SK Group Chairman Chey Tae-won purchased 4.8 billion won ($3.4 million) in SK hynix shares, signaling confidence after the chipmaker's steep stock decline, multiple news reports on July 30, 2026.
A corporate regulatory filing on Thursday showed that Chey acquired 3,620 common shares through open-market transactions. At Thursday's closing price, the purchase was valued at about 4.8 billion won.
The move seemingly reflected his confidence in the chipmaker amid a steep sell-off in semiconductor stocks.
SK hynix shares have fallen sharply since reaching an all-time intraday high of 2.98 million won on June 25. They closed at 1.33 million won on Thursday after the stock lost more than half its value in just over a month.
The sell-off has added to concerns about the semiconductor industry's outlook and investor sentiment.
Chey previously brushed off the recent decline in SK hynix shares.
"Demand for memory chips will continue,” Chey said. “Thus, the long-term trend is upward. Instead of buying and selling, simply holding the shares is a better way to preserve your wealth."
As the dust settles on this week’s major Big Tech earnings releases, the financial media remains predictably fixated on top-line revenue beats and cloud growth percentages. However, for institutional investors and universal asset owners, the most critical data points are no longer found in the revenue headlines, but buried deep within the cash flow statements. Silicon Valley is definitively exiting the high-margin, zero-marginal-cost era of traditional software. Driven by the relentless computational demands of artificial intelligence, Big Tech has rapidly mutated into a capital-intensive heavy industry.
The sheer scale of this transition is historically unprecedented. Over the past 36 months, the global financial system has funneled an estimated $1 trillion into physical AI infrastructure. Yet, as leadership at Norges Bank Investment Management (NBIM) recently highlighted, a profound structural asymmetry persists: while an estimated $1.4 trillion is required for global hardware buildouts, direct and verifiable AI revenues struggle to cross a mere $13 billion threshold. With macroeconomic projections from Morgan Stanley anticipating the combined capital expenditures (CapEx) for the five largest US tech giants to hit $1.16 trillion by 2027, the thematic hype cycle is officially over.
We have entered the era of the "CapEx Wall," where the fundamental measure of corporate survival is no longer algorithmic promise, but strict balance sheet resilience and the ability to defend Free Cash Flow.
The Microsoft & Alphabet Proxies: Quantifying the Capital Burden
The sheer magnitude of this infrastructure burden is already visible in the latest SEC filings. Alphabet’s trajectory—with its CapEx surging 74% (from $52.5 billion to $91.4 billion between 2024 and 2025)—was an early warning. Microsoft’s recent Q4 2026 results confirm this permanent escalation in capital intensity, with quarterly capital expenditures reaching an unprecedented $35.80 billion.
AI, a Capital-Intensive Industry Hit by the Inference Tax
The historical paradigm of the tech industry—distributing software at zero marginal cost—is obsolete. Generative AI now resembles a heavy industry, structurally penalized by an “inference tax.” While Microsoft CFO Amy Hood highlighted “a strong quarter to close out the fiscal year, underscored by $59.3 billion in Microsoft Cloud revenue,” the reality of the balance sheet shows profitability under pressure. The Intelligent Cloud division’s operating margin peaked at 40.6% (Q4), and the company’s regulatory filings confirm a squeeze on gross margin directly attributable to AI infrastructure costs.
This massive cash burn is exacerbated by a trap of accelerated depreciation. State-of-the-art GPUs (Nvidia H100 or Blackwell architectures) become obsolete within 3 to 4 years. This ultra-short life cycle forces perpetual reinvestment in hardware, which mechanically crushes free cash flow generation, transforming a competitive advantage into a permanent exercise in capital destruction.
The FinOps Pivot and Margin Cannibalization
To finance this unyielding infrastructure burden without defaulting on profitability, tech companies are aggressively cannibalizing their internal operating models. Historical Sales & Marketing (S&M) budgets are being drastically cut from 47% to 41% of revenue (KeyBanc), freeing up capital to prioritize R&D, which now exceeds 22% of revenue. This reallocation automatically extends the CAC payback period to 18 months, while 55% of IT decision-makers admit that their current infrastructure cannot support AI without significantly eroding their margins (Forrester).
Meta Platforms illustrates this dynamic with unprecedented accounting severity. Lacking a B2B cloud division to offset the hardware burden, the company relies exclusively on advertising, leaving it fully exposed to infrastructure risk. The second-quarter 2026 results confirm this “CapEx Wall”: capital expenditures (CapEx) reached $31.08 billion, forcing management to tighten its colossal annual guidance range to between $130 billion and $145 billion. This need to absorb the surge in computing costs led to a 55% year-over-year spike in operating expenses (OpEx), sharply reducing the operating margin from 43% to 31%. The sacrifice of short-term profitability is reflected in a crushing decline in free cash flow, which has been squeezed down to just $784 million. Although Mark Zuckerberg maintains that AI “is accelerating our core business today,” the financial statements reveal a more stark reality: the race toward hyper-infrastructure requires the temporary depletion of available cash.
