Tech and Internet Desk · Weekly Dispatch
Anthropic files for IPO at a near-trillion-dollar valuation. The H200 chip export reversal reshuffles the US-China chip war. Russia blocks Telegram and drives users to a state messenger nobody trusts. Thailand absorbs 29 billion dollars of data-center investment in a single BOI session.
Weekly Intelligence Brief | Analyst Desk | 2026-06-26
The defining AI story this week is not a model release; it is money changing hands at a scale that has no precedent in technology history. On 28 May, Anthropic closed a 65 billion dollar Series H funding round at a 965 billion dollar valuation, then filed a confidential S-1 with the SEC on 1 June to go public. Its revenue run rate is reported at 47 billion dollars annually, up from 10 billion a year ago, driven almost entirely by Claude Code. The combined hyperscaler capex commitment for 2026 now sits between 700 and 725 billion dollars across Amazon, Alphabet, Microsoft and Meta. That is more than the annual economic output of Switzerland. Electricity supply, not chip supply, is the constraint that is actually biting.
The US-China chip war shifted posture this week. The Trump administration reversed the ban on Nvidia H200 exports to China in January 2026, replacing a flat prohibition with a case-by-case licensing framework and a 25 percent tariff. Chinese firms including Alibaba, Tencent and ByteDance have been cleared to buy up to 75,000 units each, and pre-orders reportedly total more than two million units at 27,000 dollars apiece, a 54 billion dollar order book. Congress is pushing back hard. The AI OVERWATCH Act, which passed the House Foreign Affairs Committee 42-2 in January, would give both chambers a 30-day veto on any chip export to a foreign adversary and impose a two-year ban on Nvidia's Blackwell-tier chips entirely. The bill has not yet passed the full chamber. Meanwhile, Huawei's Ascend 910C is targeting 600,000 units of production this year, and HBM, the fast memory AI chips require, remains the ceiling China cannot yet clear domestically.
Russia's internet crackdown intensified further this spring. Telegram availability for users without a VPN fell to 5 percent by early April 2026, WhatsApp was cut two days after Telegram, and Roskomnadzor has now blocked 469 VPN services. The state is pushing Max, a messenger built on VK's platform, as the official replacement, but technical audits found the Android version probing for VPN activity on users' devices. Roughly 40 percent of Russian internet users now rely on VPNs despite the crackdown. Thailand sits at the other end of the spectrum: its Board of Investment cleared 29 billion dollars in digital infrastructure projects in a single session, led by TikTok's 25 billion dollar data-center expansion. A parallel controversy over a 1.62 billion baht sovereign AI contract continues despite the ministry's claim that the bidding was competitive.
North Korean hackers' running total through April 2026 holds at 577 million dollars from two attacks, still 76 percent of all crypto stolen globally this year. Israel's tech fundraising reached 8.6 billion dollars in the first half of 2026, up 45 percent, with cybersecurity investment more than doubling. One Israeli cyber startup, whose AI-powered security platform tripled its valuation to 9 billion dollars in a year, captures the dynamic: AI and defence are the two forces pulling capital in. Uzbekistan held a digital services meetup in Chicago on 25 June, pitching the IT Park export story to the diaspora and US tech buyers. Argentina's non-human corporation proposal drew formal opposition from a former lawmaker who called it 'the dissolution of all moral and legal constraints.' The bill has not passed.
