist Torsten Slok made this case in April: AI lowers the price of knowledge work, which expands the total market for knowledge work. The second position comes from PwC's 2025 Global AI Jobs Barometer. AI-augmented financial roles show fourfold productivity growth, meaning firms need one quarter of the headcount to produce the same output. Both datasets are real. Graduate unemployment is at 5.7%. Entry-level postings are down 35%. And yet overall employment is strong enough to surprise economists by 100% in a single month.

Anthropic confidentially filed its S-1 with the SEC on June 1, 2026, edging ahead of rival OpenAI in the race to public markets. The filing comes weeks after the company closed a $65 billion funding round at a post-money valuation of $965 billion, the fastest valuation doubling in startup history, rising from $380 billion in February to nearly $1 trillion in May. Backers in the most recent round include Blackstone, Brookfield, GIC, General Catalyst and Insight Partners. The company's annualised revenue run rate crossed $30 billion in April 2026, up from $9 billion at the end of 2025. Its 80% enterprise revenue mix is what separates it from OpenAI in how investors will value it. The IPO pipeline building behind this filing includes OpenAI and SpaceX, which together could raise more capital than all US public listings since 2022 combined.

Alphabet announced it will raise up to $80 billion through equity offerings to fund AI infrastructure expansion. The raise comprises a $40 billion at-the-market share programme, $30 billion in underwritten public offerings, and a $10 billion investment from Berkshire Hathaway. This is notable not because Alphabet needs the money. It has one of the deepest balance sheets in corporate history. It is notable because even that is not enough. Total AI capital expenditure across all major tech companies is expected to hit $750 billion in 2026 alone. Google's own AI capex for the year is projected at $180 to $190 billion. The AI infrastructure build-out has crossed the scale at which even the wealthiest companies in history are tapping public markets to keep pace.

At Emmanuel Macron's Choose France summit on May 30, SoftBank founder Masayoshi Son announced a commitment to invest up to 75 billion euros, approximately $87 billion, to build 5 gigawatts of AI data centre capacity in France. The first phase, 45 billion euros for 3.1 gigawatts in the Hauts-de-France region by 2031, is already underway. Sites include Dunkirk, Bosquel and Bouchain, the latter on the grounds of a former coal power station being redeveloped with state-owned nuclear operator EDF as a partner. Schneider Electric is building an AI infrastructure and robotics manufacturing hub at the Dunkirk site. Son told reporters: There is no choice. The US is going fast, China is going fast. Europe and Japan have to also go fast, not to be left out. SoftBank's net profit quadrupled to over $32 billion in the year to May 2026, driven largely by AI-related investments.

On June 1, GitHub retired its flat-rate subscription model for 4.7 million paid Copilot subscribers and replaced it with token-based AI Credits. Under the old model, a $29 per month subscription gave developers effectively unlimited access to AI coding assistance. Under the new model, the same $29 is a credit allowance. When it runs out, you pay more or stop. Developers using agentic features are projecting bills of $750 to $3,000 per month for the same usage patterns. Internal Microsoft documents showed Copilot's weekly running cost had nearly doubled between January and June 2026, driven by developers using expensive reasoning models for complex tasks. The flat subscription model was silently subsidising that usage. The per-token model ends the subsidy. The broader implication goes beyond Copilot. Every major AI tool currently on flat-rate pricing is running the same subsidy model. When the economics break, they all face the same choice GitHub just made.

Intel stock opened 2026 at $36.90. By early June it trades near $130, a gain of roughly 250% in five months. The reason is not a turnaround in Intel's traditional PC business. It is a structural shift in how AI data centres are being built. For the past three years, the AI build-out was dominated by GPU training and Nvidia owned that market. But as the industry moves from training large models to running them constantly for inference and agentic tasks, the CPU-to-GPU ratio in data centres is shifting toward parity. AI agents that act continuously need central processors handling orchestration, memory management and workflow coordination. Intel's Xeon server CPUs are sold out for all of 2026. Server CPU prices have risen 10 to 20%. Apple signed a preliminary agreement with Intel to manufacture chips for US devices, validating Intel's foundry ambitions under the CHIPS Act. AMD has similarly tripled in 2026 for the same structural reasons.