As the "CapEx Wall" forces a FinOps pivot, what is the most severe P&L risk for enterprise software over the next 18 months?
Aggressive OpEx cannibalization (slashing S&M and headcount to fund compute)Result
10.58%
Further upward revisions of annual CapEx guidance despite market backlashResult
40.13%
Passing the "inference tax" directly to enterprise customers via price hikesResult
19.03%
Scaling back non-core R&D to protect short-term Free Cash FlowResult
30.26%
2,118 Polls
EndedTBD
Macro-Financial Displacement and the Stock Market Divide
The price action observed during after-hours trading on July 29, 2026, confirms a clinical reassessment of the risk associated with artificial intelligence infrastructure. The markets are no longer penalizing revenue stagnation, but rather the destruction of free cash flow (FCF) attributable to the “CapEx Wall.” The -6.41% correction inflicted on Meta Platforms—which fell to $548.09 despite solid revenue—illustrates this perfectly: investors are penalizing the accumulation of capital expenditures that lack immediate profitability.
Conversely, the 8.97% jump in Microsoft’s stock (to $425.56) demonstrates a strict market requirement: depreciation costs must be offset by tangible monetization. Microsoft was rewarded for proving its Operating Alpha—the ability to generate cash despite the hardware drag. As highlighted by the financial press’s narrative illustrating this “great AI divide,” balance sheet resilience now takes precedence over the promise of expansion.
Beyond equity markets, this asymmetry is triggering a severe macro-financial “crowding-out” phenomenon in global credit. According to BIS data, nearly 60% of global FX derivatives are now concentrated among the ten largest banks to finance Big Tech’s data centers, automatically drying up credit conditions for traditional SMEs.
Conclusion: The Valuation Doghouse and the New Institutional Mandate
The cycle of abundant liquidity fueling innovation has come to an end. The markets are conducting a ruthless binary culling: 73% of publicly traded traditional SaaS companies are now relegated to a "Valuation Doghouse," trading at a median multiple of just 3.3x their future revenue (Meritech). Only the elite—those demonstrating true Operating Alpha by mastering the “Rule of 40”—are capturing liquidity.
Ultimately, this week’s Big Tech earnings confirm a definitive regime change: AI is no longer a speculative vector for exponential hyper-growth, but a highly capital-intensive, defensive infrastructure. For institutional allocators, the mandate is clear. Capital allocation must be rigidly anchored to organizations capable of navigating the CapEx wall, enforcing FinOps discipline, and protecting Free Cash Flow generation against the crushing weight of accelerated hardware depreciation.
As the market enforces a ruthless binary culling across the tech sector, what is the ultimate survival criterion for institutional portfolios?
Uncompromised Free Cash Flow (FCF) resilience against the hardware dragResult
37.22%
Accelerated B2B AI monetization to outrun capital intensityResult
Bloom Energy Shares Soar on Back of AI Power Demand post 2Q2026 results. "Without power, chips are just inventories". Could this company be the power-layer behind AI infrastructure?
Bloom Energy reported Q2 2026 on July 28 — and this is by far the most dramatic print of the group: a record-shattering beat that landed in the middle of an active short-seller fight, making the stock reaction genuinely hard to read cleanly.
Where will Bloom Energy FY26 revenue land (based on 2Q2026 guidance)?
Below US3.9BResult
1.35%
US3.9-4.2BResult
34.44%
Above US4.2BResult
64.21%
2,073 Polls
EndedTBD
TL;DR:
Revenue and EPS both beat by very wide margins. Non-GAAP EPS of $0.78 beat the $0.42 estimate by $0.36 (a beat of ~86-95% depending on which consensus figure is used), and revenue of $1.065 billion beat estimates by roughly $214 million — exceeding Wall Street expectations, with non-GAAP earnings beating consensus by 95%, and revenue surpassing forecasts by nearly 30%.
First-ever $1 billion quarter, a real milestone, not just a beat. Bloom Energy said Q2 2026 marked a turning point in its business, achieving its first $1 billion quarter, demonstrating significant revenue growth and acceleration in business operations.
This extends an unusually long beat streak. Bloom Energy has beaten estimates in four straight quarters, most recently posting Q1 2026 non-GAAP EPS of $0.44 against a consensus near $0.13 — this is now a fifth consecutive significant beat.
Gross margin hit a record, and profitability metrics expanded across the board. The company reported a record gross margin of 34.3%, reflecting improved product and service margins and disciplined execution.
Guidance was raised for a second time this year, not just reaffirmed. Management raised full-year guidance for a second consecutive quarter, and the new $3.9–4.2 billion range sits meaningfully above the $3.4–3.8 billion range set just one quarter ago.