| Lane / Country | The one-line read |
|---|---|
| AI (global) | Anthropic files S-1 at near-1 trillion dollar valuation. Combined hyperscaler capex 700-725 billion dollars in 2026. Power, not chips, is the binding limit. |
| Semiconductors | H200 ban reversed; case-by-case licensing plus 25% tariff. TSMC Q2 guidance 39-40.2 billion dollars. Huawei Ascend 910C targeting 600k units. HBM still China's ceiling. |
| Cyber warfare | North Korea: 577 million dollars stolen through April, 76% of all 2026 losses. Volt and Salt Typhoon still inside US infrastructure. Russia cuts Telegram and WhatsApp. |
| Splinternet | Russia blocks 469 VPNs; Telegram near-zero without VPN. China's AI-upgraded Firewall being exported. EU AI Act high-risk deadline deferred to Dec 2027. |
| United States | Anthropic S-1 filed. AI OVERWATCH Act passed committee 42-2. H200 reversal triggers congressional backlash. |
| China | Huawei Ascend 910C scaling to 600k units. H200 orders placed but HBM bottleneck persists. Great Firewall gets AI-driven upgrade. |
| Israel | H1 2026 fundraising 8.6 billion dollars, up 45%. Cyber investment more than doubled. AI valuation multiples rising sharply. |
| EU + Czechia | AI Act high-risk deadline now Dec 2027 (deferred). DSA enforcement live. Czech AI funding up 174% year-on-year through mid-2025. |
| Thailand | BOI approves 29 billion dollars in digital projects, led by TikTok 25 billion dollar data-center build. TH-AI Passport contract controversy continues. |
| Russia | Telegram near-zero; WhatsApp blocked; 469 VPNs banned. State messenger Max surveils users. 40% of Russians now use VPNs. |
| Uzbekistan | IT Park Q1 2026 exports 191.8 million dollars, up 38.5 million year-on-year. Digital services meetup held in Chicago 25 June. |
| Argentina | Non-human corporation proposal in draft law stage; no vote yet. Opposition calls it catastrophic for human dignity. Milei reaffirms AI deregulation pitch. |
Snapshot as of 26 June 2026. Each lane and country is covered in full below.
FRONTIER MODELS / COMPUTE ARMS RACE / POWER
As of late June 2026, four labs occupy a tight cluster at the top of every meaningful benchmark. Anthropic's Claude Opus 4.8 (released 28 May) scores 88.6 percent on SWE-bench Verified, the standard coding test, and leads the Artificial Analysis Intelligence Index at 61.4 points. OpenAI's GPT-5.5, released in April, follows at 60.2 on the same index and leads on creative writing and agentic computer-use tasks, including scoring 82.7 percent on Terminal-Bench 2.0 and 78.7 percent on OSWorld-Verified. Google's Gemini 3.5 Flash was announced in early June with 'frontier performance for agents and coding.' xAI's Grok 4, the latest in the Grok series trained on the Colossus cluster, is positioned as a flagship reasoning model. Microsoft unveiled its own coding-focused models in June, reducing developer reliance on OpenAI.
On the Arena Elo ratings, the competitive ranking where real users vote on which model answers better, Anthropic stands at 1,503, xAI at 1,495, Google at 1,494, OpenAI at 1,481, Alibaba at 1,449, and DeepSeek at 1,424. The gap between the top four is under 30 points; the gap between the US leaders and China's DeepSeek is 79 points, about 5 percent. Two years ago that gap was effectively uncrossable. For a non-specialist: the 'best AI' question has no clean answer because the answer depends on the task. The competitive edge has shifted from raw benchmark position to cost per useful output and reliability in long-running automated work.
Anthropic confidentially submitted a draft S-1 registration statement to the SEC on 1 June 2026, four days after closing a 65 billion dollar Series H funding round at a 965 billion dollar post-money valuation. That valuation briefly exceeded OpenAI's, making Anthropic the highest-valued private AI company in the world. The company reported an annual revenue run rate of 47 billion dollars in May, up from 10 billion dollars a year earlier, and told investors it expects to surpass 50 billion dollars in annualised run-rate revenue by end of June 2026. Morgan Stanley, Goldman Sachs and JPMorgan Chase are the lead underwriters. No listing date, share price or share count has been set.
To put the valuation in context: 965 billion dollars is approaching Apple's market cap at the time of Apple's iPhone launch era. Anthropic earns those revenues primarily through Claude Code, its AI coding assistant, and through enterprise API contracts. The IPO, when it comes, will be the first public-market test of how a frontier AI lab actually reports financials, which will set the precedent for how OpenAI and others are later valued.
US technology media treats the race as a two-horse contest between Anthropic and OpenAI. Chinese state press covers DeepSeek as proof that chip restrictions failed. The native-language delta: Chinese-language tech outlets (36Kr, Jiemian) have reported DeepSeek's internal cost disclosures with more scepticism than the English press, noting the figures are unaudited and omit compute spent on failed pretraining experiments. The same outlets note that DeepSeek's Arena Elo position, 79 points behind Anthropic, is still a meaningful gap on tasks that require precise technical output. Both framings are partially right.