AMD reported Q1 2026 revenue of $10.3 billion, up 38% year over year and ahead of the $9.89 billion consensus estimate. Data Centre segment revenue hit $5.8 billion, up 57% year over year, driven by EPYC server CPUs and Instinct AI accelerators. Free cash flow tripled to $2.6 billion, a quarterly record. AMD guided Q2 revenue at $11.2 billion, well ahead of the Street's $10.5 billion estimate. CEO Lisa Su said data centre is now the primary driver of revenue and earnings growth, with strong confidence in reaching tens of billions of dollars in data centre AI revenue in coming years. AMD stock has more than tripled in 2026. The company is no longer a distant Nvidia challenger. It is a structural beneficiary of an AI infrastructure cycle that requires every component of a data centre to be upgraded simultaneously.

Data centres now consume approximately 6% of total US electricity, roughly equivalent to the entire annual electricity consumption of Pakistan. A single hyperscale facility draws as much power as 100,000 homes. Annual global data centre spending is approaching $1 trillion, with up to $700 billion anticipated in the US alone in 2026. The scale of this build-out is colliding with organised local resistance. Ohio's governor Mike DeWine suspended the state's data centre tax incentive programme this week as projected exemption costs surged sharply. Local groups are gathering 413,000 signatures for a November ballot measure that would ban data centres with a peak load over 25 megawatts statewide. At least a dozen Ohio municipalities have imposed temporary moratoriums. Residents are not opposed to technology. They are opposed to paying higher electricity bills, absorbing water consumption that strains local supplies, and hosting massive industrial facilities that employ almost nobody locally. The pattern is repeating in Virginia, Indiana, Georgia and Arizona.

As of May 2026, Google, Microsoft, Amazon and Meta have all committed to nuclear power for their AI data centres. Thirteen announced deals. 9.8 gigawatts of nuclear capacity committed. The reason is straightforward. AI data centres need power that is always on, at massive scale, with zero carbon emissions. Solar and wind are intermittent. They produce nothing when the sun is down or the wind is calm. Nuclear runs at full capacity 24 hours a day, 365 days a year, regardless of weather. A single large nuclear plant can power multiple hyperscale data centres indefinitely. Microsoft signed a 20-year, $1.6 billion agreement with Constellation Energy to restart the Three Mile Island reactor in Pennsylvania, dormant since 2019, with first power delivery expected in 2027. Meta committed to up to 6.6 gigawatts across TerraPower, Oklo, Vistra and Constellation, the largest single corporate nuclear commitment in history. Google committed 500 megawatts from Kairos Power. Amazon invested $700 million in X-energy for small modular reactors.

Wikipedia is written entirely by unpaid volunteers. Roughly 250,000 active editors globally make approximately 10 million edits per year, covering 60 million articles across 300 languages. The Wikimedia Foundation is the nonprofit that employs around 700 paid staff to build and maintain the technical tools those volunteers use. On May 21, 2026, the Foundation dissolved its six-person Community Tech team, laying off five engineers and one manager. That team fixed bugs, built moderation tools, and responded to editor requests. Within days, over 800 editors signed a strike petition. They are threatening to stop editing, reduce anti-vandalism work, and redirect the donation banners that fund the Foundation to display critical messages instead. The anger has a specific source. The Foundation is generating growing revenue from licensing Wikipedia's content to AI companies for model training, while simultaneously cutting the engineers who support the volunteers who create that content. Several of the laid-off engineers were also reportedly involved in a nascent unionisation effort called Wiki Workers United.