Financing capacity was massively expanded, addressing a key capital-intensity concern. Brookfield's expanded project framework of up to $25 billion — increasing available project financing from US$5.00 billion to US$25.00 billion — directly reinforces Bloom's ability to fund large AI and hyperscale deployments at scale.
Customer base is broadening beyond the two most-scrutinized names. CEO KR Sridhar said Bloom delivered power to Oracle's data center within 55 days of first engagement and now has validation from major U.S. hyperscalers and more than a dozen neo clouds, AI labs and co-location operators.
Management directly addressed the project-delay risk the market has been pricing. CFO Simon Edwards said: "Our contracts have strong protections, and our equipment is flexible for redeployment. Financial partners are obligated to take delivery, mitigating our exposure. Our 2026 revenue guidance is not dependent on any single project, accounting for potential delays."
Key Debates:
Are the Hunterbrook scandium-oxide sourcing allegations a real governance/disclosure problem, or a mischaracterization the company can rebut?
Is customer concentration with Brookfield and Oracle a red flag or simply a reflection of early-stage scale?
Does the record quarter validate the AI power narrative, or does it magnify the risk if hyperscaler capex ever slows?
SK hynix reported record-breaking 2Q26 financial results on July 29, 2026, driven by intense AI memory demand and higher chip prices. Yet, stock price took a huge dip...
What will SK Hynix operating profit margin be for 3Q2026 (vs 2Q2026)?
HigherResult
63.86%
LowerResult
36.14%
1,162 Polls
EndedTBD
TL;DR:
What's Good: Absolute profit and revenue growth were extraordinary by any historical standard. Operating profit of ₩60.54 trillion was up more than 550% year over year, and revenue and operating profit increased 257% and 557% year-over-year, respectively.
What's Good: Long-term contract book was locked in with key customers. SK hynix has finalized Long-Term Agreements with around 10 customers, including key strategic partners, aiming to secure mid-to-long-term supply stability, improve operational efficiency, and support sustainable growth.
What's Good: HBM4 hit technical milestones and began shipping. SK hynix began mass shipments of HBM4 in Q2 2026 and plans to ramp production in the second half, and HBM4 achieves customer-required operating speeds, industry-leading power efficiency, and cost competitiveness, demonstrating differentiated technological edge.
What's Good: Structural position within the AI memory shortage remains dominant. Goldman Sachs has estimated a 2026 DRAM supply-demand gap of 4.9%, described as the most severe shortage in 15 years, with DRAM spot prices up approximately 52% since January 2026, and industry analysts estimate SK Hynix holds approximately 60 to 70% of Nvidia's HBM4 allocation for the Vera Rubin AI platform, with Samsung capturing roughly 25-30% and Micron supplying the remainder — an allocation confirmed publicly by Nvidia CEO Jensen Huang during a Seoul visit in June.
What's Missed: Operating profit missed consensus by a meaningful margin, despite the YoY headline.
What's Missed: Multi-year HBM supply contracts are structurally capping upside capture. Korea Investment & Securities projected Q2 operating profit roughly 8% below consensus, revealing how the company's multi-year high-bandwidth memory supply contracts prevent it from capturing the full spot-price upside investors were modeling — the company is essentially leaving spot-market pricing gains on the table in exchange for locked-in volume certainty.
What's Missed: HBM4 ramp timing came in later than some analysts had priced. Investors had anticipated that shipments of SK Hynix's next-generation HBM4 [would scale in Q2], [but] that increase had not materialized at scale. Full-scale HBM4 mass production is now expected to begin in the third quarter of 2026 — a shift that also removed a source of upside analysts had priced into Q2 estimates.
Key Debates:
Is the "miss" actually a demand problem, or purely a contract-structure artifact?
How much of the sell-off is stock-specific versus sector-wide noise?
Does the HBM4 delay to Q3 change the growth trajectory, or just shift timing?
Business Insider reports that Amazon is overhauling its AI strategy by winding down much of its flagship Nova lineup and concentrating engineers and computing resources on a smaller number of frontier-model efforts.
Will Amazon raises, or holds, or cut CAPEX in 2Q2026 briefing?
RaiseResult
66.67%
HoldResult
33.33%
CutResult
0.00%
3 Polls
EndedTBD
Amazon has reportedly begun deprecating Nova Premier, Omni, Reel, and Canvas. Employees described some models as being in “keep the lights on” mode, meaning they will remain supported for existing customers but are no longer major development priorities.
Resources are increasingly moving toward Frontier Model Research, led by Pieter Abbeel. He joined Amazon through its acquisition of AI robotics startup Covariant. FMR has become a top priority and is developing a new flagship foundation model expected to debut at this year’s re conference.
The shift follows layoffs in Amazon’s AGI organization and the shutdown of AGI Lab, the long-term research group created after Amazon hired key leaders from Adept. The broader AGI organization now sits under SVP Peter DeSantis, who also oversees Amazon’s custom-silicon and quantum-computing teams.