The IPO cycle and the model-release cycle absorbed the week's attention. Moving underneath: the US government's AI safety review framework, operational since May, now requires frontier labs to submit their most powerful models for review before release. No model has been publicly blocked. But the mechanism exists, and Anthropic's S-1 filing will for the first time require it to disclose revenue, costs and safety-review obligations in a document accessible to regulators, investors and the public simultaneously. That is a structural governance shift the product-launch coverage largely missed.
The combined hyperscaler capex commitment for 2026 sits between 700 and 725 billion dollars across the four largest cloud companies. Amazon is projecting 200 billion dollars, Alphabet 175 to 185 billion, Microsoft approximately 190 billion for the calendar year, and Meta has raised its guidance to 125 to 145 billion dollars, citing higher memory-chip prices and additional data-center costs. The 77 percent increase over 2025's 410 billion dollar total is the largest single-year capex surge in the history of the technology industry.
To say whether 725 billion is a lot: the entire global semiconductor industry's annual revenue is roughly 600 billion dollars. The hyperscalers are spending more than that building the buildings and power infrastructure that run the chips. Amazon is projected to turn free cash flow negative in 2026 as a result of this spending. This is a structural bet on future AI-related revenue, not a payoff from current revenue, which independent analysts estimate at well under 100 billion dollars across all four companies.
Every major hyperscaler has now signed at least one nuclear power deal for AI data-center capacity. Across 13 announced projects, the combined commitment stands above 9.8 gigawatts of nuclear capacity. Microsoft locked in 835 megawatts via a 16 billion dollar, 20-year agreement for the Three Mile Island Unit 1 restart. Meta leads the pack with up to 6.6 gigawatts across TerraPower Natrium, Oklo Aurora, Vistra and Constellation. Amazon secured a deal with Talen Energy for 1,920 megawatts through 2042 and led a 500 million dollar financing round for X-energy's small modular reactors. The first nuclear electrons for an AI data center are expected in 2027.
One gigawatt is roughly the output of a large conventional power plant. What the tech sector has committed is equivalent to adding ten large plants to the grid over the next several years, specifically dedicated to AI infrastructure. Nuclear power, a fuel assumed to be in terminal decline five years ago, is now the primary long-term energy strategy for the companies running the world's AI.
The spending numbers are large enough to dominate the narrative. Less examined: the spending is running far ahead of what AI currently earns. When the same companies also hold trillions in government cloud contracts, the question of which products are subsidising which is legitimate and rarely asked in the financial press. Anthropic's S-1, when public, will be the first window into whether an AI lab's unit economics are actually viable at this capex intensity.
TSMC / EXPORT CONTROLS / CHINA SELF-SUFFICIENCY
TSMC reported Q1 2026 revenue of 35.9 billion dollars, up 40.6 percent year-on-year, with a gross margin of 66.2 percent and a net profit margin of 50.5 percent. High-performance computing, meaning AI chips, accounted for 61 percent of revenue. For Q2 2026, TSMC guided revenue of 39.0 to 40.2 billion dollars, which would represent another step-up. Full-year revenue is expected to grow above 30 percent in dollar terms, putting the 2026 total on course to exceed 160 billion dollars.
On the Arizona expansion: construction of the second fab (Fab 21 Phase 2) was completed in April 2026. Equipment installation begins in the July-to-September quarter of 2026, months ahead of the original schedule, with 3nm mass production targeted for the second half of 2027. A third fab using 2nm and next-generation process technologies broke ground in April 2025 and is being built as fast as possible. All four US TSMC fabs are now reported as fully booked by customers. The first fab (4nm process) turned its first annual profit in Q1 2026.
To put these numbers in context: a 4nm chip packs roughly 100 million transistors into an area the size of a pinhead. TSMC is the only company on earth doing this at commercial scale. Its top customers are Apple and Nvidia. Nvidia's Blackwell AI accelerators, which power most large AI models, come off TSMC lines. If TSMC stops, the global AI build-out stops. The second and third Arizona fabs are the start of a diversification strategy; when complete, roughly 30 percent of TSMC's most advanced chips will be made on US soil.