The biggest story in venture capital in 2026 is not which AI app got funded. It is the structural rotation away from software entirely. Cerebras Systems, maker of a wafer-scale AI processor that runs inference at speeds 7,000 times faster than a standard GPU, raised $5.5 billion in its IPO in May. Shares opened at $385 against an IPO price of $185, valuing the company at $66 billion on its first day of trading, the largest tech IPO of 2026. Eclipse Ventures-backed companies alone have raised $14 billion this year. Anduril, building autonomous defence systems and AI-powered surveillance infrastructure, raised $5 billion at a $61 billion valuation. Mind Robotics, spun out of Rivian, raised $400 million to deploy AI robots on factory floors. The thesis driving all of it is the same. AI has made software easy to build and nearly impossible to defend as a business. Any application can be cloned in weeks using the same foundation models. The defensible businesses are the physical ones, chips, robots, power infrastructure, defence systems, that take years and billions to replicate and cannot be prompted into existence.

Announced at Google I/O 2026, Gemini Spark is a background AI agent built on Gemini 3.5 and Google's Antigravity platform. Unlike a chatbot you open and close, Spark runs continuously on Google Cloud virtual machines — even when your phone or laptop is off. It connects natively to Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube and Google Maps, and supports third-party apps via the Model Context Protocol. Users can email or text Spark directly like a human colleague. Currently in beta for Google AI Ultra subscribers in the US.

Microsoft reported Q3 FY2026 earnings with its AI business annualising at $37 billion — up 123% year over year. Paid Copilot seats crossed 20 million, up from 15 million in January, with year-over-year seat adds growing 250%. CEO Satya Nadella said weekly Copilot engagement is now at the same level as Outlook. Accenture has over 740,000 seats — the largest single deployment — while Bayer, J&J, Mercedes and Roche each committed to 90,000 or more seats this quarter.

OpenAI launched advertising inside ChatGPT on February 9, restricted to brands with a $200,000 minimum spend. On May 5 it opened a self-serve Ads Manager to all US advertisers with no minimum. Agency partners include Dentsu, Omnicom, Publicis and WPP. Ads appear in labelled boxes below AI responses and are not shown to paid subscribers. OpenAI is targeting $2.5 billion in ad revenue for 2026 with a stated ambition of $100 billion by 2030.

Bloomberg's Mark Gurman confirmed Apple is completely rebuilding Siri in iOS 27. The new Siri — code-named Campos — replaces the existing interface entirely, living in the Dynamic Island with a standalone app that functions like a modern AI chatbot. It supports back-and-forth conversation, file uploads and conversation history. Complex queries route to Google Gemini; simpler tasks run Apple's on-device model. To be announced at WWDC on June 8 and ship with iOS 27 in September.

Figure AI's Figure 03 units are commercially deployed on BMW's Spartanburg line — 40 robots billing at roughly $25 per operating hour. Tesla has deployed over 1,000 Optimus Gen 3 units across Gigafactory Texas and Fremont for parts processing and kitting, and is converting portions of Fremont to humanoid robot production. Boston Dynamics' Atlas is in enterprise pilot testing at Hyundai's Georgia facility.

TSMC's CoWoS stacks high-bandwidth memory directly on top of AI accelerators. Without it, even the most advanced chips cannot become functional AI products. TSMC's CoWoS capacity is growing at 80% CAGR but Nvidia has reserved the majority of available capacity. New facilities in Arizona and Taiwan are under construction but lead times are measured in years. The constraint is expected to persist through 2027.

Cursor reached $2 billion in annualised recurring revenue in February 2026 — up from $1 billion in November 2025 and $500 million in June 2025. No B2B software company has grown from zero to $2 billion ARR faster. Enterprise customers account for 60% of revenue, with 67% of the Fortune 500 deploying it including Nvidia, Uber and Adobe. The company is in talks to raise $2 billion at a $50 billion valuation.

The European Commission opened formal infringement proceedings against X regarding Grok on January 26, 2026, under the Digital Services Act. The investigation focuses on whether X properly assessed risks from Grok generating sexually explicit fake images of real people, including minors. Regulators in at least eight countries have confirmed action against X and xAI, including Ofcom under the UK Online Safety Act and the Irish DPC under GDPR.

Salesforce cut under 1,000 roles in February 2026 across marketing, product management, data analytics and the Agentforce team. CEO Marc Benioff confirmed AI now handles 30–50% of work at Salesforce and the company hired no new engineers in fiscal 2026. Workday cut 400 roles the same week. Meta, Amazon, Google and Atlassian all announced significant cuts in early 2026, explicitly linking them to AI productivity gains.