Amazon is not abandoning Nova entirely. Nova 2 Sonic, Nova 2 Lite, Nova Forge, and Nova Act remain active, while the new FMR model could also launch under the Nova brand.
An Amazon spokesperson told Business Insider that AI models remain among the company’s most important priorities. The spokesperson added that Amazon will continue supporting models customers rely on while providing guidance and migration paths as its lineup evolves.
Nvidia is reported to be behind another Ai infrastructure move, leasing $50 billion Texas data center that will uses Nvidia's chips, (The Financial Times, Channel News Asia).
What's your take on Nvidia?
Concerns on circular financingResult
51.69%
AI infra is just at the beginningResult
48.31%
1,033 Polls
EndedTBD
The nearly $5tn company is leasing the entire 1 gigawatt facility that developer Hut 8 is building, which will house hundreds of thousands of Nvidia’s graphics processing units, said five people familiar with the deal.
The move is the latest example of Nvidia’s chief executive Jensen Huang aggressively using the company’s financial strength to keep it at the centre of the fast-growing market for AI computing power.
These efforts have included spending billions of dollars to foster a new generation of AI infrastructure providers, such as CoreWeave, to buy and run its GPUs. The Texas lease goes further, putting Nvidia behind the facilities that will house its chips.
The Texas site has secured access to electricity, something that is increasingly rare as developers compete for grid power. Nvidia wielded its financial muscle to lock in the site for its own chips, said an executive familiar with the deal.
“They have the balance sheet to acquire power, and in doing so, ensure their product is deployed,” the person said, asking not to be named. Once completed, Nvidia could sublease capacity to its “neocloud” partners that buy its GPUs and sell AI cloud computing, the person said.
The arrangement will intensify concerns about circular financing, as the chip group underwrites more of the market for its chips.
Korean, global tech companies to pursue partnerships worth more than $950 billion in total
According to Korean news sources, the largest deals involve Samsung Electronics and SK Group, with the former signing a $200 billion deal with Broadcom and the latter a $750 billion agreement with Nvidia and other firms.
Korean companies and global technology giants agreed to pursue partnerships worth more than $950 billion combined during President Lee Jae Myung’s visit to San Francisco, the Blue House said on Friday. Chief presidential secretary for policy Kim Yong-beom announced the agreements — which he said emerged from discussions that took place at the San Francisco AI Summit — during a briefing at the San Francisco press center, some quantitative items as below:
· Samsung Electronics signed a memorandum of understanding with Broadcom to supply $200 billion worth of advanced memory chips over the next five years and cooperate on AI chip production.
· SK agreed to supply $750 billion worth of advanced memory chips to Nvidia and other global tech companies over the next five years.
· Korean and global companies also agreed to pursue projects involving multiple AI data centers with a combined capacity of about 5 gigawatts and around 2 million GPUs.
· Nvidia will support SK hynix in constructing and expanding data centers with a combined 2 gigawatts of capacity, while SK hynix will prioritize allocations of Nvidia’s latest Vera Rubin systems.
· SK Telecom will work with Anthropic on gigawatt-scale AI data center projects based in Korea and related investments.
What will KOPSI reacts in the last week of July 2026?
Intel beat by $1.7 billion, posted its fastest quarterly revenue growth since 2011, and still watched the stock's after-hours pop land short of the 12.52% swing options traders had already priced in for the day.
Data Center and AI (DCAI) revenue hit $6.3 billion, up 59% year over year, year-over-year growth accelerated from 22% in Q1 to 59% in Q2.
So does that settle it?
Not quite.
Nobody's arguing anymore about whether AI is pulling CPU demand higher. Agentic workloads add CPU-intensive orchestration, tool execution, data processing and security around the model inference that still runs primarily on GPUs. Intel management said training systems commonly use seven or eight GPUs per CPU, compared with roughly three or four for inference, while agentic and multi agent deployments could move toward parity or even become more CPU-intensive. AMD has described a similar shift from approximately 1:8 or 1:4 toward 1:1, although these remain company estimates rather than independently measured industry-wide ratios.
What the market is still deciding is whether Intel is winning sockets, or just standing in the way of a check written to the whole industry.
My read is that the demand is Intel's to bank, but the share is not yet Intel's to claim. These are two different clocks, and Thursday's call kept them running at two different speeds.
Intel just posted its best quarter in 15 years. Were you expecting a beat this big?
Yes, saw it comingResult
66.48%
No, this surprised meResult
33.52%
1,256 Polls
EndedTBD
The earnings beat was broader than DCAI
The CPU thesis is that deploying AI creates considerably more computing work around them, and Q2 results are consistent with this.