TSMC's dominance means the geopolitical risk is concentrated in one geographic location: Taiwan. Arizona is the diversification play. Even with all planned fabs complete, 70 percent of the most advanced chips will still be made in Taiwan, which is why every government with an AI strategy watches the Taiwan Strait. The Q2 guidance number, 39 to 40.2 billion dollars, is a reliable indicator that demand is holding; if that number fell, it would signal the entire AI build-out was slowing.
The strong earnings and accelerated Arizona timeline absorbed most coverage. Moving quietly: TSMC and the US Department of Commerce confirmed up to 6.6 billion dollars in proposed CHIPS Act direct funding for the Arizona complex. The funding is proposed, not disbursed; conditions attach. And a congressional standoff over the AI OVERWATCH Act (see below) creates some uncertainty about which customers will actually be able to buy the chips TSMC produces in the US.
The Trump administration reversed the Nvidia H200 export ban in January 2026, replacing a 'presumption of denial' with case-by-case review and a 25 percent tariff on each chip. The Commerce Department cleared roughly 10 Chinese firms, including Alibaba, Tencent and ByteDance, to purchase up to 75,000 H200 units each. Pre-orders reportedly exceed two million chips at 27,000 dollars apiece, an implied order book of about 54 billion dollars if fully executed. As of late June, shipments had not yet cleared Chinese customs in volume.
In parallel, the House Foreign Affairs Committee voted 42-2 in January to advance the AI OVERWATCH Act, which would give both chambers of Congress a 30-day window to block any chip export licence to a foreign adversary, impose a two-year ban on Nvidia's Blackwell architecture (one tier above H200) going to China, and close a loophole where new chips that approach but do not quite meet existing performance thresholds escape control. The bill has not passed the full House. The White House has not publicly endorsed it.
The H200 reversal is a significant policy shift that reflects the administration prioritising trade negotiation advantage over the chip-control strategy pursued from 2022 onward. The practical question is whether Chinese companies buy H200s in large volumes. If they do, Nvidia recovers a major market; if the chips sit in customs limbo for political reasons, the reversal achieves little commercially but signals diplomatic intent.
Huawei's Ascend 910C remains the default for Chinese cloud providers, hospitals and state institutions. Huawei is targeting 600,000 units of the 910C in 2026 and could distribute up to 1.6 million total Ascend dies across the product line. These chips run at 7nm process nodes, trailing TSMC's 4nm; performance is lower, but in a captive domestic market of 1.4 billion people, volume matters more than node generation. The hard ceiling: HBM (high bandwidth memory). China's chipmakers can produce the logic side of AI accelerators at scale but cannot yet produce HBM commercially. The Ascend 950DT, targeting Q4 2026, integrates HiZQ 2.0 HBM with 144 gigabytes and 4 terabytes per second bandwidth. Whether that memory is domestically produced or stockpiled foreign supply remains unconfirmed.
China's stated target is 70 percent domestic silicon wafer sourcing by end 2026. It set the same 70 percent target for 2025 and missed it. The HBM bottleneck is the single most important technical fact in the chip war and it rarely makes the front page. Without HBM at scale, China's AI chips run slower than their headline specs imply, and the gap with Nvidia's Blackwell tier is wider than node-generation numbers suggest.
STATE-SPONSORED OPERATIONS / NORTH KOREA CRYPTO / INFRASTRUCTURE
The 2026 ODNI Annual Threat Assessment confirms that China-linked groups Volt Typhoon and Salt Typhoon remain active inside US critical infrastructure. Volt Typhoon's documented goal, per NSA director testimony to Congress, is 'pre-positioning for disruption or destruction' in the event of a military conflict, not intelligence collection. The group has maintained persistent access inside US energy, water, communications and transportation networks for at least five years. Salt Typhoon is a separate operation focused on telecommunications espionage: it penetrated multiple major US carriers, stealing communications data including congressional and federal contracting records.