Stanford's Digital Economy Study found employment for software developers aged 22–25 fell nearly 20% from its late 2022 peak by July 2025. Indeed's Hiring Lab shows US software engineer postings down 36% from February 2020, with entry-level hiring down 25% year over year in 2024. New postings dropped a further 15% in early 2026. Stack Overflow's 2025 Developer Survey found 84% of developers now use AI tools. A SHRM survey found 70% of hiring managers say AI can do the work of interns.

Micron's CFO confirmed every unit of high-bandwidth memory the company will produce in 2026 is already committed to customers — primarily Nvidia, AMD and a small number of hyperscalers. Buyers who did not secure allocations earlier are now being told to wait until Q1 2027. HBM is the specialised memory stacked directly on top of AI accelerators. Without it, the most powerful chips cannot function. The constraint is not fabrication — it is the extraordinarily complex bonding and stacking process that only a handful of facilities in the world can perform.

SK Hynix announced construction of a new $7 billion fabrication facility in Cheongju, South Korea, dedicated entirely to HBM4. First commercial yield is not expected until late 2027. HBM4 offers roughly double the bandwidth of current HBM3e and will be essential for the next wave of AI accelerators from Nvidia and AMD, both already in allocation discussions with SK Hynix.

Samsung's latest attempt to qualify its HBM3e chips for Nvidia's H200 and B100 accelerators has failed for the second consecutive quarter. The failure relates primarily to power consumption and heat dissipation at sustained loads. Samsung represents roughly 30% of global HBM production capacity — and that capacity is effectively unavailable to the market's largest buyer until the qualification problem is resolved.

Anthropic closed a $3.5 billion funding round led by sovereign wealth funds including Saudi Arabia's Public Investment Fund and Abu Dhabi's Mubadala. Amazon, which has already committed $4 billion as part of a cloud partnership, did not participate. The capital will primarily fund compute — training runs for Anthropic's next frontier model are estimated at $500 million to $1 billion. Anthropic is the only major AI lab that has publicly committed to not pursuing AGI without safety guarantees in place.

Nvidia briefly crossed a $4 trillion market capitalisation — exceeding the GDP of Germany. The milestone followed quarterly revenue of $44 billion, up 78% year over year, driven almost entirely by data centre GPU sales. Nvidia now captures an estimated 85% of the global AI accelerator market by revenue. The H100 and H200 remain dominant and the next-generation Blackwell architecture is selling out as fast as it ships.

ServiceNow crossed $50 billion in annual recurring revenue — the first SaaS company in history to reach that milestone. Growth is driven almost entirely by Now Assist, its AI layer that automates IT service management, HR workflows and customer operations. The company raised full-year guidance for the third consecutive quarter. CEO Bill McDermott attributed the acceleration to AI monetisation at enterprise scale.

Goldman Sachs published an updated labour market analysis projecting AI will automate the equivalent of 300 million full-time jobs globally over the next decade — up from its 2023 estimate of 200 million. The revision is driven by faster-than-expected adoption of agentic AI in white-collar work, particularly in legal, financial analysis, software development and customer service.

Internal data leaked to The Information shows GitHub Copilot lost approximately 40% of its enterprise accounts in Q1 2026, the majority migrating to Cursor. Losses are concentrated among companies with more than 500 engineers. Microsoft has responded by announcing Copilot will be deeply embedded into VS Code — effectively conceding the standalone product market.

The US Senate passed the AI Transparency Act in a 67-33 bipartisan vote. The bill requires any AI system used in consequential decisions — hiring, lending, healthcare, criminal justice — to disclose that AI was involved and provide a mechanism for human review. It does not regulate the models themselves, only their deployment. The bill now goes to the House.

Meta released Llama 4 under an open-source licence. Independent benchmarks show it outperforms GPT-4o on most standard evaluations and matches Claude 3.5 Sonnet on reasoning tasks. A frontier-level model being freely available means any company, researcher or developer can now run this capability without API fees — putting direct pressure on OpenAI and Anthropic to justify their pricing.