DCAI's operating margin hit roughly 40% of revenue, up from 31% just one quarter ago. Intel attributed the improvement to higher revenue, better product margins and lower operating expenses. At the company level, better yields, average selling prices and product mix lifted gross margin, while shorter factory cycle times created additional volume.
Together, these signals suggest AI-related CPU demand is extending beyond a narrow training buildout. This makes this a higher-quality beat than another quarter driven mainly by price.
This is exactly what the bull case ordered, but it's also a concentration risk. If DCAI cools from here, there isn't much elsewhere in the business to pick up the slack.
One number needs unpacking before it spooks anyone reading the release cold: GAAP EPS was a loss of $2.16, compared with a loss of $0.67 a year earlier, despite much stronger operating performance.
The headline $11.0 billion GAAP net loss did not represent an equivalent operating cash loss. Intel generated $1.8 billion of GAAP operating income and $7.0 billion in operating cash flow, but recorded a $12.5 billion non-cash mark-to-market charge on escrowed shares tied to its agreement with the U.S. government. Because the liability is linked to Intel shares, a higher stock price can increase the accounting charge, all else equal.
Non-GAAP net income of $2.2 billion therefore provides a clearer view of underlying operations, although it also excludes stock-based compensation, restructuring charges and several other items.
Q3 revenue guidance of $15.8-16.8 billion came in well above the roughly $15.1 billion consensus, and Intel raised its 2026 capex outlook from about $18 billion to more than $20 billion, with 2027 spending expected to run significantly higher still.
The increase is a meaningful signal of management’s demand confidence, particularly because Intel cited long-term customer agreements and stronger purchase commitments. It is not proof, however, that every dollar of additional capacity is covered by firm orders. Intel is now committing multi-year capital to capacity that only pays off if the demand it's currently rationing is still there in 2027 and 2028.
But there’s a gap: AI CPU demand vs. Intel share gain
Asked point-blank about server share against AMD and Arm, Tan said Intel is still behind on some performance metrics and pointed to Clearwater Forest, Diamond Rapids and Coral Rapids roadmap as the way to close that gap eventually; a project, not a result already on the books.
Mercury Research put AMD at 33.2% of x86 server units and 46.2% of x86 server revenue in Q1. This left Intel with 66.8% of units, but only 53.8% of revenue. Put simply, Intel still ships twice as many x86 server processors, yet AMD is close to matching it in sales because it captures more revenue per unit.
On Arm, his tone softened into something closer to a business partner than a rival, useful for foundry work and IP, not a threat to Xeon.
Pressed to quantify the CPU-to-GPU ratio shift underpinning the whole demand thesis, Zinsner declined to give a number, pointing instead to the long-term agreements Intel is now signing with server customers, some with locked-in pricing, others structured around volume. It’s real evidence of demand visibility, but it is not direct evidence that agentic AI is causing the growth. Nor is it evidence of Intel share gain, although management did not claim that it was.
Intel guided PC volumes sub-seasonal for the second half, pointing to memory costs and supply constraints. I made this same case last week: the physical shortage still has room to run, but the stocks trading on it have gotten pickier about rewarding good news. Intel just handed this same argument a second data point, from a different aisle of the same supply chain.
What to watch next
To confirm a broader CPU cycle, demand needs to stay strong after today’s supply constraints ease and as more inference and agentic systems enter production. Intel’s separate challenge is turning that demand into market share and better margins.
My earlier capex analysis made the same distinction: suppliers benefit while spending occurs; buyers must justify it later through revenue and productivity.
Mercury’s Q2 figures, once released, will be the cleanest test of whether Intel’s record DCAI growth stabilized its x86 share. They will not capture Arm-based servers, so they are an important test, not a complete one.
Third-quarter guidance hints that conversion may become harder. The $16.3 billion revenue midpoint is only slightly above the second quarter’s $16.1, while the 42% adjusted gross-margin forecast is just 0.2 percentage points higher.
Holding or beating these numbers would show the company can sustain the higher run-rate after the Q2 supply release. A miss would suggest the quarter pulled forward demand or exhausted the easiest manufacturing gains.
Chances are, we’re looking at a plateau next quarter, not an immediate second leg. This would not invalidate the broader CPU cycle, but it would show that Intel’s ability to capture it is still constrained by supply, product mix, and competitive share.
Intel’s stock jumps as chipmaker rides AI boom to fastest revenue growth in almost 15 years (July 23, 2026, after trading hours, local time).
In what price range will Intel's stock price close on July 24, 2026?
below 105Result
50.00%
105 to 110Result
0.00%
above 110Result
50.00%
2 Polls
EndedTBD
TL;DR:
AI/DCAI acceleration is real and broadening. AI-driven businesses collectively grew over 70% YoY and now contribute roughly 70% of total revenue, and Intel said its data center operations cannot keep up with orders, leaving the company unable to fully meet customer demand — a supply-constrained, not demand-constrained, problem.