The Polish Internal Security Agency (ABW) warned in May 2026 that state-sponsored cyber attacks are 'shifting from espionage and data theft toward physical disruption of critical infrastructure.' CISA launched a new critical-infrastructure initiative in May, focused on proactive isolation and recovery planning. A Russian-speaking ransomware group attacked a German political party in March. State-sponsored and state-adjacent cyber activity is now effectively continuous across Europe and North America.
The 'pre-positioning for destruction' framing in US official statements is calibrated language. Calling it destruction rather than espionage activates different legal and military response authorities. The practical reality is that the implants are in the networks, access exists, and removing them from complex legacy telecom infrastructure is slow. CISA and the FBI continue to urge defensive action against exploitation of end-of-life edge devices: routers, firewalls, VPN gateways. These are the entry points because they are rarely monitored closely and hard to patch.
Attribution of cyber incidents is politically useful. Naming Chinese groups generates congressional support for defence budgets. This does not mean the threat is fabricated: the evidence base for Volt Typhoon is detailed and corroborated by Five Eyes partners including Australia and the UK. But the public framing consistently foregrounds Chinese actors while Russian groups' ongoing activity against European political and infrastructure targets receives less prominent coverage in US media.
North Korean-linked hackers stole approximately 577 million dollars in cryptocurrency through April 2026, capturing 76 percent of all crypto hack losses in that period, via just two incidents. The Drift Protocol breach on 1 April netted 285 million dollars after three weeks of staging and in-person social engineering with Drift employees. On-chain preparation reportedly began 11 March. The KelpDAO exploit on 18 April took 292 million dollars by compromising two internal RPC nodes and launching a denial-of-service campaign against external nodes, forcing the bridge's single verifier to rely on poisoned data sources. The Arbitrum Security Council froze roughly 75 million dollars of the stolen KelpDAO funds still on the network using emergency powers.
The 577 million dollar figure is large in absolute terms. Context: global crypto market cap sits around 3 to 3.5 trillion dollars, so this is about 0.02 percent of total holdings. But for a country with a GDP estimated at approximately 18 billion dollars, 577 million in four months is a revenue stream that funds weapons programmes under strict international sanctions. North Korea's cumulative crypto theft since 2017 now exceeds 6 billion dollars.
TRM Labs reports North Korean operators are now using AI tools in reconnaissance and social engineering phases. The operations are becoming more precise: fewer attacks, higher yield, more sophisticated laundering. The KelpDAO attack exploited a bridge protocol design flaw, not a conventional phishing attack. This is specialised financial crime at state scale, not opportunistic hacking.
RUSSIA RUNET / CHINA FIREWALL EXPORTS / EU DSA / THAILAND
Russia's sovereign internet law, codified in Government Decree No. 1667 effective 1 March 2026, gives Roskomnadzor legal authority to reroute or isolate internet traffic in real time. The weeks after the law took effect brought a rapid escalation. Telegram throttling began around 10 February 2026; by early April, Telegram availability for users without a VPN dropped to 5 percent. WhatsApp was blocked two days after Telegram. YouTube remains throttled. Roskomnadzor has blocked 469 VPN services as of February 2026. In mid-April, at the Ministry of Digital Development's instruction, more than 20 major Russian websites and e-commerce platforms, including Gosuslugi, Ozon, Wildberries, Yandex, Sberbank and VK, began blocking access for users connecting via VPN.
The state's alternative is Max, a messenger built on VK's platform and embedded into official Russian life by a Putin decree signed in June 2025. Max has not gained meaningful voluntary adoption. A technical audit of the Android version found it probing users' devices for VPN activity and sending traffic to third-party servers to check domain accessibility. Roughly 40 percent of Russian internet users now rely on VPNs, a dramatic increase, despite the crackdown. The state is both blocking VPNs and watching who uses them.
Russian state media frames the runet law as a 'stability and security' measure protecting citizens from external attack. Independent Russian-language outlets (Meduza, zona.media) frame it as censorship infrastructure formalising what was already happening informally. The technical architecture at its core, SORM deep-packet-inspection nodes installed on every ISP, has been in place since 2019. The 2026 decree and the subsequent blocking wave are the enforcement phase of an infrastructure built over seven years. The current VPN crackdown is not just about hiding content; it is about creating a surveillance record of who tried to circumvent the system.