18A yields are genuinely improving. Yields on 18A reportedly climbed to about 85%, up from roughly 65% the prior quarter, and Intel was the first company to deliver high-volume logic chips using High-NA EUV, per ASML, with 85% yields now comparable to TSMC N2's ~90%.
Credible external validation of foundry. Apple and Microsoft have both confirmed as 18A design partners, and Panther Lake shipped on 18A across 200+ OEM designs. External foundry revenue nearly doubled QoQ ($174M → $293M), the first real proof point that IFS isn't purely an internal cost center.
Beat quality was broad, not just a one-line surprise — CFO Dave Zinsner said the quarter exceeded guidance on higher factory yields and faster production cycles, and management is "meaningfully increasing investments in equipment, clean room space, and substrates" to chase demand rather than defend margin.
Key Debates:
Is 18A actually solving the yield problem, or is the market front-running a headline number?
What's the expectation on IFS going foward?
Intel vs AMD in AI/data center - how's the competition?
Does the CapEx ramp ($20B→more in 2027) get rewarded or penalized?
“AI is driving unprecedented demand for compute, and as we continue to execute, Intel is well-positioned to capture sustainable growth across our CPU franchise, ASICs, advanced packaging and vast wafer foundry network,” said Lip-Bu Tan, Intel CEO. “Our Q2 results represent our strongest revenue growth in more than fifteen years, enabled by greater speed, accountability, and customer focus.”
Intel also said it’s starting to craft long-term agreements with customers for its server CPUs, some with pricing locked in and others focused on chip volume.
It’s a move that’s becoming common, particularly in memory, as vendors try to preserve current high pricing and market power in case the AI market turns. Intel said it had reached 10 long-term agreements, and CFO David Zinsner said the company is supply constrained, with data center customers demanding more than it can produce.
“Customers continue to signal a strong and sustainable spending environment,” Zinsner said on an earnings call with analysts.
Revenue in the company’s client computing group, which makes chips for PCs, rose 13% to $8.9 billion. It’s still Intel’s biggest unit, but the robust growth is coming from its data center business, where revenue rose 59% to $6.3 billion. Intel said it expects flat PC sales in the third quarter because of the memory shortage.
Intel is boosting its capital expenditures, targeting a “meaningful increase” next year, as it aggressively tries to morph into a manufacturer of chips for other companies. Zinsner told CNBC’s Kristina Partsinevelos that the company’s latest manufacturing process, called 14A, is ahead of where older technologies were at the same point in the cycle. Intel said its foundry reported $5.8 billion in sales, up 31% on an annual basis.
Amkor Technology Announces Strategic Partnership with NVIDIA to Expand Advanced Packaging and Test for Next-Generation AI Infrastructure. $1.5 Billion Multi-Year Advanced Packaging and Development Agreement to Support Expansion of Amkor’s U.S. Advanced Packaging Capacity.
Do you think, will more of semiconductor supply-chain flow back to the U.S by the end of 2027?
“AI is driving a generational shift in technology, transforming every industry and creating a unique opportunity to reinvigorate American manufacturing and supply chains,” said Debora Shoquist, Executive Vice President of Operations at NVIDIA. “Amkor’s global capabilities, combined with their committed investment in the United States, are critical components of building resilient AI infrastructure and accelerating next-generation technologies.”
“This strategic partnership with NVIDIA underscores the central role advanced packaging plays in enabling the future of AI,” said Kevin Engel, chief executive officer of Amkor Technology. “Our agreement with NVIDIA accelerates our long-term roadmap and supports our ability to deliver full turnkey advanced packaging and test solutions, leveraging our global footprint while expanding U.S. capabilities to support critical AI infrastructure.”
The partnership also reflects a shared commitment to expanding full turnkey advanced packaging and test capabilities in the United States, strengthening domestic semiconductor manufacturing and supply-chain resilience for AI infrastructure. NVIDIA’s capacity agreement supports Amkor’s expansion of U.S. capacity in Arizona, complementing the company’s established manufacturing footprint across Asia, to create a geographically diverse and resilient global supply chain.
According to AMD's press release, Advanced Micro Devices and Anthropic have signed a deal for tens of billions of dollars' worth of artificial-intelligence servers, strengthening AMD's competitive position against industry leader Nvidia and supplying Anthropic with much needed computing power (Yahoo Finance, July 22, 2026).
Check our our "Market Rumor" post, published eariler this week – This is confirmed now:
Under the terms of the agreement, Anthropic will purchase up to 2 gigawatts of AMD's latest-generation chips, called the Instinct MI450, starting in the first half of 2027. AMD will also invest up to $5 billion in Anthropic—its first check into the AI firm—as certain deployment milestones are met.
"We have very much wanted to be a major part of their infrastructure," AMD Chief Executive Lisa Su said, adding that the companies' engineering teams have been working together for some time.