The Telegram and WhatsApp blocks generated significant coverage in European media. Less covered: the same week Russia blocked Telegram, it signed agreements to expand SORM installation requirements to new categories of internet service provider, extending the surveillance architecture deeper into the network. The block is the visible story; the architecture expansion is the structural one.
Between 2025 and 2026, the Great Firewall added AI-driven traffic-classification layers: QUIC SNI inspection for nationwide encrypted connection analysis, DoH identification to block DNS-over-HTTPS servers, and fully encrypted traffic detection using heuristic rules. The result is that standard VPN protocols that worked inside China in 2023 are now blocked at a significantly higher rate; the Q2 2026 failure rate for common protocols has increased substantially. Separately, a leak of documents from Geedge Networks, linked to a co-architect of the Great Firewall, confirmed exports of 'secure gateway' censorship technology to Pakistan, Myanmar, Ethiopia and Kazakhstan. Myanmar reportedly used the hardware to monitor up to 81 million connections simultaneously.
China is not only maintaining its own firewall; it is selling the means to build equivalent systems elsewhere. The domestic and export products share the same underlying architecture, AI-driven content classification and deep-packet inspection built into network hardware. The buyer governments are all currently facing domestic protest movements or insurgencies, which contextualises the surveillance interest. The leak's authenticity has not been formally contested by Geedge or Chinese authorities.
The Geedge leak was reported primarily by Western investigative journalists and technology press. Chinese state media did not cover it. PBS NewsHour's broadcast segment on the leak had wider reach than the original written reports. Treat the four named buyer countries as unconfirmed in the absence of formal acknowledgment, though the pattern is consistent with prior Belt and Road digital-infrastructure deals.
Under the Digital Omnibus agreement reached on 7 May 2026 and pending formal EU adoption, the deadline for high-risk AI system compliance under the AI Act is deferred from 2 August 2026 to 2 December 2027. This gives enterprise software developers and operators an additional 16 months to audit their systems, write risk documentation and establish conformity procedures. The deferral applies to high-risk AI systems listed in Annex III: AI used in hiring, credit scoring, benefits decisions and similar high-stakes contexts.
What is not deferred: the fine structure. Violations of banned AI practices (enforceable since 2 February 2025) already carry fines of up to 35 million euros or 7 percent of global annual turnover, whichever is higher. That 7 percent ceiling is the harshest digital regulation penalty in the world, above GDPR's 4 percent ceiling. The DSA is also fully operational: the 200 million euro fine on Temu (issued May 2026 for inadequate risk assessments and product safety failures) remains the largest DSA penalty to date. The previous largest was the 120 million euro fine on X in December 2025.
Thailand's Board of Investment approved six projects worth a combined 958 billion baht, or approximately 29 billion dollars, in May 2026. TikTok's parent ByteDance received approval for an 842 billion baht (about 25 billion dollar) data-center expansion across Bangkok, Samut Prakan and Chachoengsao. This is materially larger than ByteDance's prior cumulative commitments to Thailand of roughly 8.8 billion dollars over five years. The project qualifies for a five-year corporate income tax exemption. Chachoengsao is Thailand's main data-center cluster, where TikTok, Gulf Development and other hyperscaler tenants are converging. Taken together with earlier investments by Amazon AWS (live January 2026), Microsoft (1 billion dollar commitment, partnership with Charoen Pokphand Group and True Corporation) and Gulf Development (4.3 billion dollar five-year expansion), makes Thailand one of the most active data-center markets in Southeast Asia.
The structural risks are power and heat. Data centers at the scale being built require liquid cooling infrastructure that Thailand's existing sites do not widely have. The country's power grid is already stretched during dry season, when hydroelectric output drops. Regulatory readiness and skilled labour supply are secondary constraints that the government has not yet squarely addressed.
A parallel governance dispute: the Ministry of Digital Economy and Society refused in June 2026 to halt the 1.62 billion baht ($45 million) 'TH-AI Passport' sovereign AI model project after opposition parties filed an anti-corruption investigation request. The contract was signed on 7 April 2026 with TH Consortium (Turnkey Communication Services and Human Intelligence Co Ltd, a Thai-Hong Kong joint venture). Opposition parties question the transparency of the terms of reference, potential conflicts of interest, and the use of extra-budgetary funds. Thai-language business press (Bangkok Biz, Prachachat) has covered this more directly than English-language outlets. The ministry defends the contract as transparent and legally compliant.
Russia's tech sector is fragmenting on two tracks simultaneously. On the censorship track, the Telegram and WhatsApp blocks, 469 banned VPN services, and the Max surveillance-messenger push (described above in Lane 4) represent the most aggressive enforcement of the runet isolation architecture since it was built. On the domestic-AI track, Yandex's YandexGPT and Sberbank's GigaChat are the main large language models in production. Both are constrained by chip access: Nvidia's advanced GPUs are under export control, domestic chip production (Elbrus, MCST) runs several generations behind, and the US June 2026 policy reversal on H200 exports, while it loosens restrictions for China, does not benefit Russia, which remains under full-spectrum sanctions.
The IT sector absorbed significant talent departure after the 2022 invasion of Ukraine, with estimates of 50,000 to 100,000 tech workers having left, though some returned. The current VPN crackdown is beginning to affect the tools that allowed Russian developers to access GitHub, foreign cloud services and professional networks. The long-term effect on Russia's domestic tech capacity is a slow-burn story that the blocking headlines do not capture.
Uzbekistan's IT Park reported Q1 2026 service exports of 191.8 million dollars, up 38.5 million year-on-year. The zone hosts more than 650 resident companies, with a significant share being foreign firms drawn by the tax and customs exemptions running until 2028. On 25 June 2026, IT Park held a Uzbekistan Digital Services Meetup in Chicago, bringing together Uzbek diaspora, tech companies and logistics firms to pitch the export growth story to the US market. A parallel business visit to Poland included participation in the Uzbekistan-Poland Partnership Forum in Gdansk on 16 June.
The government's long-term target is 5 billion dollars in annual IT exports by 2030, against a Q1 2026 annualised run rate of roughly 767 million dollars. That is a large gap requiring roughly 6.5x growth in four years. Uzbek-language outlets (Zamin.uz, Spot.uz) cover the IT Park story as a national economic success. The structural risks remain: dependence on Russian IT workers who relocated after 2022 and whose status could change under political pressure, and on a Western-oriented business model that could be squeezed if Russia-West tensions deepen further.
Israeli high-tech companies raised approximately 8.6 billion dollars in the first half of 2026, up about 45 percent from the same period in 2025. The number of funding rounds fell by roughly 35 percent, meaning investors are concentrating larger sums in fewer companies. Cybersecurity investment more than doubled compared to H1 2025. In May alone, Israeli startups raised nearly 1 billion dollars, with AI and defence as the dominant sectors. One AI security startup saw its valuation triple to 9 billion dollars in a year, capturing the premium investors are placing on AI-native defence tools.
The post-war recovery thesis is running in parallel with structural AI tailwinds. War accelerated defence-tech and cyber investment while the conflict was active; as it winds down, commercial tech investment is also returning. Both forces are present simultaneously. Calcalist (Hebrew-language) and the Jerusalem Post have covered the funding surge. Israeli founders cite the global AI boom as a structural driver independent of the local war dynamic. Israel has approximately 2,300 active AI startups, a density that rivals any geography outside the US and China.
The remaining headwind: a 15 percent US tariff on most Israeli goods is in place unless Israel strikes a deal with Washington. The tariff deadline is early July. Tech exports (software, services, IP royalties) are largely exempt from goods tariffs, so the direct impact on the tech sector is limited, but macroeconomic pressure on the broader economy filters into consumer tech spending and talent retention.
President Javier Milei reaffirmed in June 2026 his commitment to three pillars for attracting AI investment: a promise to keep AI unregulated, a new corporate category called the 'non-human corporation' (an entity managed exclusively by AI with no human ownership requirement), and a low corporate tax rate. Argentina is the first country to formally propose this category in draft legislation. The bill has not passed.