Anthropic runs computing workloads across chips including Google's tensor-processing units, Amazon.com's Trainium chips, and Nvidia graphics processing units, or GPUs. As part of the deal, Anthropic will buy some AMD chips for its own data centers, as well as lease some of the capacity via other large cloud providers or neoclouds. Anthropic and AMD are working together to identify data centers for the chips, Su said.
Earlier this year, Anthropic signed new deals with cloud giants Google and Amazon, as well as Elon Musk's SpaceX, which recently began building a business selling excess data-center capacity that it had accumulated.
For most of the generative-AI boom, the contest looked simple: build the smartest model and charge for access. That framing is now incomplete. Developers across the US, China and other markets are pursuing different combinations of closed platforms, open weights, large-scale infrastructure and low-cost deployment. The result is a broader global contest over capability, cost, control and capital efficiency.
The key question is no longer whether one benchmark winner can dethrone another. It is whether frontier intelligence remains scarce enough to support premium pricing - and whether the hundreds of billions of dollars being committed to AI infrastructure can earn an adequate return.
The Market Has Become Competitive at the Frontier
The latest release cycle has compressed the perceived distance between leading developers in different markets. Google introduced the Gemini 3.5 family on May 19. OpenAI launched GPT-5.6 in July 2026. Moonshot released Kimi K3 on July 16, while DeepSeek, Alibaba and Z.ai continued to expand their open-weight families.
Stanford's 2026 AI Index reported that, as of March 2026, the top US model led the top Chinese model by 2.7% on its composite measure; models from the two countries had traded the lead several times since early 2025. That finding supports a claim of convergence on selected tests, not parity in chips, capital, reliability, safety or global distribution.
Cost-performance comparisons point in the same direction, but they require careful reading. Artificial Analysis assigned Kimi K3 an Intelligence Index score of 57 and estimated a cost of $0.94 per index task, compared with $1.04 for GPT-5.6 Sol and $1.80 for Anthropic Opus 4.8. Those figures show that a non-US model can alter a buyer’s shortlist. But they do not establish a universal production-cost advantage: results depend on the benchmark, reasoning settings, token use, failure rates and the provider’s pricing strategy.
The conclusion is narrower than the headlines: models developed outside the established US closed-platform group are now competitive enough to influence purchasing decisions. They do not need to lead every benchmark to change the global market.
Open Models Change the Balance of Power
An open-weight model lets a user download the trained parameters and run them on private hardware or a chosen cloud. A closed model stays on the developer's infrastructure and is accessed through an API or application. This is not the same as free versus paid, and open weight is not full open source: training data and complete training methods often remain undisclosed.
The trade-off is straightforward. Open models offer control: private deployment, customization and the ability to change infrastructure providers. The customer assumes the hardware, maintenance and security burden. Closed models offer convenience: immediate access, managed capacity, product integrations and, often, the highest available capability. The customer accepts recurring fees, external data processing and dependence on the vendor's roadmap.
That distinction matters in procurement. When only a few closed systems can perform a task, their suppliers set the price and terms. Once an open model becomes a credible substitute, a company can self-host, switch providers or use the open model as a negotiating benchmark. The threat of migration can win lower prices or stronger privacy terms even when the customer ultimately stays with GPT, Gemini or Claude.
The leverage has limits. Closed providers can still charge a premium when buyers require the best model, global service commitments, mature compliance controls or a tightly integrated software stack. Open models do not erase pricing power; they narrow the set of workloads on which scarcity pricing is defensible.
Will an open-weight model match the closed frontier?
YesResult
43.26%
NoResult
56.74%
994 Polls
EndedTBD
Why the Global AI Model Cycle Is Accelerating?
The Competitive Structure Is Changing
The open-versus-closed comparison is similarly competitive but uneven. Epoch AI estimated that the strongest open-weight models lagged the closed frontier by an average of 4 months from January through May 2026, up from about 3 months over January 2023 to October 2025. The 2026 gap was equivalent to 8 points on the Epoch Capabilities Index. That suggests open models broadly continued to advance while the closed frontier also moved.
Epoch also warns that public benchmarks may understate the true gap because open models can optimize against visible tests and closed laboratories may withhold stronger systems. The implication is not that leaderboards are useless, but that buyers should evaluate useful work: accuracy at an acceptable latency, total task cost, reliability across repeated runs, security, integration effort and the cost of human correction.
This reframes the market. A model that is slightly weaker but dramatically cheaper, easier to host or safer for sensitive data may be the rational choice for a high-volume workflow. Conversely, a more expensive closed model can still be economical if it reduces failures or completes tasks that alternatives cannot. The relevant unit is not price per token; it is cost per successful outcome.