The political opposition is vocal. Former lawmaker Elisa Carrio said the proposal would turn Argentina into 'a catastrophic experiment for human dignity' and argued that unregulated AI would mean 'the dissolution of all moral and legal constraints, the end of the state, and the privatization of military and police power in the hands of private groups.' Opposition senator Agustin Rossi has filed a formal information request on the government's 'social digital twin' programme, an AI-based simulation of Argentine society used to model policy effects, asking for the legal basis, data protections and citizen rights framework. Spanish-language outlets (Ambito, Infobae) covered the twin programme in detail; international tech press focused almost entirely on the non-human corporation proposal.
The context: Argentina's stablecoin and crypto adoption is among the highest in the world, driven by structural distrust of the peso. The deregulation pitch follows the same pattern: local instability creates unexpected demand for alternative frameworks. Whether investors act on the pitch before the legal framework is enacted remains the open question.
Hyperscaler capex produces productivity gains that begin to justify the spending by 2028. Power supply expands through nuclear and grid investment. The H200 reversal eases diplomatic friction with China without materially closing the AI capability gap. The Anthropic IPO prices well and establishes a framework for valuing frontier AI labs that sustains investment. The splinternet stabilises into three blocs (US-allies, China, Russia-adjacent) that trade selectively. In this path, the countries that built data-center capacity in 2025 to 2026 (Thailand, Malaysia, Poland) capture disproportionate economic gains.
AI capex runs so far ahead of revenue that several hyperscalers cut guidance in late 2026 or 2027, triggering a repricing of AI-related equities. The Anthropic IPO disappoints or is delayed, undermining the near-trillion-dollar valuation benchmark. Power shortages slow the build-out before nuclear capacity arrives. Congress passes the AI OVERWATCH Act, creating legal uncertainty that freezes chip orders. China closes the HBM gap faster than expected, undermining the chip-control strategy. A major state-sponsored cyberattack on critical infrastructure forces a regulatory reset that slows AI deployment broadly.
For readers who follow this desk's cycle lens, kept strictly to pattern recognition, not prediction. Three significant astrological events fire in quick succession at the close of this week. Mercury stations retrograde on 29 June 2026 at 26 degrees Cancer, beginning a retrograde period through 23 July. Cancer governs domestic accumulation, data, and the private sphere. The classic pattern for Mercury retrograde is contract delays, systems audits, discovered errors in documents, and communications snags. For the AI and chip sector specifically, the timing coincides with H200 shipment paperwork clearing customs and the pre-IPO quiet period for Anthropic; both are exactly the kind of administrative-review processes that this window historically makes slower and more error-prone.
Jupiter enters Leo on 30 June, the first time since August 2014, beginning a 12-month transit. Leo governs spectacle, public performance, leadership and being seen. Jupiter amplifies whatever sign it occupies. The pattern maps onto the Anthropic IPO arc: a near-trillion-dollar company making its first public appearance on Wall Street is Jupiter-in-Leo energy, concentrated. The shift from Jupiter in Cancer (private accumulation, the Series H rounds that built the war chests) to Jupiter in Leo (public stage, investor scrutiny) is an archetypal transition from the funding phase to the performance phase.
A Capricorn Full Moon also falls on 29 June, opposite the Sun in Cancer. Full moons mark culminations; Capricorn governs institutions, structures and accountability. The institutional accountability pattern shows up as: congressional hearings on AI chip exports (AI OVERWATCH Act), the DSA Temu fine setting precedent for institutional enforcement against tech, and Russia's formal legal codification of internet control. Saturn and Neptune remain in Aries, their conjunction of limits meeting dissolution mapping onto the AI moment's core tension: enormous projections running into hard physical walls (power supply, HBM, chip nodes). The next major eclipse falls 12 August in Leo, amplifying the Jupiter-in-Leo themes. This week is the pre-turn moment before that louder cycle begins.
Primary sources, regulator releases, company filings and native-language outlets listed by section. Where a publication has a known editorial position, it is noted.
Prepared by the News Feed analyst desk. Checked against company filings, regulator releases, and native-language sources as of 26 June 2026. Where figures are estimates or unverified, this is noted above. Verify before acting.