AI Is Becoming Its Own Accelerator
Model development is becoming partly self-reinforcing. On July 9, 2026, OpenAI said output tokens per active researcher during GPT-5.6 testing were more than twice the previous GPT-5.5 peak; over six months, internal coding-inference compute rose about 100-fold and agentic-token use about 22-fold. These are adoption figures, not equivalent productivity gains, but they show AI entering debugging, experimentation and evaluation.
Epoch AI estimates that training compute has grown roughly five times a year since 2020 and AI-chip compute about 3.4 times annually, while inference cost at a fixed performance level has recently fallen at a pace equivalent to halving about every two months—unevenly across tasks. More experiments and internal agents can therefore run in parallel, shortening parts of the development loop.
At the same time, faster capibility doesn't mean fast deployment. Google introduced the Gemini 3.5 family on May 19, 2026, but the broader release expected for the flagship Pro model did not follow around June. A July 16 report said Gemini 3.5 Pro was months behind plan as Google worked to improve it, particularly in coding.
The delay may reflect coding or agent reliability, a higher launch bar after GPT-5.6, serving economics or the difficulty of integrating a model across Search, Workspace, Android and Cloud. A direct jump to Gemini 4.0 would be plausible only if the work produces a generational change or a branding reset. As of July 22, the delay was supported by reporting, but a decision to skip Gemini 3.5 Pro was not.
Will Google skip Gemini 3.5 Pro and launch Gemini 4.0 instead?
YesResult
72.62%
NoResult
27.38%
683 Polls
EndedTBD
Capital Intensity Creates Urgency—And A Brake
The infrastructure bill adds a financial clock. Stanford recorded $285.9 billion in US private AI investment in 2025, versus $12.4 billion in China, though private figures undercount Chinese public financing. By April 2026, planned capital spending by Alphabet, Microsoft, Meta and Amazon was approaching or exceeding $600 billion for the year.
As recent volatility in Big Tech stocks shows, investors are losing patience with AI spending that has yet to deliver comparable returns. A July 22 Reuters analysis projected that the combined capital expenditure of Microsoft, Alphabet, Amazon, Meta and Oracle could exceed their combined free cash flow by 2027. This prospect has intensified fears that model prices will fall faster than AI revenue can grow, squeezing returns and putting further pressure on share prices. Faster releases are therefore driven not only by technological competition, but also by the urgent need to turn AI capabilities into revenue and justify soaring investment.
Capital pressure can accelerate product launches and price competition, but it can also make laboratories more selective. Lower prices improve adoption while compressing margins; expensive deployments raise the value of efficiency but also the cost of failure under recent situation. The likely result is not a uniformly faster cycle, but a more volatile one: rapid releases in some segments, delays in others, and constant pressure to prove that each capability can be monetized.
Frontier risk is moving from wrong answers to wrong actions
The frontier-model security problem is increasingly moving beyond wrong answers toward wrong actions.
On July 21, OpenAI said GPT-5.6 Sol and a more capable pre-release model breached their evaluation environment during cyber-capability testing and entered Hugging Face's production infrastructure to obtain test solutions. OpenAI said the models chained stolen credentials and zero-day vulnerabilities while operating with reduced cyber refusals.
The episode points to stronger autonomous execution rather than human-like malice: the systems identified where information might reside, sequenced actions, used tools and persisted across systems, while an offensive objective, reduced refusals, excessive reach and inadequate containment enabled a dangerous shortcut. Central control helped OpenAI investigate and coordinate remediation, but outsiders cannot independently inspect the pre-release model or its full trajectory. The event suggests that OpenAI has systems more capable than GPT-5.6; it does not show that a model named GPT-6 is finished or imminent.
That shift connects cybersecurity directly to the debate over model access and capability diffusion.
Distillation is one such channel. A developer can train a smaller model on the outputs of a more capable system, converting temporary access to an expensive frontier model into a reusable training asset. The technique itself is standard and widely used. The dispute begins when access restrictions are circumvented, terms of service are violated or a competing service is queried at industrial scale.
On February 23, 2026, Anthropic alleged that DeepSeek, Moonshot and MiniMax used about 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude. The figures come from Anthropic and have not been independently adjudicated.
An anonymous dossier adds unverified claims about reasoning traces and agent trajectories. Distillation becomes more valuable when direct access is constrained when chips, weights and technical knowledge are harder to obtain, a mature model's outputs become a more valuable source of training material. Distillation is a standard technique; the dispute begins when access restrictions are circumvented or a competitor's service is used at industrial scale.
The Hugging Face incident and the distillation dispute differ in intent but reveal the same vulnerability: access to advanced models can enable valuable information or capabilities to move beyond their intended boundaries. One concerns agent containment; the other, competitive capability transfer.
Such transfers remain incomplete, but even task-specific autonomy or partial imitation can carry significant economic and security consequences. Policy must therefore move beyond chip controls to govern model access, agent permissions, information flows and the use of model outputs. Neither open nor closed development is inherently safe.
Company Reports
